XccesS Wiki - Benutzerbeiträge [de]

SAP

2026-06-17T07:46:21Z

Hendrik: /* JAVA */

===Beschreibung===
SAP ist ein deutsches Software-Unternehmen, das sich auf die Entwicklung und den Verkauf von Unternehmenssoftware spezialisiert hat. Die Software-Lösungen von SAP decken eine Vielzahl von Geschäftsbereichen ab, wie z.B. Finanzwesen, Personalwesen, Einkauf und Vertrieb. Die Produkte von SAP werden von Unternehmen aller Größenordnungen genutzt, um Geschäftsprozesse zu optimieren, Daten zu verwalten und bessere Entscheidungen zu treffen. Zu den bekanntesten SAP-Produkten gehören SAP ERP, SAP HANA und SAP S/4HANA.
===SWPM starten===
<syntaxhighlight lang="bash">
./sapinst SAPINST_STACK_XML=stack.xml SAPINST_USE_HOSTNAME=
</syntaxhighlight>
===SUM starten===

====Linux====
=====ABAP=====
<syntaxhighlight lang="bash">
cd /usr/sap/$SID/SUM
rm -R ./SUM
SAPCAR -xvf ./SUM11SP05_0-80006800.SAR
cd SUM/abap/
sudo ./SUMSTART confighostagent $SAPSYSTEMNAME
</syntaxhighlight>
=====JAVA=====
<syntaxhighlight lang="bash">
cd /usr/sap/$SID/SUM
rm -R ./SUM
SAPCAR -xvf ./SUM*.SAR
cd SUM
sudo ./STARTUP confighostagent $SAPSYSTEMNAME
</syntaxhighlight>
====Windows====
<syntaxhighlight lang="bash">
STARTUP.BAT confighostagent <SID>
</syntaxhighlight>

===SAP Host Agent===
====Version überprüfen====
[https://me.sap.com/notes/0002032385 Note 0002032385]

Windows:
<syntaxhighlight lang="bash">
%Program Files%\SAP\hostctrl\exe\saphostexec.exe -version
</syntaxhighlight>
Linux:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostexec -version
</syntaxhighlight>

====Aktualisieren====
SAR Archiv herunterladen und nach /usr/sap/Z36/SUM kopieren
als root am System anmelden
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostexec -upgrade -archive /usr/sap/<SID>/SUM/<ARCHIV>.SAR
</syntaxhighlight>
===Kernelupdate===
''Siehe auch Workitem 2L-DB-SAP-00026: SAP Kerneltausch''

Aktuellen Kernel herausfinden
<syntaxhighlight lang="bash">
disp+work -v | grep "patch number"
</syntaxhighlight>
Kernel Backup erstellen und alte Backups löschen
<syntaxhighlight lang="bash">
cd /sapmnt/*/exe/uc/
rm -R linuxx86_64_*
cp -R linuxx86_64 linuxx86_64_<DATUM>
</syntaxhighlight>
[https://me.sap.com/softwarecenter/support/index SAP Downloadportal] aufrufen und nach "SAP KERNEL 7.85 64-BIT UNICODE" suchen
Die Archive herunterladen
Nach /sapmnt/SID/exe/uc/linuxx86_64/ kopieren
Ggf. noch Dateiberechtigungen korrigieren
<syntaxhighlight lang="bash">
chown -R <SIDADM>:sapsys *.SAR && chown -R <SIDADM>:sapsys *.sar
</syntaxhighlight>
In der richtigen Reihenfolge entpacken:
# SAPEXEDB.SAR
# SAPEXE.SAR
# dw_utils
# dw
# R3Trans
# tp
<syntaxhighlight lang="bash">
cd /sapmnt/*/exe/uc/linuxx86_64/
SAPCAR -xvf SAPEXEDB_*.SAR
SAPCAR -xvf SAPEXE_*.SAR
SAPCAR -xvf R3trans_*.SAR
SAPCAR -xvf tp_*.sar
SAPCAR -xvf sapftp_*.sar
SAPCAR -xvf dw_*-*.sar
SAPCAR -xvf dw_utils*.sar
SAPCAR -xvf lib_dbsl*.sar
SAPCAR -xvf sapwebgui*.sar
SAPCAR -xvf saphttp_*.sar
SAPCAR -xvf abap2vcs_*.sar
SAPCAR -xvf SYBCTRL*.SAR
SAPCAR -xvf sapnwrfc_*.sar
SAPCAR -xvf ENSA2_*.SAR
SAPCAR -xvf saplicense_*.sar
SAPCAR -xvf enserver_*.sar
</syntaxhighlight>
Aufräumen und alte SAR Archive löschen:
<syntaxhighlight lang="bash">
cd /sapmnt/*/exe/uc/linuxx86_64/ && rm *.SAR && rm *.sar
</syntaxhighlight>
Oder in Kurzform:
<syntaxhighlight lang="bash">
ls -r | grep -i '.*[0-9].*\.sar$' | xargs -I {} SAPCAR -xvf {}
</syntaxhighlight>Dies entpackt alle sar Archive, die eine Versionsnummer im Namen enthalten.

''Hinweis: vorher Kontrollieren ob mit der Ausgabe ls -r tatsächlich SAPEXEDB '''vor''' SAPEXE ausgegeben wird.''

SAP neustarten:
<syntaxhighlight lang="bash">
sapcontrol -nr 00 -function RestartSystem && watch -n 1 sapcontrol -nr 00 -function GetProcessList
</syntaxhighlight>

====Oracle Nacharbeiten====
Falls als DB die Oracle DB verwendet wird, müssen noch Berechtigungen nachgezogen werden, da es sonst zu Problemen mit BRTOOLS kommen kann.
Als root unter /sapmnt/<SID>/exe/nuc/linuxx86_64:
<syntaxhighlight lang="bash">
chmod -R 4775 brrestore brspace brrecover brconnect brbackup brarchive
chown -R oracle:oinstall brrestore brspace brrecover brconnect brbackup
</syntaxhighlight>

===Memory Parameter prüfen===
<syntaxhighlight lang="bash">
cdpro
sappfpar check pf=<profile_file>
</syntaxhighlight>

===JAVA===
====Sysinfo.xml generieren====
https://me.sap.com/notes/2293050
====Java patchen====
Patches unter /usr/sap/<SID>/SUM/ ablegen

Als <SID>adm anmelden
<syntaxhighlight lang="bash">
telnet localhost 50108
Administrator
<PW>
VERSION
add deploy
deploy /usr/sap/<SID>/SUM/ version_rule=all
</syntaxhighlight>
Eventuell SAP stoppen und wieder starten

<syntaxhighlight lang="bash">
dura9708:djpadm 73> ls -ltr
total 169596
-rw-r--r-- 1 djpadm sapsys 18675603 Apr 28 10:29 MESSAGING24P_21-80000682.SCA
-rw-r--r-- 1 djpadm sapsys 25564663 Apr 28 10:29 J2EEFRMW24P_4-80000439.SCA
-rw-r--r-- 1 djpadm sapsys 14020864 Apr 28 10:29 ENGINEAPI24P_4-80000618.SCA
-rw-r--r-- 1 djpadm sapsys 115196425 Apr 28 10:29 SERVERCORE24P_15-80000485.SCA
</syntaxhighlight>

===Codeschnipsel===

===Nützliche Links===

Adobe Document Server

2026-06-17T07:46:08Z

Hendrik: /* Update */

=== Beschreibung ===

=== Download ===

=== Installation ===

=== Konfiguration ===

=== Update ===
Aktuellen Patch herunterladen https://me.sap.com/softwarecenter/support/index

ADOBE DOCUMENT SERVICES 7.50 ADSSAP27P_2-80000623.SCA

SCA Archiv nach /usr/sap/<SID>/SUM/ hochladen

telnet localhost 50008

VERSION

add deploy

deploy /usr/sap/<SID>/SUM/ version_rule=all

=== Test ===

=== Codeschnipsel ===

=== Nützliche Links ===

SAP Cloud Connector

2026-06-16T13:06:24Z

Hendrik: /* Linux (SUSE/RHEL) */

== Beschreibung ==
Der '''SAP Cloud Connector''' ist eine on-premise Komponente, die eine sichere Verbindung zwischen SAP Business Technology Platform (BTP) Cloud-Anwendungen und on-premise SAP-Systemen herstellt. Er fungiert als sicherer Tunnel und Reverse-Proxy für die Kommunikation zwischen Cloud und On-Premise Umgebungen.

'''Hauptfunktionen:'''
* Sichere Verbindung zwischen SAP BTP und On-Premise Systemen
* SSL/TLS verschlüsselte Kommunikation
* Access Control für Backend-Systeme
* High Availability Unterstützung (Master/Shadow)

== Installation ==

=== Download ===
* URL: [https://tools.hana.ondemand.com/#cloud/ SAP Development Tools]
* Navigation: SAP BTP → Cloud Connector
* Aktuelle Version: 2.18.x oder höher

=== Linux (SUSE/RHEL) ===
<syntaxhighlight lang="bash">
# Download und Installation
sudo rpm -U com.sap.scc-ui-*.x86_64.rpm && sudo systemctl restart scc_daemon

# Version verifizieren
sudo cat /opt/sap/scc/version.properties
</syntaxhighlight>

=== Windows ===
<syntaxhighlight lang="cmd">
REM MSI Installer als Administrator ausführen
sapcc-*.msi

REM Service wird automatisch installiert und gestartet
REM Standardpfad: C:\SAP\scc

REM Service Status prüfen
sc query "SAP Cloud Connector"
</syntaxhighlight>

=== Initial Setup ===
<syntaxhighlight lang="bash">
# Admin UI aufrufen
# URL: https://localhost:8443

# Initial Credentials:
# User: Administrator
# Password: manage

# WICHTIG: Password sofort über "Change Password" ändern!
</syntaxhighlight>

== Update ==
<syntaxhighlight lang="bash">
rpm -U com.sap.scc-ui-<version>.rpm
</syntaxhighlight>

== Konfiguration ==
=== Passwort ändern ===
Anmelden und links im Menü auf "Configuration" klicken. Dann unter User Administration auf den Stift klicken.

=== Zertifikate erneuern ===
==== Systemzertifikat ====
Das Systemzertifikat ist self-signed und kann unter Configuration - On-Premises mit dem "Neu" Symbol einfach aktualisiert werden. DIe alten Daten werden hierbei übernommen.
==== Subaccount Zertifikat und Verbindung ====
Dies kann ebenfalls erneuert werden wenn man einen Subaccount auswählt und auf das Symbol "Zertifikat erneuern" klickt. Hierbei sind die Zugangsdaten vom entsprechenden Account einzugeben.

== Nützliche Links ==
* [https://help.sap.com/docs/connectivity/sap-btp-connectivity-cf/cloud-connector Cloud Connector Administration Guide]
* [https://tools.hana.ondemand.com/ SAP Development Tools Portal]
* [https://userapps.support.sap.com/sap/support/knowledge/en/2539713 SAP Note 2539713 - Cloud Connector Upgrade]
* [https://community.sap.com/topics/cloud-connector SAP Community - Cloud Connector]

Llama.cpp

2026-06-03T12:43:59Z

Hendrik:

'''Hinweis:''' Alle Pfade beziehen sich auf <code>/home/hendrik/Programme/llama.cpp/</code> als Installationsverzeichnis. Bitte entsprechend anpassen.

== Beschreibung ==

'''llama.cpp''' ist eine C/C++ Implementierung für Inference von Large Language Models (LLMs). Es unterstützt verschiedene Backends (CPU, Vulkan, ROCm, CUDA) und ermöglicht das Ausführen von quantisierten Modellen im GGUF-Format.

'''ik_llama.cpp''' ist ein optimierter Fork von llama.cpp mit zusätzlichen Performance-Verbesserungen und Unterstützung für neuere GPU-Architekturen (z.B. gfx1151 / Strix Halo).

== Download ==

'''llama.cpp:'''

<syntaxhighlight lang="bash" line="1">
git clone https://github.com/ggml-org/llama.cpp ~/Programme/llama.cpp
cd ~/Programme/llama.cpp
</syntaxhighlight>

'''ik_llama.cpp:'''

<syntaxhighlight lang="bash" line="1">
git clone https://github.com/ikawrakow/ik_llama.cpp ~/Programme/llama.cpp
cd ~/Programme/llama.cpp
</syntaxhighlight>

== Installation ==

=== Voraussetzungen ===

<syntaxhighlight lang="bash" line="1">
sudo dnf install libcurl-devel -y
</syntaxhighlight>

{| class="wikitable"
|-
! GPU !! Architektur !! Standard-Target
|-
| RX 7900 XTX || gfx1100 || '''ja'''
|-
| Strix Halo || gfx1151 || '''ja'''
|-
| MI50 / MI60 || gfx906 || separater Build (siehe unten)
|}

=== Vulkan Build ===

''Gilt für llama.cpp und ik_llama.cpp.''

[https://vulkan.lunarg.com/sdk/home Vulkan SDK] herunterladen, entpacken, ins Verzeichnis wechseln und <code>source setup-env.sh</code> ausführen. Dann ins llama.cpp-Verzeichnis wechseln:

<syntaxhighlight lang="bash" line="1">
source ~/Programme/Vulkan_SDK/1.4.335.0/setup-env.sh
cd ~/Programme/llama.cpp/
rm -R build-vulkan
cmake -B build-vulkan -DGGML_VULKAN=1 -DGGML_RPC=ON # llama.cpp
# cmake -B build-vulkan -DLLAMA_VULKAN=on -DGGML_RPC=ON # ik_llama.cpp
cmake --build build-vulkan --config Release -- -j $(nproc)
</syntaxhighlight>

=== ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0) ===

==== ROCm Quick-Install ====

Alle Befehle von der [https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/quick-start.html ROCm Quick-Start Seite] ausführen.

'''Wichtig für MI50 (gfx906) mit ROCm 6.4:''' 
Vor dem Reboot die rocblas-Bibliothek aus dem AUR ergänzen:

# [https://archlinux.org/packages/extra/x86_64/rocblas/ rocblas aus dem AUR] herunterladen (Version 6.4)
# Entpacken
# Alle Tensor-Dateien mit <code>gfx906</code> im Namen von <code>rocblas-…/opt/rocm/lib/rocblas/library</code> nach <code>/opt/rocm/lib/rocblas/library</code> kopieren
# Reboot
# Prüfen: <code>sudo update-alternatives --display rocm</code>

==== ROCm Umgebung einrichten (optional, bei Fehlern) ====

<syntaxhighlight lang="bash" line="1">
echo 'export LD_LIBRARY_PATH=/opt/rocm/lib:$LD_LIBRARY_PATH' >> ~/.bashrc
echo 'export HSA_OVERRIDE_GFX_VERSION=9.0.6' >> ~/.bashrc # Für MI50/MI60
source ~/.bashrc
sudo dnf install rocwmma-devel -y
</syntaxhighlight>

==== llama.cpp kompilieren (ROCm) ====

<syntaxhighlight lang="bash" line="1">
cd ~/Programme/llama.cpp/ && git pull && rm -R build-rocm-old && cp -R build-rocm build-rocm-old
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" cmake -B build-rocm -DGGML_HIP=ON -DGGML_RPC=ON -DAMDGPU_TARGETS=gfx1100,gfx1151 -DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 16
</syntaxhighlight>

'''Für MI50 (gfx906) — separater Build:'''

<syntaxhighlight lang="bash" line="1">
cd ~/Programme/llama.cpp/ && rm -R build-rocm-old && cp -R build-rocm build-rocm-old
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" cmake -B build-rocm -DGGML_HIP=ON -DGGML_RPC=ON -DAMDGPU_TARGETS=gfx906 -DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 16
</syntaxhighlight>

==== ik_llama.cpp kompilieren (ROCm) ====

<syntaxhighlight lang="bash" line="1">
cd ~/Programme/llama.cpp/ && rm -R build-rocm-old && cp -R build-rocm build-rocm-old
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" cmake -B build-rocm -DGGML_HIP=ON -DGGML_RPC=ON -DAMDGPU_TARGETS=gfx1100,gfx1151 -DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 16
</syntaxhighlight>

=== Build prüfen ===

<syntaxhighlight lang="bash" line="1">
./build/bin/llama-server --version
</syntaxhighlight>

== Konfiguration ==

=== llama-server als systemd Service einrichten ===

'''Service-Datei erstellen:'''

<syntaxhighlight lang="bash" line="1">
sudo nano /etc/systemd/system/llama-server.service
</syntaxhighlight>

'''Service-Datei Inhalt (Multi-GPU mit ROCm):'''

<syntaxhighlight lang="ini" line="1">
[Unit]
Description=Llama.cpp ROCm Multi-GPU Server
After=network.target

[Service]
Type=simple
User=hendrik
Group=hendrik
WorkingDirectory=/home/hendrik/Programme/llama.cpp/build/bin

# Multi-GPU Konfiguration
Environment="HIP_VISIBLE_DEVICES=0,1"
Environment="HSA_OVERRIDE_GFX_VERSION=9.0.6"
Environment="PATH=/opt/rocm/bin:/usr/local/bin:/usr/bin:/bin"
Environment="LD_LIBRARY_PATH=/opt/rocm/lib"

# Server mit optimalen Multi-GPU Einstellungen
ExecStart=/home/hendrik/Programme/llama.cpp/build/bin/llama-server \
-m /home/hendrik/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja

Restart=always
RestartSec=10
LimitNOFILE=65535
LimitMEMLOCK=infinity
StandardOutput=journal
StandardError=journal
SyslogIdentifier=llama-server

[Install]
WantedBy=multi-user.target
</syntaxhighlight>

'''Service aktivieren und starten:'''

<syntaxhighlight lang="bash" line="1">
sudo systemctl daemon-reload
sudo systemctl enable llama-server # Auto-Start beim Boot
sudo systemctl start llama-server
sudo systemctl status llama-server # Status prüfen
sudo journalctl -u llama-server -f # Logs verfolgen
</syntaxhighlight>

'''Service verwalten:'''

<syntaxhighlight lang="bash" line="1">
sudo systemctl stop llama-server # Stoppen
sudo systemctl restart llama-server # Neustarten
sudo systemctl disable llama-server # Autostart deaktivieren
</syntaxhighlight>

=== Manuelle Server-Starts ===

'''Multi-GPU (ROCm) — optimal:'''

<syntaxhighlight lang="bash" line="1">
HIP_VISIBLE_DEVICES=0,1 ./llama-server \
-m ~/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja
</syntaxhighlight>

=== RPC — Verteiltes Inference über mehrere Hosts ===

Mit <code>-DGGML_RPC=ON</code> wird der <code>rpc-server</code> mitgebaut, der es ermöglicht, die Inferenz über das Netzwerk auf mehrere Maschinen zu verteilen. Jeder Remote-Host stellt seine GPU(s) über einen rpc-server zur Verfügung; der Main-Host verbindet sich darüber mit <code>--rpc</code>.

'''Voraussetzung:''' Auf '''allen''' Hosts muss llama.cpp mit <code>-DGGML_RPC=ON</code> kompiliert sein (siehe Build-Abschnitte oben). Die Netzwerkverbindung zwischen den Hosts muss bestehen.

==== 1. RPC-Server auf allen Remote-Hosts starten ====

Auf jedem Host, der seine GPU zur Verfügung stellen soll:

<syntaxhighlight lang="bash" line="1">
cd ~/Programme/llama.cpp/
./build/bin/rpc-server --host 0.0.0.0
</syntaxhighlight>

Standardmäßig lauscht der rpc-server auf Port '''50052'''. Mit <code>--port</code> kann ein anderer Port gewählt werden.

==== 2. llama-server auf dem Main-Host starten ====

Auf dem Host, der die Inferenz koordiniert:

<syntaxhighlight lang="bash" line="1">
./llama-server \
--model ~/.lmstudio/models/unsloth/Step-3.7-Flash-GGUF/Step-3.7-Flash-UD-IQ4_XS-00001-of-00003.gguf \
-ngl 99 \
--rpc localhost:50052,192.168.1.19:50052
</syntaxhighlight>

Dabei ist <code>localhost:50052</code> die lokale GPU und <code>192.168.1.19:50052</code> die Remote-GPU.

==== rpc-server als systemd Service (Remote-Host) ====

'''Service-Datei erstellen:'''

<syntaxhighlight lang="bash" line="1">
sudo nano /etc/systemd/system/llama-rpc.service
</syntaxhighlight>

'''Service-Datei Inhalt:'''

<syntaxhighlight lang="ini" line="1">
[Unit]
Description=Llama.cpp RPC Server (Remote GPU)
After=network.target

[Service]
Type=simple
User=hendrik
Group=hendrik
WorkingDirectory=/home/hendrik/Programme/llama.cpp/build/bin

Environment="HIP_VISIBLE_DEVICES=0"
Environment="PATH=/opt/rocm/bin:/usr/local/bin:/usr/bin:/bin"
Environment="LD_LIBRARY_PATH=/opt/rocm/lib"

ExecStart=/home/hendrik/Programme/llama.cpp/build/bin/rpc-server --host 0.0.0.0

Restart=always
RestartSec=10
StandardOutput=journal
StandardError=journal
SyslogIdentifier=llama-rpc

[Install]
WantedBy=multi-user.target
</syntaxhighlight>

'''Aktivieren:'''

<syntaxhighlight lang="bash" line="1">
sudo systemctl daemon-reload
sudo systemctl enable --now llama-rpc
</syntaxhighlight>

== Update ==

<syntaxhighlight lang="bash" line="1">
cd ~/Programme/llama.cpp
git pull
cmake --build build --config Release -- -j $(nproc)

# Services neu starten falls aktiv
sudo systemctl restart llama-server
sudo systemctl restart llama-rpc # falls RPC-Service läuft
</syntaxhighlight>

== Parameter-Referenz ==

{| class="wikitable"
|-
! Parameter !! Beschreibung
|-
| <code>-ngl 99</code> || Alle Layer auf GPU auslagern (Langform: <code>--gpu-layers</code>)
|-
| <code>-fa 1</code> || Flash Attention aktivieren
|-
| <code>-c 32768</code> || Kontextfenster-Größe
|-
| <code>-b 2048</code> || Batch-Size
|-
| <code>-ub 2048</code> || Physical USB batch size
|-
| <code>--split-mode row</code> || Tensor-Split-Modus (Multi-GPU)
|-
| <code>--tensor-split 0.5,0.5</code> || Gleichmäßige Aufteilung auf 2 GPUs
|-
| <code>--parallel 1</code> || Anzahl paralleler Sequenzen
|-
| <code>--jinja</code> || Jinja-Template-Support für Chat
|-
| <code>--rpc host:port,…</code> || RPC-Backends für verteiltes Inference
|-
| <code>-DGGML_RPC=ON</code> || CMake-Flag: RPC-Support einkompilieren
|}

Llama.cpp

2026-06-03T12:38:30Z

Hendrik:

== Beschreibung ==

'''llama.cpp''' ist eine C/C++ Implementierung für Inference von Large Language Models (LLMs). Es unterstützt verschiedene Backends (CPU, Vulkan, ROCm, CUDA) und ermöglicht das Ausführen von quantisierten Modellen im GGUF-Format.

'''ik_llama.cpp''' ist ein optimierter Fork von llama.cpp mit zusätzlichen Performance-Verbesserungen und Unterstützung für neuere GPU-Architekturen (z.B. gfx1151 / Strix Halo).

== Download ==

<syntaxhighlight lang="bash" line="1">
git clone https://github.com/ggml-org/llama.cpp
cd llama.cpp
</syntaxhighlight>

Für ik_llama.cpp den entsprechenden Fork klonen.

== Installation ==

=== Vulkan Build ===

''Gilt für llama.cpp und ik_llama.cpp.''

<syntaxhighlight lang="bash" line="1">
sudo dnf install libcurl-devel -y
</syntaxhighlight>

[https://vulkan.lunarg.com/sdk/home Vulkan SDK] herunterladen, entpacken, ins Verzeichnis wechseln und <code>source setup-env.sh</code> ausführen. Dann ins llama.cpp-Verzeichnis wechseln:

<syntaxhighlight lang="bash" line="1">
source /home/hendrik/Programme/Vulkan_SDK/1.4.335.0/setup-env.sh
cd /home/hendrik/Programme/llama.cpp/
rm -R build-vulkan
cmake -B build-vulkan -DGGML_VULKAN=1 -DGGML_RPC=ON # llama.cpp
# cmake -B build-vulkan -DLLAMA_VULKAN=on -DGGML_RPC=ON # ik_llama.cpp
cmake --build build-vulkan --config Release -- -j $(nproc)
</syntaxhighlight>

=== ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0) ===

{| class="wikitable"
|-
! GPU !! Architektur !! Bemerkung
|-
| MI50 / MI60 || gfx906 || nur llama.cpp
|-
| RX 7900 XTX || gfx1100 ||
|-
| Strix Halo || gfx1151 || nur ik_llama.cpp
|}

==== ROCm Quick-Install ===

Alle Befehle von der [https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/quick-start.html ROCm Quick-Start Seite] ausführen.

'''Wichtig für MI50 (gfx906) mit ROCm 6.4:''' 
Vor dem Reboot die rocblas-Bibliothek aus dem AUR ergänzen:

# [https://archlinux.org/packages/extra/x86_64/rocblas/ rocblas aus dem AUR] herunterladen (Version 6.4)
# Entpacken
# Alle Tensor-Dateien mit <code>gfx906</code> im Namen von <code>rocblas-…/opt/rocm/lib/rocblas/library</code> nach <code>/opt/rocm/lib/rocblas/library</code> kopieren
# Reboot
# Prüfen: <code>sudo update-alternatives --display rocm</code>

==== ROCm Umgebung einrichten (optional, bei Fehlern) ====

<syntaxhighlight lang="bash" line="1">
echo 'export LD_LIBRARY_PATH=/opt/rocm/lib:$LD_LIBRARY_PATH' >> ~/.bashrc
echo 'export HSA_OVERRIDE_GFX_VERSION=9.0.6' >> ~/.bashrc # Für MI50/MI60
source ~/.bashrc
</syntaxhighlight>

<syntaxhighlight lang="bash" line="1">
sudo dnf install rocwmma-devel -y
</syntaxhighlight>

==== llama.cpp kompilieren (ROCm) ====

<syntaxhighlight lang="bash" line="1">
cd /home/hendrik/Programme/llama.cpp/ && git pull && rm -R build-rocm-old && cp -R build-rocm build-rocm-old
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" cmake -B build-rocm -DGGML_HIP=ON -DGGML_RPC=ON -DAMDGPU_TARGETS=gfx1100 -DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 16
</syntaxhighlight>

Für mehrere GPU-Architekturen gleichzeitig:

<syntaxhighlight lang="bash" line="1">
# Für MI50 (gfx906) und RX 7900 XTX (gfx1100)
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
cmake -S . -B build -DGGML_HIP=ON -DGGML_RPC=ON -DAMDGPU_TARGETS="gfx906;gfx1100" \
-DCMAKE_BUILD_TYPE=Release && \
cmake --build build --config Release -- -j 8
</syntaxhighlight>

==== ik_llama.cpp kompilieren (ROCm) ====

<syntaxhighlight lang="bash" line="1">
cd /home/hendrik/Programme/llama.cpp/ && rm -R build-rocm-old && cp -R build-rocm build-rocm-old
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" cmake -B build-rocm -DGGML_HIP=ON -DGGML_RPC=ON -DAMDGPU_TARGETS=gfx1100,gfx1151 -DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 16
</syntaxhighlight>

== Konfiguration ==

=== llama-server als systemd Service einrichten ===

'''Service-Datei erstellen:'''

<syntaxhighlight lang="bash" line="1">
sudo nano /etc/systemd/system/llama-server.service
</syntaxhighlight>

'''Service-Datei Inhalt (Multi-GPU mit ROCm):'''

<syntaxhighlight lang="ini" line="1">
[Unit]
Description=Llama.cpp ROCm Multi-GPU Server
After=network.target

[Service]
Type=simple
User=username
Group=username
WorkingDirectory=/home/username/Programme/llama.cpp/build/bin

# Multi-GPU Konfiguration
Environment="HIP_VISIBLE_DEVICES=0,1"
Environment="HSA_OVERRIDE_GFX_VERSION=9.0.6"
Environment="PATH=/opt/rocm/bin:/usr/local/bin:/usr/bin:/bin"
Environment="LD_LIBRARY_PATH=/opt/rocm/lib"

# Server mit optimalen Multi-GPU Einstellungen
ExecStart=/home/username/Programme/llama.cpp/build/bin/llama-server \
-m /home/username/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja

Restart=always
RestartSec=10
LimitNOFILE=65535
LimitMEMLOCK=infinity
StandardOutput=journal
StandardError=journal
SyslogIdentifier=llama-server

[Install]
WantedBy=multi-user.target
</syntaxhighlight>

'''Service aktivieren und starten:'''

<syntaxhighlight lang="bash" line="1">
sudo systemctl daemon-reload
sudo systemctl enable llama-server # Auto-Start beim Boot
sudo systemctl start llama-server
sudo systemctl status llama-server # Status prüfen
sudo journalctl -u llama-server -f # Logs verfolgen
</syntaxhighlight>

'''Service verwalten:'''

<syntaxhighlight lang="bash" line="1">
sudo systemctl stop llama-server # Stoppen
sudo systemctl restart llama-server # Neustarten
sudo systemctl disable llama-server # Autostart deaktivieren
</syntaxhighlight>

=== Manuelle Server-Starts ===

'''Multi-GPU (ROCm) — optimal:'''

<syntaxhighlight lang="bash" line="1">
HIP_VISIBLE_DEVICES=0,1 ./llama-server \
-m ~/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja
</syntaxhighlight>

=== RPC — Verteiltes Inference über mehrere Hosts ===

Mit <code>-DGGML_RPC=ON</code> wird der <code>rpc-server</code> mitgebaut, der es ermöglicht, die Inferenz über das Netzwerk auf mehrere Maschinen zu verteilen. Jeder Remote-Host stellt seine GPU(s) über einen rpc-server zur Verfügung; der Main-Host verbindet sich darüber mit <code>--rpc</code>.

'''Voraussetzung:''' Auf '''allen''' Hosts muss llama.cpp mit <code>-DGGML_RPC=ON</code> kompiliert sein (siehe Build-Abschnitte oben). Die Netzwerkverbindung zwischen den Hosts muss bestehen.

==== 1. RPC-Server auf allen Remote-Hosts starten ====

Auf jedem Host, der seine GPU zur Verfügung stellen soll:

<syntaxhighlight lang="bash" line="1">
cd /home/hendrik/Programme/llama.cpp/
./build/bin/rpc-server --host 0.0.0.0
</syntaxhighlight>

Standardmäßig lauscht der rpc-server auf Port '''50052'''. Mit <code>--port</code> kann ein anderer Port gewählt werden.

==== 2. llama-server auf dem Main-Host starten ====

Auf dem Host, der die Inferenz koordiniert:

<syntaxhighlight lang="bash" line="1">
./llama-server \
--model /home/hendrik/.lmstudio/models/unsloth/Step-3.7-Flash-GGUF/Step-3.7-Flash-UD-IQ4_XS-00001-of-00003.gguf \
--gpu-layers 99 \
--rpc localhost:50052,192.168.1.19:50052
</syntaxhighlight>

Dabei ist <code>localhost:50052</code> die lokale GPU und <code>192.168.1.19:50052</code> die Remote-GPU.

==== rpc-server als systemd Service (Remote-Host) ====

'''Service-Datei erstellen:'''

<syntaxhighlight lang="bash" line="1">
sudo nano /etc/systemd/system/llama-rpc.service
</syntaxhighlight>

'''Service-Datei Inhalt:'''

<syntaxhighlight lang="ini" line="1">
[Unit]
Description=Llama.cpp RPC Server (Remote GPU)
After=network.target

[Service]
Type=simple
User=username
Group=username
WorkingDirectory=/home/username/Programme/llama.cpp/build/bin

Environment="HIP_VISIBLE_DEVICES=0"
Environment="PATH=/opt/rocm/bin:/usr/local/bin:/usr/bin:/bin"
Environment="LD_LIBRARY_PATH=/opt/rocm/lib"

ExecStart=/home/username/Programme/llama.cpp/build/bin/rpc-server --host 0.0.0.0

Restart=always
RestartSec=10
StandardOutput=journal
StandardError=journal
SyslogIdentifier=llama-rpc

[Install]
WantedBy=multi-user.target
</syntaxhighlight>

'''Aktivieren:'''

<syntaxhighlight lang="bash" line="1">
sudo systemctl daemon-reload
sudo systemctl enable --now llama-rpc
</syntaxhighlight>

== Update ==

<syntaxhighlight lang="bash" line="1">
cd ~/llama.cpp
git pull
cmake --build build --config Release -- -j $(nproc)

# Service neu starten falls aktiv
sudo systemctl restart llama-server
sudo systemctl restart llama-rpc # falls RPC-Service läuft
</syntaxhighlight>

== Parameter-Referenz ==

{| class="wikitable"
|-
! Parameter !! Beschreibung
|-
| <code>-ngl 99</code> || Alle Layer auf GPU auslagern
|-
| <code>-fa 1</code> || Flash Attention aktivieren
|-
| <code>-c 32768</code> || Kontextfenster-Größe
|-
| <code>-b 2048</code> || Batch-Size
|-
| <code>-ub 2048</code> || Physical USB batch size
|-
| <code>--split-mode row</code> || Tensor-Split-Modus (Multi-GPU)
|-
| <code>--tensor-split 0.5,0.5</code> || Gleichmäßige Aufteilung auf 2 GPUs
|-
| <code>--parallel 1</code> || Anzahl paralleler Sequenzen
|-
| <code>--jinja</code> || Jinja-Template-Support für Chat
|-
| <code>--rpc host:port,…</code> || RPC-Backends für verteiltes Inference
|-
| <code>-DGGML_RPC=ON</code> || CMake-Flag: RPC-Support einkompilieren
|}

Llama.cpp

2026-05-31T11:26:21Z

Hendrik: /* ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 */

=== Beschreibung ===
llama.cpp ist eine C/C++ Implementierung für Inference von Large Language Models (LLMs). Es unterstützt verschiedene Backends (CPU, Vulkan, ROCm, CUDA) und ermöglicht das Ausführen von quantisierten Modellen im GGUF-Format.

=== Download ===
<syntaxhighlight lang="bash" line="1">
git clone https://github.com/ggml-org/llama.cpp
cd llama.cpp
</syntaxhighlight>

=== Installation ===
==== Vulkan Build ====
<syntaxhighlight lang="bash" line="1">
sudo dnf install libcurl-devel -y
</syntaxhighlight>
[https://vulkan.lunarg.com/sdk/home Vulkan SDK] herunterladen
entpacken, ins Verzeichnis wechseln und source setup-env.sh ausführen. Dann ins Verzeichnis wechseln wo llama.cpp installiert werden soll.
Dort dann
<syntaxhighlight lang="bash" line="1">
source /home/hendrik/Programme/Vulkan_SDK/1.4.335.0/setup-env.sh
cd /home/hendrik/Programme/llama.cpp/
rm -R build-vulkan
cmake -B build-vulkan -DGGML_VULKAN=1
cmake --build build-vulkan --config Release -- -j $(nproc)
</syntaxhighlight>

==== ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 ====
gfx906 = Mi 50 Support 
gfx1100 = 7900 XTX Support

<syntaxhighlight lang="bash" line="1">
Copy & paste all the commands from the quick install https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/quick-start.html
Before rebooting to complete the install, download the 6.4 rocblas from the AUR: https://archlinux.org/packages/extra/x86_64/rocblas/
Extract it
Copy all tensor files that contain gfx906 in rocblas-6.4.3-3-x86_64.pkg/opt/rocm/lib/rocblas/library to /opt/rocm/lib/rocblas/library
Reboot
Check if it worked by running sudo update-alternatives --display rocm
</syntaxhighlight>
# ROCm Umgebung einrichten (optional falls Fehler auftreten)
<syntaxhighlight lang="bash" line="1">
echo 'export LD_LIBRARY_PATH=/opt/rocm/lib:$LD_LIBRARY_PATH' >> ~/.bashrc
echo 'export HSA_OVERRIDE_GFX_VERSION=9.0.6' >> ~/.bashrc # Für MI50/MI60
source ~/.bashrc
</syntaxhighlight>
<syntaxhighlight lang="bash" line="1">
sudo dnf install rocwmma-devel -y
</syntaxhighlight>
# llama.cpp kompilieren
<syntaxhighlight lang="bash" line="1">
cd /home/hendrik/Programme/llama.cpp/ && git pull && rm -R build-rocm-old && cp -R build-rocm build-rocm-old
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" cmake -B build-rocm -DGGML_HIP=ON -DAMDGPU_TARGETS=gfx1100 -DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 16
</syntaxhighlight>

Für mehrere GPU-Architekturen gleichzeitig:
<syntaxhighlight lang="bash" line="1">
# Für MI50 (gfx906) und RX 7900 XTX (gfx1100)
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
cmake -S . -B build -DGGML_HIP=ON -DAMDGPU_TARGETS="gfx906;gfx1100" \
-DCMAKE_BUILD_TYPE=Release && \
cmake --build build --config Release -- -j 8
</syntaxhighlight>

=== ik_llama.cpp ===
==== Vulkan Build ====
<syntaxhighlight lang="bash" line="1">
sudo dnf install libcurl-devel -y
</syntaxhighlight>
[https://vulkan.lunarg.com/sdk/home Vulkan SDK] herunterladen
entpacken, ins Verzeichnis wechseln und source setup-env.sh ausführen. Dann ins Verzeichnis wechseln wo llama.cpp installiert werden soll.
Dort dann
<syntaxhighlight lang="bash" line="1">
source /home/hendrik/Programme/Vulkan_SDK/1.4.335.0/setup-env.sh
cd /home/hendrik/Programme/llama.cpp/
rm -R build-vulkan
cmake -B build-vulkan -DLLAMA_VULKAN=on
cmake --build build-vulkan --config Release -- -j $(nproc)
</syntaxhighlight>

==== ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 ====
gfx906 = Mi 50 Support 
gfx1100 = 7900 XTX Support
gfx1151 = Strix Halo
# llama.cpp kompilieren
<syntaxhighlight lang="bash" line="1">
cd /home/hendrik/Programme/llama.cpp/ && rm -R build-rocm-old && cp -R build-rocm build-rocm-old
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" cmake -B build-rocm -DGGML_HIP=ON -DAMDGPU_TARGETS=gfx1100,gfx1151 -DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 16
</syntaxhighlight>

=== Konfiguration ===

==== llama-server als systemd Service einrichten ====

'''Service-Datei erstellen:'''
<syntaxhighlight lang="bash" line="1">
sudo nano /etc/systemd/system/llama-server.service
</syntaxhighlight>

'''Service-Datei Inhalt (Multi-GPU mit ROCm):'''
<syntaxhighlight lang="ini" line="1">
[Unit]
Description=Llama.cpp ROCm Multi-GPU Server
After=network.target

[Service]
Type=simple
User=username
Group=username
WorkingDirectory=/home/username/llama.cpp/build/bin

# Multi-GPU Konfiguration
Environment="HIP_VISIBLE_DEVICES=0,1"
Environment="HSA_OVERRIDE_GFX_VERSION=9.0.6"
Environment="PATH=/opt/rocm/bin:/usr/local/bin:/usr/bin:/bin"
Environment="LD_LIBRARY_PATH=/opt/rocm/lib"

# Server mit optimalen Multi-GPU Einstellungen
ExecStart=/home/username/llama.cpp/build/bin/llama-server \
-m /home/username/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja

Restart=always
RestartSec=10
LimitNOFILE=65535
LimitMEMLOCK=infinity
StandardOutput=journal
StandardError=journal
SyslogIdentifier=llama-server

[Install]
WantedBy=multi-user.target
</syntaxhighlight>

'''Service aktivieren und starten:'''
<syntaxhighlight lang="bash" line="1">
# Service neu laden
sudo systemctl daemon-reload

# Service aktivieren (Auto-Start beim Boot)
sudo systemctl enable llama-server

# Service starten
sudo systemctl start llama-server

# Status prüfen
sudo systemctl status llama-server

# Logs anzeigen
sudo journalctl -u llama-server -f
</syntaxhighlight>

'''Service Management:'''
<syntaxhighlight lang="bash" line="1">
# Service stoppen
sudo systemctl stop llama-server

# Service neustarten
sudo systemctl restart llama-server

# Service deaktivieren
sudo systemctl disable llama-server

==== Manuelle Server-Starts ====
'''Multi-GPU (ROCm) - Optimal:'''
<syntaxhighlight lang="bash" line="1">
HIP_VISIBLE_DEVICES=0,1 ./llama-server \
-m ~/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja
</syntaxhighlight>

=== Update ===
<syntaxhighlight lang="bash" line="1">
cd ~/llama.cpp
git pull
cmake --build build --config Release -- -j $(nproc)

# Service neu starten falls aktiv
sudo systemctl restart llama-server
</syntaxhighlight>
=== Befehlszeilenargumente ===
llama-server
<syntaxhighlight lang="bash" line="1">
-h, --help, --usage print usage and exit
--version show version and build info
-cl, --cache-list show list of models in cache
--completion-bash print source-able bash completion script for llama.cpp
--verbose-prompt print a verbose prompt before generation (default: false)
-t, --threads N number of CPU threads to use during generation (default: -1)
(env: LLAMA_ARG_THREADS)
-tb, --threads-batch N number of threads to use during batch and prompt processing (default:
same as --threads)
-C, --cpu-mask M CPU affinity mask: arbitrarily long hex. Complements cpu-range
(default: "")
-Cr, --cpu-range lo-hi range of CPUs for affinity. Complements --cpu-mask
--cpu-strict <0|1> use strict CPU placement (default: 0)
--prio N set process/thread priority : low(-1), normal(0), medium(1), high(2),
realtime(3) (default: 0)
--poll <0...100> use polling level to wait for work (0 - no polling, default: 50)
-Cb, --cpu-mask-batch M CPU affinity mask: arbitrarily long hex. Complements cpu-range-batch
(default: same as --cpu-mask)
-Crb, --cpu-range-batch lo-hi ranges of CPUs for affinity. Complements --cpu-mask-batch
--cpu-strict-batch <0|1> use strict CPU placement (default: same as --cpu-strict)
--prio-batch N set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime
(default: 0)
--poll-batch <0|1> use polling to wait for work (default: same as --poll)
-c, --ctx-size N size of the prompt context (default: 4096, 0 = loaded from model)
(env: LLAMA_ARG_CTX_SIZE)
-n, --predict, --n-predict N number of tokens to predict (default: -1, -1 = infinity)
(env: LLAMA_ARG_N_PREDICT)
-b, --batch-size N logical maximum batch size (default: 2048)
(env: LLAMA_ARG_BATCH)
-ub, --ubatch-size N physical maximum batch size (default: 512)
(env: LLAMA_ARG_UBATCH)
--keep N number of tokens to keep from the initial prompt (default: 0, -1 =
all)
--swa-full use full-size SWA cache (default: false)
[(more
info)](https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
(env: LLAMA_ARG_SWA_FULL)
--kv-unified, -kvu use single unified KV buffer for the KV cache of all sequences
(default: false)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/14363)
(env: LLAMA_ARG_KV_SPLIT)
-fa, --flash-attn [on|off|auto] set Flash Attention use ('on', 'off', or 'auto', default: 'auto')
(env: LLAMA_ARG_FLASH_ATTN)
--no-perf disable internal libllama performance timings (default: false)
(env: LLAMA_ARG_NO_PERF)
-e, --escape process escapes sequences (\n, \r, \t, \', \", \\) (default: true)
--no-escape do not process escape sequences
--rope-scaling {none,linear,yarn} RoPE frequency scaling method, defaults to linear unless specified by
the model
(env: LLAMA_ARG_ROPE_SCALING_TYPE)
--rope-scale N RoPE context scaling factor, expands context by a factor of N
(env: LLAMA_ARG_ROPE_SCALE)
--rope-freq-base N RoPE base frequency, used by NTK-aware scaling (default: loaded from
model)
(env: LLAMA_ARG_ROPE_FREQ_BASE)
--rope-freq-scale N RoPE frequency scaling factor, expands context by a factor of 1/N
(env: LLAMA_ARG_ROPE_FREQ_SCALE)
--yarn-orig-ctx N YaRN: original context size of model (default: 0 = model training
context size)
(env: LLAMA_ARG_YARN_ORIG_CTX)
--yarn-ext-factor N YaRN: extrapolation mix factor (default: -1.0, 0.0 = full
interpolation)
(env: LLAMA_ARG_YARN_EXT_FACTOR)
--yarn-attn-factor N YaRN: scale sqrt(t) or attention magnitude (default: -1.0)
(env: LLAMA_ARG_YARN_ATTN_FACTOR)
--yarn-beta-slow N YaRN: high correction dim or alpha (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_SLOW)
--yarn-beta-fast N YaRN: low correction dim or beta (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_FAST)
-nkvo, --no-kv-offload disable KV offload
(env: LLAMA_ARG_NO_KV_OFFLOAD)
-nr, --no-repack disable weight repacking
(env: LLAMA_ARG_NO_REPACK)
--no-host bypass host buffer allowing extra buffers to be used
(env: LLAMA_ARG_NO_HOST)
-ctk, --cache-type-k TYPE KV cache data type for K
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K)
-ctv, --cache-type-v TYPE KV cache data type for V
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V)
-dt, --defrag-thold N KV cache defragmentation threshold (DEPRECATED)
(env: LLAMA_ARG_DEFRAG_THOLD)
-np, --parallel N number of parallel sequences to decode (default: 1)
(env: LLAMA_ARG_N_PARALLEL)
--mlock force system to keep model in RAM rather than swapping or compressing
(env: LLAMA_ARG_MLOCK)
--no-mmap do not memory-map model (slower load but may reduce pageouts if not
using mlock)
(env: LLAMA_ARG_NO_MMAP)
--numa TYPE attempt optimizations that help on some NUMA systems
- distribute: spread execution evenly over all nodes
- isolate: only spawn threads on CPUs on the node that execution
started on
- numactl: use the CPU map provided by numactl
if run without this previously, it is recommended to drop the system
page cache before using this
see https://github.com/ggml-org/llama.cpp/issues/1437
(env: LLAMA_ARG_NUMA)
-dev, --device <dev1,dev2,..> comma-separated list of devices to use for offloading (none = don't
offload)
use --list-devices to see a list of available devices
(env: LLAMA_ARG_DEVICE)
--list-devices print list of available devices and exit
--override-tensor, -ot <tensor name pattern>=<buffer type>,...
override tensor buffer type
--cpu-moe, -cmoe keep all Mixture of Experts (MoE) weights in the CPU
(env: LLAMA_ARG_CPU_MOE)
--n-cpu-moe, -ncmoe N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU
(env: LLAMA_ARG_N_CPU_MOE)
-ngl, --gpu-layers, --n-gpu-layers N max. number of layers to store in VRAM (default: -1)
(env: LLAMA_ARG_N_GPU_LAYERS)
-sm, --split-mode {none,layer,row} how to split the model across multiple GPUs, one of:
- none: use one GPU only
- layer (default): split layers and KV across GPUs
- row: split rows across GPUs
(env: LLAMA_ARG_SPLIT_MODE)
-ts, --tensor-split N0,N1,N2,... fraction of the model to offload to each GPU, comma-separated list of
proportions, e.g. 3,1
(env: LLAMA_ARG_TENSOR_SPLIT)
-mg, --main-gpu INDEX the GPU to use for the model (with split-mode = none), or for
intermediate results and KV (with split-mode = row) (default: 0)
(env: LLAMA_ARG_MAIN_GPU)
--check-tensors check model tensor data for invalid values (default: false)
--override-kv KEY=TYPE:VALUE advanced option to override model metadata by key. may be specified
multiple times.
types: int, float, bool, str. example: --override-kv
tokenizer.ggml.add_bos_token=bool:false
--no-op-offload disable offloading host tensor operations to device (default: false)
--lora FNAME path to LoRA adapter (can be repeated to use multiple adapters)
--lora-scaled FNAME SCALE path to LoRA adapter with user defined scaling (can be repeated to use
multiple adapters)
--control-vector FNAME add a control vector
note: this argument can be repeated to add multiple control vectors
--control-vector-scaled FNAME SCALE add a control vector with user defined scaling SCALE
note: this argument can be repeated to add multiple scaled control
vectors
--control-vector-layer-range START END
layer range to apply the control vector(s) to, start and end inclusive
-m, --model FNAME model path (default: `models/$filename` with filename from `--hf-file`
or `--model-url` if set, otherwise models/7B/ggml-model-f16.gguf)
(env: LLAMA_ARG_MODEL)
-mu, --model-url MODEL_URL model download url (default: unused)
(env: LLAMA_ARG_MODEL_URL)
-dr, --docker-repo [<repo>/]<model>[:quant]
Docker Hub model repository. repo is optional, default to ai/. quant
is optional, default to :latest.
example: gemma3
(default: unused)
(env: LLAMA_ARG_DOCKER_REPO)
-hf, -hfr, --hf-repo <user>/<model>[:quant]
Hugging Face model repository; quant is optional, case-insensitive,
default to Q4_K_M, or falls back to the first file in the repo if
Q4_K_M doesn't exist.
mmproj is also downloaded automatically if available. to disable, add
--no-mmproj
example: unsloth/phi-4-GGUF:q4_k_m
(default: unused)
(env: LLAMA_ARG_HF_REPO)
-hfd, -hfrd, --hf-repo-draft <user>/<model>[:quant]
Same as --hf-repo, but for the draft model (default: unused)
(env: LLAMA_ARG_HFD_REPO)
-hff, --hf-file FILE Hugging Face model file. If specified, it will override the quant in
--hf-repo (default: unused)
(env: LLAMA_ARG_HF_FILE)
-hfv, -hfrv, --hf-repo-v <user>/<model>[:quant]
Hugging Face model repository for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_REPO_V)
-hffv, --hf-file-v FILE Hugging Face model file for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_FILE_V)
-hft, --hf-token TOKEN Hugging Face access token (default: value from HF_TOKEN environment
variable)
(env: HF_TOKEN)
--log-disable Log disable
--log-file FNAME Log to file
--log-colors [on|off|auto] Set colored logging ('on', 'off', or 'auto', default: 'auto')
'auto' enables colors when output is to a terminal
(env: LLAMA_LOG_COLORS)
-v, --verbose, --log-verbose Set verbosity level to infinity (i.e. log all messages, useful for
debugging)
--offline Offline mode: forces use of cache, prevents network access
(env: LLAMA_OFFLINE)
-lv, --verbosity, --log-verbosity N Set the verbosity threshold. Messages with a higher verbosity will be
ignored.
(env: LLAMA_LOG_VERBOSITY)
--log-prefix Enable prefix in log messages
(env: LLAMA_LOG_PREFIX)
--log-timestamps Enable timestamps in log messages
(env: LLAMA_LOG_TIMESTAMPS)
-ctkd, --cache-type-k-draft TYPE KV cache data type for K for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K_DRAFT)
-ctvd, --cache-type-v-draft TYPE KV cache data type for V for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V_DRAFT)

----- sampling params -----

--samplers SAMPLERS samplers that will be used for generation in the order, separated by
';'
(default:
penalties;dry;top_n_sigma;top_k;typ_p;top_p;min_p;xtc;temperature)
-s, --seed SEED RNG seed (default: -1, use random seed for -1)
--sampling-seq, --sampler-seq SEQUENCE
simplified sequence for samplers that will be used (default:
edskypmxt)
--ignore-eos ignore end of stream token and continue generating (implies
--logit-bias EOS-inf)
--temp N temperature (default: 0.8)
--top-k N top-k sampling (default: 40, 0 = disabled)
--top-p N top-p sampling (default: 0.9, 1.0 = disabled)
--min-p N min-p sampling (default: 0.1, 0.0 = disabled)
--top-nsigma N top-n-sigma sampling (default: -1.0, -1.0 = disabled)
--xtc-probability N xtc probability (default: 0.0, 0.0 = disabled)
--xtc-threshold N xtc threshold (default: 0.1, 1.0 = disabled)
--typical N locally typical sampling, parameter p (default: 1.0, 1.0 = disabled)
--repeat-last-n N last n tokens to consider for penalize (default: 64, 0 = disabled, -1
= ctx_size)
--repeat-penalty N penalize repeat sequence of tokens (default: 1.0, 1.0 = disabled)
--presence-penalty N repeat alpha presence penalty (default: 0.0, 0.0 = disabled)
--frequency-penalty N repeat alpha frequency penalty (default: 0.0, 0.0 = disabled)
--dry-multiplier N set DRY sampling multiplier (default: 0.0, 0.0 = disabled)
--dry-base N set DRY sampling base value (default: 1.75)
--dry-allowed-length N set allowed length for DRY sampling (default: 2)
--dry-penalty-last-n N set DRY penalty for the last n tokens (default: -1, 0 = disable, -1 =
context size)
--dry-sequence-breaker STRING add sequence breaker for DRY sampling, clearing out default breakers
('\n', ':', '"', '*') in the process; use "none" to not use any
sequence breakers
--dynatemp-range N dynamic temperature range (default: 0.0, 0.0 = disabled)
--dynatemp-exp N dynamic temperature exponent (default: 1.0)
--mirostat N use Mirostat sampling.
Top K, Nucleus and Locally Typical samplers are ignored if used.
(default: 0, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)
--mirostat-lr N Mirostat learning rate, parameter eta (default: 0.1)
--mirostat-ent N Mirostat target entropy, parameter tau (default: 5.0)
-l, --logit-bias TOKEN_ID(+/-)BIAS modifies the likelihood of token appearing in the completion,
i.e. `--logit-bias 15043+1` to increase likelihood of token ' Hello',
or `--logit-bias 15043-1` to decrease likelihood of token ' Hello'
--grammar GRAMMAR BNF-like grammar to constrain generations (see samples in grammars/
dir) (default: '')
--grammar-file FNAME file to read grammar from
-j, --json-schema SCHEMA JSON schema to constrain generations (https://json-schema.org/), e.g.
`{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead
-jf, --json-schema-file FILE File containing a JSON schema to constrain generations
(https://json-schema.org/), e.g. `{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead

----- example-specific params -----

--ctx-checkpoints, --swa-checkpoints N
max number of context checkpoints to create per slot (default: 8)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/15293)
(env: LLAMA_ARG_CTX_CHECKPOINTS)
--cache-ram, -cram N set the maximum cache size in MiB (default: 8192, -1 - no limit, 0 -
disable)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/16391)
(env: LLAMA_ARG_CACHE_RAM)
--no-context-shift disables context shift on infinite text generation (default: enabled)
(env: LLAMA_ARG_NO_CONTEXT_SHIFT)
--context-shift enables context shift on infinite text generation (default: disabled)
(env: LLAMA_ARG_CONTEXT_SHIFT)
-r, --reverse-prompt PROMPT halt generation at PROMPT, return control in interactive mode
-sp, --special special tokens output enabled (default: false)
--no-warmup skip warming up the model with an empty run
--spm-infill use Suffix/Prefix/Middle pattern for infill (instead of
Prefix/Suffix/Middle) as some models prefer this. (default: disabled)
--pooling {none,mean,cls,last,rank} pooling type for embeddings, use model default if unspecified
(env: LLAMA_ARG_POOLING)
-cb, --cont-batching enable continuous batching (a.k.a dynamic batching) (default: enabled)
(env: LLAMA_ARG_CONT_BATCHING)
-nocb, --no-cont-batching disable continuous batching
(env: LLAMA_ARG_NO_CONT_BATCHING)
--mmproj FILE path to a multimodal projector file. see tools/mtmd/README.md
note: if -hf is used, this argument can be omitted
(env: LLAMA_ARG_MMPROJ)
--mmproj-url URL URL to a multimodal projector file. see tools/mtmd/README.md
(env: LLAMA_ARG_MMPROJ_URL)
--no-mmproj explicitly disable multimodal projector, useful when using -hf
(env: LLAMA_ARG_NO_MMPROJ)
--no-mmproj-offload do not offload multimodal projector to GPU
(env: LLAMA_ARG_NO_MMPROJ_OFFLOAD)
--image-min-tokens N minimum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MIN_TOKENS)
--image-max-tokens N maximum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MAX_TOKENS)
--override-tensor-draft, -otd <tensor name pattern>=<buffer type>,...
override tensor buffer type for draft model
--cpu-moe-draft, -cmoed keep all Mixture of Experts (MoE) weights in the CPU for the draft
model
(env: LLAMA_ARG_CPU_MOE_DRAFT)
--n-cpu-moe-draft, -ncmoed N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU for the draft model
(env: LLAMA_ARG_N_CPU_MOE_DRAFT)
-a, --alias STRING set alias for model name (to be used by REST API)
(env: LLAMA_ARG_ALIAS)
--host HOST ip address to listen, or bind to an UNIX socket if the address ends
with .sock (default: 127.0.0.1)
(env: LLAMA_ARG_HOST)
--port PORT port to listen (default: 8080)
(env: LLAMA_ARG_PORT)
--path PATH path to serve static files from (default: )
(env: LLAMA_ARG_STATIC_PATH)
--api-prefix PREFIX prefix path the server serves from, without the trailing slash
(default: )
(env: LLAMA_ARG_API_PREFIX)
--no-webui Disable the Web UI (default: enabled)
(env: LLAMA_ARG_NO_WEBUI)
--embedding, --embeddings restrict to only support embedding use case; use only with dedicated
embedding models (default: disabled)
(env: LLAMA_ARG_EMBEDDINGS)
--reranking, --rerank enable reranking endpoint on server (default: disabled)
(env: LLAMA_ARG_RERANKING)
--api-key KEY API key to use for authentication (default: none)
(env: LLAMA_API_KEY)
--api-key-file FNAME path to file containing API keys (default: none)
--ssl-key-file FNAME path to file a PEM-encoded SSL private key
(env: LLAMA_ARG_SSL_KEY_FILE)
--ssl-cert-file FNAME path to file a PEM-encoded SSL certificate
(env: LLAMA_ARG_SSL_CERT_FILE)
--chat-template-kwargs STRING sets additional params for the json template parser
(env: LLAMA_CHAT_TEMPLATE_KWARGS)
-to, --timeout N server read/write timeout in seconds (default: 600)
(env: LLAMA_ARG_TIMEOUT)
--threads-http N number of threads used to process HTTP requests (default: -1)
(env: LLAMA_ARG_THREADS_HTTP)
--cache-reuse N min chunk size to attempt reusing from the cache via KV shifting
(default: 0)
[(card)](https://ggml.ai/f0.png)
(env: LLAMA_ARG_CACHE_REUSE)
--metrics enable prometheus compatible metrics endpoint (default: disabled)
(env: LLAMA_ARG_ENDPOINT_METRICS)
--props enable changing global properties via POST /props (default: disabled)
(env: LLAMA_ARG_ENDPOINT_PROPS)
--slots enable slots monitoring endpoint (default: enabled)
(env: LLAMA_ARG_ENDPOINT_SLOTS)
--no-slots disables slots monitoring endpoint
(env: LLAMA_ARG_NO_ENDPOINT_SLOTS)
--slot-save-path PATH path to save slot kv cache (default: disabled)
--jinja use jinja template for chat (default: disabled)
(env: LLAMA_ARG_JINJA)
--reasoning-format FORMAT controls whether thought tags are allowed and/or extracted from the
response, and in which format they're returned; one of:
- none: leaves thoughts unparsed in `message.content`
- deepseek: puts thoughts in `message.reasoning_content`
- deepseek-legacy: keeps `<think>` tags in `message.content` while
also populating `message.reasoning_content`
(default: auto)
(env: LLAMA_ARG_THINK)
--reasoning-budget N controls the amount of thinking allowed; currently only one of: -1 for
unrestricted thinking budget, or 0 to disable thinking (default: -1)
(env: LLAMA_ARG_THINK_BUDGET)
--chat-template JINJA_TEMPLATE set custom jinja chat template (default: template taken from model's
metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE)
--chat-template-file JINJA_TEMPLATE_FILE
set custom jinja chat template file (default: template taken from
model's metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE_FILE)
--no-prefill-assistant whether to prefill the assistant's response if the last message is an
assistant message (default: prefill enabled)
when this flag is set, if the last message is an assistant message
then it will be treated as a full message and not prefilled

(env: LLAMA_ARG_NO_PREFILL_ASSISTANT)
-sps, --slot-prompt-similarity SIMILARITY
how much the prompt of a request must match the prompt of a slot in
order to use that slot (default: 0.10, 0.0 = disabled)
--lora-init-without-apply load LoRA adapters without applying them (apply later via POST
/lora-adapters) (default: disabled)
-td, --threads-draft N number of threads to use during generation (default: same as
--threads)
-tbd, --threads-batch-draft N number of threads to use during batch and prompt processing (default:
same as --threads-draft)
--draft-max, --draft, --draft-n N number of tokens to draft for speculative decoding (default: 16)
(env: LLAMA_ARG_DRAFT_MAX)
--draft-min, --draft-n-min N minimum number of draft tokens to use for speculative decoding
(default: 0)
(env: LLAMA_ARG_DRAFT_MIN)
--draft-p-min P minimum speculative decoding probability (greedy) (default: 0.8)
(env: LLAMA_ARG_DRAFT_P_MIN)
-cd, --ctx-size-draft N size of the prompt context for the draft model (default: 0, 0 = loaded
from model)
(env: LLAMA_ARG_CTX_SIZE_DRAFT)
-devd, --device-draft <dev1,dev2,..> comma-separated list of devices to use for offloading the draft model
(none = don't offload)
use --list-devices to see a list of available devices
-ngld, --gpu-layers-draft, --n-gpu-layers-draft N
number of layers to store in VRAM for the draft model
(env: LLAMA_ARG_N_GPU_LAYERS_DRAFT)
-md, --model-draft FNAME draft model for speculative decoding (default: unused)
(env: LLAMA_ARG_MODEL_DRAFT)
--spec-replace TARGET DRAFT translate the string in TARGET into DRAFT if the draft model and main
model are not compatible
-mv, --model-vocoder FNAME vocoder model for audio generation (default: unused)
--tts-use-guide-tokens Use guide tokens to improve TTS word recall
--embd-gemma-default use default EmbeddingGemma model (note: can download weights from the
internet)
--fim-qwen-1.5b-default use default Qwen 2.5 Coder 1.5B (note: can download weights from the
internet)
--fim-qwen-3b-default use default Qwen 2.5 Coder 3B (note: can download weights from the
internet)
--fim-qwen-7b-default use default Qwen 2.5 Coder 7B (note: can download weights from the
internet)
--fim-qwen-7b-spec use Qwen 2.5 Coder 7B + 0.5B draft for speculative decoding (note: can
download weights from the internet)
--fim-qwen-14b-spec use Qwen 2.5 Coder 14B + 0.5B draft for speculative decoding (note:
can download weights from the internet)
--fim-qwen-30b-default use default Qwen 3 Coder 30B A3B Instruct (note: can download weights
from the internet)
--gpt-oss-20b-default use gpt-oss-20b (note: can download weights from the internet)
--gpt-oss-120b-default use gpt-oss-120b (note: can download weights from the internet)
--vision-gemma-4b-default use Gemma 3 4B QAT (note: can download weights from the internet)
--vision-gemma-12b-default use Gemma 3 12B QAT (note: can download weights from the internet)
</syntaxhighlight>
=== Test ===
=== Bekannte Probleme ===
<syntaxhighlight lang="bash" line="1">
Error 1 [ 34%] Linking HIP shared library ../../../bin/libggml-hip.so [ 34%] Built target ggml-hip gmake[1]: *** [CMakeFiles/Makefile2:1775: ggml/src/CMakeFiles/ggml-cpu.dir/all] Error 2 gmake: *** [Makefile:146: all] Error 2
</syntaxhighlight>
Lösung:
<syntaxhighlight lang="bash" line="1">
GCC 13.3.0 konnte damit nicht umgehen, aber GCC 14.2.0 hat bessere C++-Standard-Compliance und kann diesen Header-Konflikt korrekt auflösen – deshalb funktioniert die Kompilierung jetzt mit der neueren Compiler-Version.
sudo update-alternatives --remove gcc /usr/bin/gcc-13
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-14 100
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-13 50
sudo update-alternatives --config gcc
# Wähle gcc-14

sudo update-alternatives --remove g++ /usr/bin/g++-13
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-14 100
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-13 50
sudo update-alternatives --config g++
# Wähle g++-14
</syntaxhighlight>
=== Nützliche Links ===
* [https://github.com/ggml-org/llama.cpp Offizielles llama.cpp Repository]
* [https://github.com/ggml-org/llama.cpp/blob/master/examples/server/README.md llama-server Dokumentation]
* [https://github.com/ggml-org/llama.cpp/discussions llama.cpp Discussions]
* [https://huggingface.co/models?library=gguf GGUF Modelle auf Hugging Face]
* [https://docs.rocm.com/ ROCm Dokumentation]
* [https://vulkan.lunarg.com/doc/sdk Vulkan SDK Dokumentation]

Llama.cpp

2026-05-31T10:50:21Z

Hendrik: /* ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 */

=== Beschreibung ===
llama.cpp ist eine C/C++ Implementierung für Inference von Large Language Models (LLMs). Es unterstützt verschiedene Backends (CPU, Vulkan, ROCm, CUDA) und ermöglicht das Ausführen von quantisierten Modellen im GGUF-Format.

=== Download ===
<syntaxhighlight lang="bash" line="1">
git clone https://github.com/ggml-org/llama.cpp
cd llama.cpp
</syntaxhighlight>

=== Installation ===
==== Vulkan Build ====
<syntaxhighlight lang="bash" line="1">
sudo dnf install libcurl-devel -y
</syntaxhighlight>
[https://vulkan.lunarg.com/sdk/home Vulkan SDK] herunterladen
entpacken, ins Verzeichnis wechseln und source setup-env.sh ausführen. Dann ins Verzeichnis wechseln wo llama.cpp installiert werden soll.
Dort dann
<syntaxhighlight lang="bash" line="1">
source /home/hendrik/Programme/Vulkan_SDK/1.4.335.0/setup-env.sh
cd /home/hendrik/Programme/llama.cpp/
rm -R build-vulkan
cmake -B build-vulkan -DGGML_VULKAN=1
cmake --build build-vulkan --config Release -- -j $(nproc)
</syntaxhighlight>

==== ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 ====
gfx906 = Mi 50 Support 
gfx1100 = 7900 XTX Support

<syntaxhighlight lang="bash" line="1">
Copy & paste all the commands from the quick install https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/quick-start.html
Before rebooting to complete the install, download the 6.4 rocblas from the AUR: https://archlinux.org/packages/extra/x86_64/rocblas/
Extract it
Copy all tensor files that contain gfx906 in rocblas-6.4.3-3-x86_64.pkg/opt/rocm/lib/rocblas/library to /opt/rocm/lib/rocblas/library
Reboot
Check if it worked by running sudo update-alternatives --display rocm
</syntaxhighlight>
# ROCm Umgebung einrichten (optional falls Fehler auftreten)
<syntaxhighlight lang="bash" line="1">
echo 'export LD_LIBRARY_PATH=/opt/rocm/lib:$LD_LIBRARY_PATH' >> ~/.bashrc
echo 'export HSA_OVERRIDE_GFX_VERSION=9.0.6' >> ~/.bashrc # Für MI50/MI60
source ~/.bashrc
</syntaxhighlight>
<syntaxhighlight lang="bash" line="1">
sudo dnf install rocwmma-devel -y
</syntaxhighlight>
# llama.cpp kompilieren
<syntaxhighlight lang="bash" line="1">
cd /home/hendrik/Programme/llama.cpp/ && git pull && rm -R build-rocm-old && cp -R build-rocm build-rocm-old
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" cmake -B build-rocm -DGGML_HIP=ON -DAMDGPU_TARGETS=gfx1100 -DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 16
</syntaxhighlight>

Für mehrere GPU-Architekturen gleichzeitig:
<syntaxhighlight lang="bash" line="1">
# Für MI50 (gfx906) und RX 7900 XTX (gfx1100)
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
cmake -S . -B build -DGGML_HIP=ON -DAMDGPU_TARGETS="gfx906;gfx1100" \
-DCMAKE_BUILD_TYPE=Release && \
cmake --build build --config Release -- -j 8
</syntaxhighlight>

=== ik_llama.cpp ===
==== Vulkan Build ====
<syntaxhighlight lang="bash" line="1">
sudo dnf install libcurl-devel -y
</syntaxhighlight>
[https://vulkan.lunarg.com/sdk/home Vulkan SDK] herunterladen
entpacken, ins Verzeichnis wechseln und source setup-env.sh ausführen. Dann ins Verzeichnis wechseln wo llama.cpp installiert werden soll.
Dort dann
<syntaxhighlight lang="bash" line="1">
source /home/hendrik/Programme/Vulkan_SDK/1.4.335.0/setup-env.sh
cd /home/hendrik/Programme/llama.cpp/
rm -R build-vulkan
cmake -B build-vulkan -DLLAMA_VULKAN=on
cmake --build build-vulkan --config Release -- -j $(nproc)
</syntaxhighlight>

==== ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 ====
gfx906 = Mi 50 Support 
gfx1100 = 7900 XTX Support
gfx1151 = Strix Halo
# llama.cpp kompilieren
<syntaxhighlight lang="bash" line="1">
cd /home/hendrik/Programme/llama.cpp/ && rm -R build-rocm-old && cp -R build-rocm build-rocm-old
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" cmake -B build-rocm -DLLAMA_HIPBLAS=on -DAMDGPU_TARGETS=gfx1100,gfx1151 -DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 16
</syntaxhighlight>

=== Konfiguration ===

==== llama-server als systemd Service einrichten ====

'''Service-Datei erstellen:'''
<syntaxhighlight lang="bash" line="1">
sudo nano /etc/systemd/system/llama-server.service
</syntaxhighlight>

'''Service-Datei Inhalt (Multi-GPU mit ROCm):'''
<syntaxhighlight lang="ini" line="1">
[Unit]
Description=Llama.cpp ROCm Multi-GPU Server
After=network.target

[Service]
Type=simple
User=username
Group=username
WorkingDirectory=/home/username/llama.cpp/build/bin

# Multi-GPU Konfiguration
Environment="HIP_VISIBLE_DEVICES=0,1"
Environment="HSA_OVERRIDE_GFX_VERSION=9.0.6"
Environment="PATH=/opt/rocm/bin:/usr/local/bin:/usr/bin:/bin"
Environment="LD_LIBRARY_PATH=/opt/rocm/lib"

# Server mit optimalen Multi-GPU Einstellungen
ExecStart=/home/username/llama.cpp/build/bin/llama-server \
-m /home/username/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja

Restart=always
RestartSec=10
LimitNOFILE=65535
LimitMEMLOCK=infinity
StandardOutput=journal
StandardError=journal
SyslogIdentifier=llama-server

[Install]
WantedBy=multi-user.target
</syntaxhighlight>

'''Service aktivieren und starten:'''
<syntaxhighlight lang="bash" line="1">
# Service neu laden
sudo systemctl daemon-reload

# Service aktivieren (Auto-Start beim Boot)
sudo systemctl enable llama-server

# Service starten
sudo systemctl start llama-server

# Status prüfen
sudo systemctl status llama-server

# Logs anzeigen
sudo journalctl -u llama-server -f
</syntaxhighlight>

'''Service Management:'''
<syntaxhighlight lang="bash" line="1">
# Service stoppen
sudo systemctl stop llama-server

# Service neustarten
sudo systemctl restart llama-server

# Service deaktivieren
sudo systemctl disable llama-server

==== Manuelle Server-Starts ====
'''Multi-GPU (ROCm) - Optimal:'''
<syntaxhighlight lang="bash" line="1">
HIP_VISIBLE_DEVICES=0,1 ./llama-server \
-m ~/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja
</syntaxhighlight>

=== Update ===
<syntaxhighlight lang="bash" line="1">
cd ~/llama.cpp
git pull
cmake --build build --config Release -- -j $(nproc)

# Service neu starten falls aktiv
sudo systemctl restart llama-server
</syntaxhighlight>
=== Befehlszeilenargumente ===
llama-server
<syntaxhighlight lang="bash" line="1">
-h, --help, --usage print usage and exit
--version show version and build info
-cl, --cache-list show list of models in cache
--completion-bash print source-able bash completion script for llama.cpp
--verbose-prompt print a verbose prompt before generation (default: false)
-t, --threads N number of CPU threads to use during generation (default: -1)
(env: LLAMA_ARG_THREADS)
-tb, --threads-batch N number of threads to use during batch and prompt processing (default:
same as --threads)
-C, --cpu-mask M CPU affinity mask: arbitrarily long hex. Complements cpu-range
(default: "")
-Cr, --cpu-range lo-hi range of CPUs for affinity. Complements --cpu-mask
--cpu-strict <0|1> use strict CPU placement (default: 0)
--prio N set process/thread priority : low(-1), normal(0), medium(1), high(2),
realtime(3) (default: 0)
--poll <0...100> use polling level to wait for work (0 - no polling, default: 50)
-Cb, --cpu-mask-batch M CPU affinity mask: arbitrarily long hex. Complements cpu-range-batch
(default: same as --cpu-mask)
-Crb, --cpu-range-batch lo-hi ranges of CPUs for affinity. Complements --cpu-mask-batch
--cpu-strict-batch <0|1> use strict CPU placement (default: same as --cpu-strict)
--prio-batch N set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime
(default: 0)
--poll-batch <0|1> use polling to wait for work (default: same as --poll)
-c, --ctx-size N size of the prompt context (default: 4096, 0 = loaded from model)
(env: LLAMA_ARG_CTX_SIZE)
-n, --predict, --n-predict N number of tokens to predict (default: -1, -1 = infinity)
(env: LLAMA_ARG_N_PREDICT)
-b, --batch-size N logical maximum batch size (default: 2048)
(env: LLAMA_ARG_BATCH)
-ub, --ubatch-size N physical maximum batch size (default: 512)
(env: LLAMA_ARG_UBATCH)
--keep N number of tokens to keep from the initial prompt (default: 0, -1 =
all)
--swa-full use full-size SWA cache (default: false)
[(more
info)](https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
(env: LLAMA_ARG_SWA_FULL)
--kv-unified, -kvu use single unified KV buffer for the KV cache of all sequences
(default: false)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/14363)
(env: LLAMA_ARG_KV_SPLIT)
-fa, --flash-attn [on|off|auto] set Flash Attention use ('on', 'off', or 'auto', default: 'auto')
(env: LLAMA_ARG_FLASH_ATTN)
--no-perf disable internal libllama performance timings (default: false)
(env: LLAMA_ARG_NO_PERF)
-e, --escape process escapes sequences (\n, \r, \t, \', \", \\) (default: true)
--no-escape do not process escape sequences
--rope-scaling {none,linear,yarn} RoPE frequency scaling method, defaults to linear unless specified by
the model
(env: LLAMA_ARG_ROPE_SCALING_TYPE)
--rope-scale N RoPE context scaling factor, expands context by a factor of N
(env: LLAMA_ARG_ROPE_SCALE)
--rope-freq-base N RoPE base frequency, used by NTK-aware scaling (default: loaded from
model)
(env: LLAMA_ARG_ROPE_FREQ_BASE)
--rope-freq-scale N RoPE frequency scaling factor, expands context by a factor of 1/N
(env: LLAMA_ARG_ROPE_FREQ_SCALE)
--yarn-orig-ctx N YaRN: original context size of model (default: 0 = model training
context size)
(env: LLAMA_ARG_YARN_ORIG_CTX)
--yarn-ext-factor N YaRN: extrapolation mix factor (default: -1.0, 0.0 = full
interpolation)
(env: LLAMA_ARG_YARN_EXT_FACTOR)
--yarn-attn-factor N YaRN: scale sqrt(t) or attention magnitude (default: -1.0)
(env: LLAMA_ARG_YARN_ATTN_FACTOR)
--yarn-beta-slow N YaRN: high correction dim or alpha (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_SLOW)
--yarn-beta-fast N YaRN: low correction dim or beta (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_FAST)
-nkvo, --no-kv-offload disable KV offload
(env: LLAMA_ARG_NO_KV_OFFLOAD)
-nr, --no-repack disable weight repacking
(env: LLAMA_ARG_NO_REPACK)
--no-host bypass host buffer allowing extra buffers to be used
(env: LLAMA_ARG_NO_HOST)
-ctk, --cache-type-k TYPE KV cache data type for K
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K)
-ctv, --cache-type-v TYPE KV cache data type for V
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V)
-dt, --defrag-thold N KV cache defragmentation threshold (DEPRECATED)
(env: LLAMA_ARG_DEFRAG_THOLD)
-np, --parallel N number of parallel sequences to decode (default: 1)
(env: LLAMA_ARG_N_PARALLEL)
--mlock force system to keep model in RAM rather than swapping or compressing
(env: LLAMA_ARG_MLOCK)
--no-mmap do not memory-map model (slower load but may reduce pageouts if not
using mlock)
(env: LLAMA_ARG_NO_MMAP)
--numa TYPE attempt optimizations that help on some NUMA systems
- distribute: spread execution evenly over all nodes
- isolate: only spawn threads on CPUs on the node that execution
started on
- numactl: use the CPU map provided by numactl
if run without this previously, it is recommended to drop the system
page cache before using this
see https://github.com/ggml-org/llama.cpp/issues/1437
(env: LLAMA_ARG_NUMA)
-dev, --device <dev1,dev2,..> comma-separated list of devices to use for offloading (none = don't
offload)
use --list-devices to see a list of available devices
(env: LLAMA_ARG_DEVICE)
--list-devices print list of available devices and exit
--override-tensor, -ot <tensor name pattern>=<buffer type>,...
override tensor buffer type
--cpu-moe, -cmoe keep all Mixture of Experts (MoE) weights in the CPU
(env: LLAMA_ARG_CPU_MOE)
--n-cpu-moe, -ncmoe N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU
(env: LLAMA_ARG_N_CPU_MOE)
-ngl, --gpu-layers, --n-gpu-layers N max. number of layers to store in VRAM (default: -1)
(env: LLAMA_ARG_N_GPU_LAYERS)
-sm, --split-mode {none,layer,row} how to split the model across multiple GPUs, one of:
- none: use one GPU only
- layer (default): split layers and KV across GPUs
- row: split rows across GPUs
(env: LLAMA_ARG_SPLIT_MODE)
-ts, --tensor-split N0,N1,N2,... fraction of the model to offload to each GPU, comma-separated list of
proportions, e.g. 3,1
(env: LLAMA_ARG_TENSOR_SPLIT)
-mg, --main-gpu INDEX the GPU to use for the model (with split-mode = none), or for
intermediate results and KV (with split-mode = row) (default: 0)
(env: LLAMA_ARG_MAIN_GPU)
--check-tensors check model tensor data for invalid values (default: false)
--override-kv KEY=TYPE:VALUE advanced option to override model metadata by key. may be specified
multiple times.
types: int, float, bool, str. example: --override-kv
tokenizer.ggml.add_bos_token=bool:false
--no-op-offload disable offloading host tensor operations to device (default: false)
--lora FNAME path to LoRA adapter (can be repeated to use multiple adapters)
--lora-scaled FNAME SCALE path to LoRA adapter with user defined scaling (can be repeated to use
multiple adapters)
--control-vector FNAME add a control vector
note: this argument can be repeated to add multiple control vectors
--control-vector-scaled FNAME SCALE add a control vector with user defined scaling SCALE
note: this argument can be repeated to add multiple scaled control
vectors
--control-vector-layer-range START END
layer range to apply the control vector(s) to, start and end inclusive
-m, --model FNAME model path (default: `models/$filename` with filename from `--hf-file`
or `--model-url` if set, otherwise models/7B/ggml-model-f16.gguf)
(env: LLAMA_ARG_MODEL)
-mu, --model-url MODEL_URL model download url (default: unused)
(env: LLAMA_ARG_MODEL_URL)
-dr, --docker-repo [<repo>/]<model>[:quant]
Docker Hub model repository. repo is optional, default to ai/. quant
is optional, default to :latest.
example: gemma3
(default: unused)
(env: LLAMA_ARG_DOCKER_REPO)
-hf, -hfr, --hf-repo <user>/<model>[:quant]
Hugging Face model repository; quant is optional, case-insensitive,
default to Q4_K_M, or falls back to the first file in the repo if
Q4_K_M doesn't exist.
mmproj is also downloaded automatically if available. to disable, add
--no-mmproj
example: unsloth/phi-4-GGUF:q4_k_m
(default: unused)
(env: LLAMA_ARG_HF_REPO)
-hfd, -hfrd, --hf-repo-draft <user>/<model>[:quant]
Same as --hf-repo, but for the draft model (default: unused)
(env: LLAMA_ARG_HFD_REPO)
-hff, --hf-file FILE Hugging Face model file. If specified, it will override the quant in
--hf-repo (default: unused)
(env: LLAMA_ARG_HF_FILE)
-hfv, -hfrv, --hf-repo-v <user>/<model>[:quant]
Hugging Face model repository for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_REPO_V)
-hffv, --hf-file-v FILE Hugging Face model file for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_FILE_V)
-hft, --hf-token TOKEN Hugging Face access token (default: value from HF_TOKEN environment
variable)
(env: HF_TOKEN)
--log-disable Log disable
--log-file FNAME Log to file
--log-colors [on|off|auto] Set colored logging ('on', 'off', or 'auto', default: 'auto')
'auto' enables colors when output is to a terminal
(env: LLAMA_LOG_COLORS)
-v, --verbose, --log-verbose Set verbosity level to infinity (i.e. log all messages, useful for
debugging)
--offline Offline mode: forces use of cache, prevents network access
(env: LLAMA_OFFLINE)
-lv, --verbosity, --log-verbosity N Set the verbosity threshold. Messages with a higher verbosity will be
ignored.
(env: LLAMA_LOG_VERBOSITY)
--log-prefix Enable prefix in log messages
(env: LLAMA_LOG_PREFIX)
--log-timestamps Enable timestamps in log messages
(env: LLAMA_LOG_TIMESTAMPS)
-ctkd, --cache-type-k-draft TYPE KV cache data type for K for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K_DRAFT)
-ctvd, --cache-type-v-draft TYPE KV cache data type for V for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V_DRAFT)

----- sampling params -----

--samplers SAMPLERS samplers that will be used for generation in the order, separated by
';'
(default:
penalties;dry;top_n_sigma;top_k;typ_p;top_p;min_p;xtc;temperature)
-s, --seed SEED RNG seed (default: -1, use random seed for -1)
--sampling-seq, --sampler-seq SEQUENCE
simplified sequence for samplers that will be used (default:
edskypmxt)
--ignore-eos ignore end of stream token and continue generating (implies
--logit-bias EOS-inf)
--temp N temperature (default: 0.8)
--top-k N top-k sampling (default: 40, 0 = disabled)
--top-p N top-p sampling (default: 0.9, 1.0 = disabled)
--min-p N min-p sampling (default: 0.1, 0.0 = disabled)
--top-nsigma N top-n-sigma sampling (default: -1.0, -1.0 = disabled)
--xtc-probability N xtc probability (default: 0.0, 0.0 = disabled)
--xtc-threshold N xtc threshold (default: 0.1, 1.0 = disabled)
--typical N locally typical sampling, parameter p (default: 1.0, 1.0 = disabled)
--repeat-last-n N last n tokens to consider for penalize (default: 64, 0 = disabled, -1
= ctx_size)
--repeat-penalty N penalize repeat sequence of tokens (default: 1.0, 1.0 = disabled)
--presence-penalty N repeat alpha presence penalty (default: 0.0, 0.0 = disabled)
--frequency-penalty N repeat alpha frequency penalty (default: 0.0, 0.0 = disabled)
--dry-multiplier N set DRY sampling multiplier (default: 0.0, 0.0 = disabled)
--dry-base N set DRY sampling base value (default: 1.75)
--dry-allowed-length N set allowed length for DRY sampling (default: 2)
--dry-penalty-last-n N set DRY penalty for the last n tokens (default: -1, 0 = disable, -1 =
context size)
--dry-sequence-breaker STRING add sequence breaker for DRY sampling, clearing out default breakers
('\n', ':', '"', '*') in the process; use "none" to not use any
sequence breakers
--dynatemp-range N dynamic temperature range (default: 0.0, 0.0 = disabled)
--dynatemp-exp N dynamic temperature exponent (default: 1.0)
--mirostat N use Mirostat sampling.
Top K, Nucleus and Locally Typical samplers are ignored if used.
(default: 0, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)
--mirostat-lr N Mirostat learning rate, parameter eta (default: 0.1)
--mirostat-ent N Mirostat target entropy, parameter tau (default: 5.0)
-l, --logit-bias TOKEN_ID(+/-)BIAS modifies the likelihood of token appearing in the completion,
i.e. `--logit-bias 15043+1` to increase likelihood of token ' Hello',
or `--logit-bias 15043-1` to decrease likelihood of token ' Hello'
--grammar GRAMMAR BNF-like grammar to constrain generations (see samples in grammars/
dir) (default: '')
--grammar-file FNAME file to read grammar from
-j, --json-schema SCHEMA JSON schema to constrain generations (https://json-schema.org/), e.g.
`{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead
-jf, --json-schema-file FILE File containing a JSON schema to constrain generations
(https://json-schema.org/), e.g. `{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead

----- example-specific params -----

--ctx-checkpoints, --swa-checkpoints N
max number of context checkpoints to create per slot (default: 8)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/15293)
(env: LLAMA_ARG_CTX_CHECKPOINTS)
--cache-ram, -cram N set the maximum cache size in MiB (default: 8192, -1 - no limit, 0 -
disable)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/16391)
(env: LLAMA_ARG_CACHE_RAM)
--no-context-shift disables context shift on infinite text generation (default: enabled)
(env: LLAMA_ARG_NO_CONTEXT_SHIFT)
--context-shift enables context shift on infinite text generation (default: disabled)
(env: LLAMA_ARG_CONTEXT_SHIFT)
-r, --reverse-prompt PROMPT halt generation at PROMPT, return control in interactive mode
-sp, --special special tokens output enabled (default: false)
--no-warmup skip warming up the model with an empty run
--spm-infill use Suffix/Prefix/Middle pattern for infill (instead of
Prefix/Suffix/Middle) as some models prefer this. (default: disabled)
--pooling {none,mean,cls,last,rank} pooling type for embeddings, use model default if unspecified
(env: LLAMA_ARG_POOLING)
-cb, --cont-batching enable continuous batching (a.k.a dynamic batching) (default: enabled)
(env: LLAMA_ARG_CONT_BATCHING)
-nocb, --no-cont-batching disable continuous batching
(env: LLAMA_ARG_NO_CONT_BATCHING)
--mmproj FILE path to a multimodal projector file. see tools/mtmd/README.md
note: if -hf is used, this argument can be omitted
(env: LLAMA_ARG_MMPROJ)
--mmproj-url URL URL to a multimodal projector file. see tools/mtmd/README.md
(env: LLAMA_ARG_MMPROJ_URL)
--no-mmproj explicitly disable multimodal projector, useful when using -hf
(env: LLAMA_ARG_NO_MMPROJ)
--no-mmproj-offload do not offload multimodal projector to GPU
(env: LLAMA_ARG_NO_MMPROJ_OFFLOAD)
--image-min-tokens N minimum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MIN_TOKENS)
--image-max-tokens N maximum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MAX_TOKENS)
--override-tensor-draft, -otd <tensor name pattern>=<buffer type>,...
override tensor buffer type for draft model
--cpu-moe-draft, -cmoed keep all Mixture of Experts (MoE) weights in the CPU for the draft
model
(env: LLAMA_ARG_CPU_MOE_DRAFT)
--n-cpu-moe-draft, -ncmoed N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU for the draft model
(env: LLAMA_ARG_N_CPU_MOE_DRAFT)
-a, --alias STRING set alias for model name (to be used by REST API)
(env: LLAMA_ARG_ALIAS)
--host HOST ip address to listen, or bind to an UNIX socket if the address ends
with .sock (default: 127.0.0.1)
(env: LLAMA_ARG_HOST)
--port PORT port to listen (default: 8080)
(env: LLAMA_ARG_PORT)
--path PATH path to serve static files from (default: )
(env: LLAMA_ARG_STATIC_PATH)
--api-prefix PREFIX prefix path the server serves from, without the trailing slash
(default: )
(env: LLAMA_ARG_API_PREFIX)
--no-webui Disable the Web UI (default: enabled)
(env: LLAMA_ARG_NO_WEBUI)
--embedding, --embeddings restrict to only support embedding use case; use only with dedicated
embedding models (default: disabled)
(env: LLAMA_ARG_EMBEDDINGS)
--reranking, --rerank enable reranking endpoint on server (default: disabled)
(env: LLAMA_ARG_RERANKING)
--api-key KEY API key to use for authentication (default: none)
(env: LLAMA_API_KEY)
--api-key-file FNAME path to file containing API keys (default: none)
--ssl-key-file FNAME path to file a PEM-encoded SSL private key
(env: LLAMA_ARG_SSL_KEY_FILE)
--ssl-cert-file FNAME path to file a PEM-encoded SSL certificate
(env: LLAMA_ARG_SSL_CERT_FILE)
--chat-template-kwargs STRING sets additional params for the json template parser
(env: LLAMA_CHAT_TEMPLATE_KWARGS)
-to, --timeout N server read/write timeout in seconds (default: 600)
(env: LLAMA_ARG_TIMEOUT)
--threads-http N number of threads used to process HTTP requests (default: -1)
(env: LLAMA_ARG_THREADS_HTTP)
--cache-reuse N min chunk size to attempt reusing from the cache via KV shifting
(default: 0)
[(card)](https://ggml.ai/f0.png)
(env: LLAMA_ARG_CACHE_REUSE)
--metrics enable prometheus compatible metrics endpoint (default: disabled)
(env: LLAMA_ARG_ENDPOINT_METRICS)
--props enable changing global properties via POST /props (default: disabled)
(env: LLAMA_ARG_ENDPOINT_PROPS)
--slots enable slots monitoring endpoint (default: enabled)
(env: LLAMA_ARG_ENDPOINT_SLOTS)
--no-slots disables slots monitoring endpoint
(env: LLAMA_ARG_NO_ENDPOINT_SLOTS)
--slot-save-path PATH path to save slot kv cache (default: disabled)
--jinja use jinja template for chat (default: disabled)
(env: LLAMA_ARG_JINJA)
--reasoning-format FORMAT controls whether thought tags are allowed and/or extracted from the
response, and in which format they're returned; one of:
- none: leaves thoughts unparsed in `message.content`
- deepseek: puts thoughts in `message.reasoning_content`
- deepseek-legacy: keeps `<think>` tags in `message.content` while
also populating `message.reasoning_content`
(default: auto)
(env: LLAMA_ARG_THINK)
--reasoning-budget N controls the amount of thinking allowed; currently only one of: -1 for
unrestricted thinking budget, or 0 to disable thinking (default: -1)
(env: LLAMA_ARG_THINK_BUDGET)
--chat-template JINJA_TEMPLATE set custom jinja chat template (default: template taken from model's
metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE)
--chat-template-file JINJA_TEMPLATE_FILE
set custom jinja chat template file (default: template taken from
model's metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE_FILE)
--no-prefill-assistant whether to prefill the assistant's response if the last message is an
assistant message (default: prefill enabled)
when this flag is set, if the last message is an assistant message
then it will be treated as a full message and not prefilled

(env: LLAMA_ARG_NO_PREFILL_ASSISTANT)
-sps, --slot-prompt-similarity SIMILARITY
how much the prompt of a request must match the prompt of a slot in
order to use that slot (default: 0.10, 0.0 = disabled)
--lora-init-without-apply load LoRA adapters without applying them (apply later via POST
/lora-adapters) (default: disabled)
-td, --threads-draft N number of threads to use during generation (default: same as
--threads)
-tbd, --threads-batch-draft N number of threads to use during batch and prompt processing (default:
same as --threads-draft)
--draft-max, --draft, --draft-n N number of tokens to draft for speculative decoding (default: 16)
(env: LLAMA_ARG_DRAFT_MAX)
--draft-min, --draft-n-min N minimum number of draft tokens to use for speculative decoding
(default: 0)
(env: LLAMA_ARG_DRAFT_MIN)
--draft-p-min P minimum speculative decoding probability (greedy) (default: 0.8)
(env: LLAMA_ARG_DRAFT_P_MIN)
-cd, --ctx-size-draft N size of the prompt context for the draft model (default: 0, 0 = loaded
from model)
(env: LLAMA_ARG_CTX_SIZE_DRAFT)
-devd, --device-draft <dev1,dev2,..> comma-separated list of devices to use for offloading the draft model
(none = don't offload)
use --list-devices to see a list of available devices
-ngld, --gpu-layers-draft, --n-gpu-layers-draft N
number of layers to store in VRAM for the draft model
(env: LLAMA_ARG_N_GPU_LAYERS_DRAFT)
-md, --model-draft FNAME draft model for speculative decoding (default: unused)
(env: LLAMA_ARG_MODEL_DRAFT)
--spec-replace TARGET DRAFT translate the string in TARGET into DRAFT if the draft model and main
model are not compatible
-mv, --model-vocoder FNAME vocoder model for audio generation (default: unused)
--tts-use-guide-tokens Use guide tokens to improve TTS word recall
--embd-gemma-default use default EmbeddingGemma model (note: can download weights from the
internet)
--fim-qwen-1.5b-default use default Qwen 2.5 Coder 1.5B (note: can download weights from the
internet)
--fim-qwen-3b-default use default Qwen 2.5 Coder 3B (note: can download weights from the
internet)
--fim-qwen-7b-default use default Qwen 2.5 Coder 7B (note: can download weights from the
internet)
--fim-qwen-7b-spec use Qwen 2.5 Coder 7B + 0.5B draft for speculative decoding (note: can
download weights from the internet)
--fim-qwen-14b-spec use Qwen 2.5 Coder 14B + 0.5B draft for speculative decoding (note:
can download weights from the internet)
--fim-qwen-30b-default use default Qwen 3 Coder 30B A3B Instruct (note: can download weights
from the internet)
--gpt-oss-20b-default use gpt-oss-20b (note: can download weights from the internet)
--gpt-oss-120b-default use gpt-oss-120b (note: can download weights from the internet)
--vision-gemma-4b-default use Gemma 3 4B QAT (note: can download weights from the internet)
--vision-gemma-12b-default use Gemma 3 12B QAT (note: can download weights from the internet)
</syntaxhighlight>
=== Test ===
=== Bekannte Probleme ===
<syntaxhighlight lang="bash" line="1">
Error 1 [ 34%] Linking HIP shared library ../../../bin/libggml-hip.so [ 34%] Built target ggml-hip gmake[1]: *** [CMakeFiles/Makefile2:1775: ggml/src/CMakeFiles/ggml-cpu.dir/all] Error 2 gmake: *** [Makefile:146: all] Error 2
</syntaxhighlight>
Lösung:
<syntaxhighlight lang="bash" line="1">
GCC 13.3.0 konnte damit nicht umgehen, aber GCC 14.2.0 hat bessere C++-Standard-Compliance und kann diesen Header-Konflikt korrekt auflösen – deshalb funktioniert die Kompilierung jetzt mit der neueren Compiler-Version.
sudo update-alternatives --remove gcc /usr/bin/gcc-13
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-14 100
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-13 50
sudo update-alternatives --config gcc
# Wähle gcc-14

sudo update-alternatives --remove g++ /usr/bin/g++-13
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-14 100
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-13 50
sudo update-alternatives --config g++
# Wähle g++-14
</syntaxhighlight>
=== Nützliche Links ===
* [https://github.com/ggml-org/llama.cpp Offizielles llama.cpp Repository]
* [https://github.com/ggml-org/llama.cpp/blob/master/examples/server/README.md llama-server Dokumentation]
* [https://github.com/ggml-org/llama.cpp/discussions llama.cpp Discussions]
* [https://huggingface.co/models?library=gguf GGUF Modelle auf Hugging Face]
* [https://docs.rocm.com/ ROCm Dokumentation]
* [https://vulkan.lunarg.com/doc/sdk Vulkan SDK Dokumentation]

SAP HANA

2026-05-05T10:48:50Z

Hendrik:

=== Beschreibung ===
SAP HANA ist eine In-Memory-Datenbank-Plattform, die von SAP entwickelt wurde. Sie ermöglicht es Unternehmen große Datenmengen in Echtzeit zu verarbeiten und zu analysieren. Die Plattform unterstützt sowohl transaktionale als auch analytische Anwendungen und ermöglicht es den Benutzern auf umfassende Echtzeit-Analysen zuzugreifen.

Die In-Memory-Technologie von SAP HANA sorgt dafür, dass Daten in einem schnellen Zugriffsspeicher gespeichert werden, anstatt auf langsameren Festplatten. Dies führt zu schnelleren Datenzugriffszeiten und ermöglicht es Unternehmen Echtzeit-Transaktionen und Analysen durchzuführen.

SAP HANA bietet auch eine Vielzahl von Tools und Anwendungen um Unternehmen bei der Verwaltung von Daten zu unterstützen. Dazu gehören Tools zur Datenintegration, Datenbereinigung, Datenmodellierung und Analyse. Es unterstützt auch eine Vielzahl von Programmiersprachen und Entwicklungsplattformen, einschließlich SAPUI5, Java, Python uvm.
=== Download ===
[https://me.sap.com/softwarecenter/support/index SAP Downloadportal]

<nowiki>https://launchpad.support.sap.com/#/softwarecenter</nowiki> SUPPORT PACKAGES & PATCHES By Category SAP IN-MEMORY (SAP HANA ) HANA PLATFORM EDITION SAP HANA PLATFORM EDITION SAP HANA PLATFORM EDITION 2.0 SAP HANA CLIENT 2.0

=== Installation ===
hdblcm nach installieren:
https://me.sap.com/notes/2651885<syntaxhighlight lang="bash">
<install_media>/SAP_HANA_DATABASE/hdblcm --sid=<SID> --action=update --components=hdblcm --ignore=check_signature_file --verify_signature=off
</syntaxhighlight>

=== Konfiguration===

===SAP HANA Update===
Aktuelle Version als <SID>ADM mit "HDB version" anzeigen lassen

IMDB_CLIENT<VERSION> und IMDB_SERVER<VERSION> von SAP herunterladen und nach /usr/sap/SUM kopieren
<syntaxhighlight lang="bash">
sudo su - root
cd /usr/sap/SUM
chmod +x SAPCAR && ./SAPCAR -xvf IMDB_CLIENT*
chmod +x SAPCAR && ./SAPCAR -xvf IMDB_SERVER*
cd /hana/shared/*/hdblcm && ./hdblcm --action=update --ignore=check_signature_file --verify_signature=off --lss_trust_unsigned_components
Enter directory root to search for components: /usr/sap/SUM/
</syntaxhighlight>
Mit "1" alles updaten und weiter den Anweisungen folgen.

===SAP HANA Sizing Report===
Quelle: [https://blogs.sap.com/2021/05/06/how-to-install-run-the-abap-on-hana-sizing-report-sap-note-1872170-a-step-by-step-guide/ HANA Sizing Report]
# Note updaten. Prüfen, ob 2810633 einbaubar ist, ansonsten 2504480 einbauen. Wenn beide nicht einbaubar sind ist das System zu alt oder bereits aktuell.
# SA38 aufrufen und /SDF/HDB_SIZING ausführen
# "Number of parallel dialog processes" auf 3 setzen
# Ausführen oder als Hintergrundjob einplanen

=== Kennwort ändern ===
<syntaxhighlight lang="bash">
ALTER USER <USER> PASSWORD "<new_password>";
</syntaxhighlight>
Oder direkt mit dem <SID>adm in der Shell:
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER <USER> PASSWORD <PW>"
</syntaxhighlight>
Hinweise:
* für System@TENANT und SYSTEM@SYSTEMDB muss der HDBUSERSTORE für das Backupscript aktualisiert werden und für die Anmeldung im HDB Studio

Lösungsschritte:

Prüfen, welche Keys angelegt sind:

<syntaxhighlight lang="bash">
hdbuserstore list
</syntaxhighlight>
Alten Key löschen (Beispiel: BACKUPKEY für SystemDB):

<syntaxhighlight lang="bash">
hdbuserstore delete BACKUP
</syntaxhighlight>
Neuen Key anlegen und dabei das aktualisierte SYSTEM‑Passwort verwenden:

<syntaxhighlight lang="bash">
hdbuserstore set BACKUP <FQDN>:30013 SYSTEM "<PW>"
</syntaxhighlight>

====SAPABAP1 Kennwort ändern====

Direkt auf Applikationsserver:
<syntaxhighlight lang="bash">
hdbsql -n <FQDN>:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPABAP1 PASSWORD <PW>"
hdbuserstore set default <FQDN>:30013@<TENANT> SAPABAP1 "<PW>"
</syntaxhighlight>
Oder im HDB Studio:
<syntaxhighlight lang="bash">
ALTER USER SAPABAP1 PASSWORD "<PW>";
</syntaxhighlight>
[https://me.sap.com/notes/0002750829 2750829 - Ändern des SAPABAP1-Schemakennworts in SAP HANA]
Anschließend siehe "SAP HANA HDB Userstore neu setzen"

==== SAPDBCTRL (SAP Host Agent) ====
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d SYSTEMDB -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
</syntaxhighlight>
Dann als root fortfahren.
Für SystemDB, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname SYSTEMDB@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>
Für Tenant, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname IMP@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>

=== Encryption root keys backup password setzen ===
<syntaxhighlight lang="bash">
ALTER SYSTEM SET ENCRYPTION ROOT KEYS BACKUP PASSWORD "<PW>";
</syntaxhighlight>
===SAP HANA HDB Userstore neu setzen===
<syntaxhighlight lang="bash">
hdbuserstore SET DEFAULT <FQDN:30013>@TENANT SAPABAP1 'PASSWORD'
</syntaxhighlight>Hinweis: Auch wenn Kennwörter geändert werden, wie z.B. vom Backupuser, dann muss der hdbuserstore neu gesetzt werden.<syntaxhighlight lang="bash">
hdbuserstore SET BACKUP <FQDN>:30013@SYSTEMDB BACKUP_OPERATOR 'PASSWORD'
</syntaxhighlight>

===Primary/Secondary Schwenk===
ACHTUNG!! NUR CODESCHNIPSEL!!! Anleitung unvollständig
<syntaxhighlight lang="bash">
Primary
hdbnsutil -sr_enable --name=S4PDB1

Secondary
hdbnsutil -sr_unregister
hdbnsutil -sr_register --remoteHost=simas4pdb --remoteInstance=00 --replicationMode=sync --operationMode=logreplay --name=S4PDB2 --force_full_replica --online
HDB start
</syntaxhighlight>

===Test===
<syntaxhighlight lang="bash">
HDB info
HDB version
</syntaxhighlight>

=== SAP User anzeigen lassen ===
<syntaxhighlight lang="sql">
hdbsql -n localhost:30013 -d <SID>-u SAPABAP1 -p '<PW>' "SELECT BNAME, USTYP FROM SAPABAP1.USR02 WHERE MANDT = '000'"
</syntaxhighlight>

=== SAP User kopieren ===
Dieser Befehl kopiert den User von 000 nach 210
<syntaxhighlight lang="sql">
hdbsql -n localhost:30013 -d <SID>-u SAPABAP1 -p '<PW>' "DELETE FROM SAPABAP1.USR02 WHERE MANDT = '210' AND BNAME = '<USER>'"; INSERT INTO SAPABAP1.USR02 (MANDT, BNAME, BCODE, GLTGV, GLTGB, USTYP, CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT, TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2, BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4, BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN, CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE, RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE, PWDSALTEDHASH, SECURITY_POLICY ) SELECT '210', BNAME, BCODE, GLTGV, GLTGB, USTYP, CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT, TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2, BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4, BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN, CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE, RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE, PWDSALTEDHASH, SECURITY_POLICY FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<USER>';"
</syntaxhighlight>

=== Passwort Hash eines einzelnen Users kopieren ===
<syntaxhighlight lang="sql">
UPDATE SAPABAP1.USR02
SET
BCODE = (SELECT BCODE FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<NAME>'),
PASSCODE = (SELECT PASSCODE FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<NAME>'),
PWDSALTEDHASH = (SELECT PWDSALTEDHASH FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<NAME>'),
CODVN = (SELECT CODVN FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<NAME>'),
PWDINITIAL = 0,
UFLAG = 0,
GLTGB = '99991231'
WHERE MANDT = '321'
AND BNAME = '<NAME>';
COMMIT;

</syntaxhighlight>

=== Passwort Hashes kompletter Mandanten kopieren ===
<syntaxhighlight lang="sql">
UPDATE SAPABAP1.USR02 T_ZIEL
SET T_ZIEL.PWDSALTEDHASH = (
SELECT T_QUELLE.PWDSALTEDHASH
FROM SAPABAP1.USR02 T_QUELLE
WHERE T_QUELLE.MANDT = '300'
AND T_QUELLE.BNAME = T_ZIEL.BNAME
)
WHERE T_ZIEL.MANDT = '301'
AND EXISTS (
SELECT 1
FROM SAPABAP1.USR02 T_CHECK
WHERE T_CHECK.MANDT = '300'
AND T_CHECK.BNAME = T_ZIEL.BNAME
);
COMMIT;
</syntaxhighlight>

=== Backup abbrechen ===
<syntaxhighlight lang="sql">
BACKUP CANCEL FOR <SYSTEMDB/TENANT> <ID>
</syntaxhighlight>
===Störungen===
==== * 447: backup could not be completed: [110164] Encryption root keys backup password doesn't exist. It must be either set in the database or must be given with the backup/recover SQL statement SQLSTATE: HY000 ====
Die Datenbank wurde mit aktivierten Encryption root keys backup password installiert, aber sind noch nicht in der Datenbank gesetzt. Dazu auf der SYSTEMDB und TENANT ausführen:
<syntaxhighlight lang="sql">
ALTER SYSTEM SET ENCRYPTION ROOT KEYS BACKUP PASSWORD "<PW>";
</syntaxhighlight>
====Tabelle BALDAT====
in der SLG2 je Mandant alles löschen und den Job SBAL_DELETE einplanen
<syntaxhighlight lang="bash">
</syntaxhighlight>
====Transparent Huge Pages (THP) are activated or not readable====
HANA Cockpit -> SystemDB -> Database Overview -> Alerting and Diagnostics -> Alert Definitions
ID 116 auswählen Transparent Huge Pages status ( ID 116)
Edit und den Schalter "Schedule Active" auf "no" setzen und "save"
===Codeschnipsel===
SELECT * FROM SYS.M_TABLES where SCHEMA_NAME ='SAPHANADB' AND TABLE_NAME = 'LMSEMAPHORE';

===Nützliche Links===

SAP MaxDB

2026-04-10T14:28:52Z

Hendrik: /* Logbackups */

=== Beschreibung ===

=== Download ===

=== Installation ===

=== Konfiguration ===

=== Backups ===
==== Fullbackup ====
Als Einzeiler
<syntaxhighlight lang="bash" line="1">
dbmcli -d SID -u superdba,PW -uUTL -c backup_start SID_PIPE_FULL
</syntaxhighlight>
Oder
<syntaxhighlight lang="bash" line="1">
dbmcli -d <SID> -u superdba,<PW>
util_connect
backup_start <SID>_PIPE_FULL
</syntaxhighlight>
==== Logbackups ====
<syntaxhighlight lang="bash" line="1">
dbmcli -d <SID> -u superdba,<PW>
util_connect
autolog_on SID_STAGE_AUTOLOG INTERVAL 1800
</syntaxhighlight>

==== Kennwörter ändern ====
===== CONTROL =====
Hinweis: Wenn Monitoring-Checks "lila" sind, ist noch folgendes auszuführen: sudo /opt/nrpe/addxuser
<syntaxhighlight lang="bash" line="1">
dbmcli -d <SID> -u control,<passwort_alt> user_put control password=<passwort_neu>
xuser -U c clear
xuser -U c -u CONTROL,<passwort> -d SID -n Hostname -S INTERNAL
dbmcli -d $SID -u superdba,<PW> db_clear
</syntaxhighlight>

===== SUPERDBA =====
<syntaxhighlight lang="bash" line="1">
dbmcli -d <SID> -u control,<PW> user_put superdba password=<PW>
dbmcli -d <SID> -u control,<PW> load_systab -u superdba,<PW>
xuser -U w clear
xuser -U w -u superdba,<PW> -d <SID> -n <HOST> -S INTERNAL
dbmcli -d $SID -u superdba,<PW> db_clear
</syntaxhighlight>

===== SAPSID =====
<syntaxhighlight lang="bash" line="1">
stopsap R3
dbmcli -d <SID> -u superdba,<PW> -uSQL SAP<SID>,<PW> sql_execute ALTER PASSWORD <PWALT> TO <PWNEW>
xuser -U DEFAULT clear
xuser -U DEFAULT -u SAP<SID>,<PW> -d <SID> -n <HOST> -S SAPR3 -t 0
dbmcli -d $SID -u superdba,<PW> db_clear
startsap R3
</syntaxhighlight>

===== SAPSIDDB (Java oder ADS) =====
<syntaxhighlight lang="bash" line="1">
stopsap J2EE
dbmcli -d <SID> -u superdba,<PW> -uSQL SAPADKDB,<PWALT> sql_execute ALTER PASSWORD <PWALT> TO <PWNEW>
xuser -U DEFAULT clear
xuser -U DEFAULT -u SAP<SID>DB,<PW> -d <SID> -n <HOST> -S SAPR3 -t 0
</syntaxhighlight>
Im Configtool unter Connection Pools "Password" das neue Passwort eintragen
<syntaxhighlight lang="bash" line="1">
cd /usr/sap/<SID>/J00/j2ee/configtool/ && ./configtool.sh
</syntaxhighlight>
<syntaxhighlight lang="bash" line="1">
startsap J2EE
</syntaxhighlight>

=== Update ===

=== Test ===

=== Fehlerbehebung ===
==== Disk autolog voll und AUTOSAVE IS OFF ====
Die Datenbank Logs der MaxDB werden automatisch in das Verzeichnis /sapdb/<SID>/autolog geschrieben. Sofern es Fehler bei der Sicherung dieser Dateien gibt, kann die Platte schnell volllaufen.
===== Lösung =====
<syntaxhighlight lang="bash" line="1">
dbmcli -d <SID> -u superdba,<PW>
dbmcli on IGK>db_state
OK
State
ONLINE

---
dbmcli on IGK>autolog_show
OK
AUTOSAVE IS OFF

---
dbmcli on IGK>medium_getall
OK
IGK_PIPE_AUTOLOG /backup/pipes/IGK_AUTOLOG PIPE LOG 0 8 NO NO 20230714144617 20230714144617 BACK 0
IGK_STAGE_AUTOLOG /sapdb/IGK/autolog/IGK_AUTOLOG FILE AUTO 0 8 NO NO 20230714144855 20230714144855 NONE 0
IGK_PIPE_INC\P01 /backup/pipes/IGK_INC_01,* PIPE PAGES 0 8 NO NO 20230714145322 20230714145322 BACK 0
IGK_PIPE_INC\P02 /backup/pipes/IGK_INC_02,* PIPE PAGES 0 8 NO NO 20230714145322 20230714145322 BACK 0
IGK_PIPE_FULL\P01 /backup/pipes/IGK_FULL_01,* PIPE DATA 0 8 NO NO 20230714145434 20230714145434 BACK 0
IGK_PIPE_FULL\P02 /backup/pipes/IGK_FULL_02,* PIPE DATA 0 8 NO NO 20230714145434 20230714145434 BACK 0
FULL_INITIAL /dev/null FILE DATA 0 8 YES NO 20230714145632 20230714145632 NONE 0

---
dbmcli on IGK>autolog_on IGK_STAGE_AUTOLOG INTERVAL 1800
OK

---
dbmcli on IGK>autolog_show
OK
AUTOSAVE IS ON
</syntaxhighlight>
Anschließend am besten nochmal alle xuser mit xuser_list und "dbmcli -U <XUSER_KEY> db_state" überprüfen
Falls autosave weiterhin off ist zunächst autolog erweitern

=== Codeschnipsel ===

=== Nützliche Links ===

=== Codeschnipsel ===

=== Nützliche Links ===

SAP MaxDB

2026-04-10T14:28:18Z

Hendrik: /* Backups */

=== Beschreibung ===

=== Download ===

=== Installation ===

=== Konfiguration ===

=== Backups ===
==== Fullbackup ====
Als Einzeiler
<syntaxhighlight lang="bash" line="1">
dbmcli -d SID -u superdba,PW -uUTL -c backup_start SID_PIPE_FULL
</syntaxhighlight>
Oder
<syntaxhighlight lang="bash" line="1">
dbmcli -d <SID> -u superdba,<PW>
util_connect
backup_start <SID>_PIPE_FULL
</syntaxhighlight>
==== Logbackups ====
<syntaxhighlight lang="bash" line="1">
dbmcli -d <SID> -u superdba,<PW>
util_connect
backup_start <SID>_PIPE_FULL
</syntaxhighlight>
==== Kennwörter ändern ====
===== CONTROL =====
Hinweis: Wenn Monitoring-Checks "lila" sind, ist noch folgendes auszuführen: sudo /opt/nrpe/addxuser
<syntaxhighlight lang="bash" line="1">
dbmcli -d <SID> -u control,<passwort_alt> user_put control password=<passwort_neu>
xuser -U c clear
xuser -U c -u CONTROL,<passwort> -d SID -n Hostname -S INTERNAL
dbmcli -d $SID -u superdba,<PW> db_clear
</syntaxhighlight>

===== SUPERDBA =====
<syntaxhighlight lang="bash" line="1">
dbmcli -d <SID> -u control,<PW> user_put superdba password=<PW>
dbmcli -d <SID> -u control,<PW> load_systab -u superdba,<PW>
xuser -U w clear
xuser -U w -u superdba,<PW> -d <SID> -n <HOST> -S INTERNAL
dbmcli -d $SID -u superdba,<PW> db_clear
</syntaxhighlight>

===== SAPSID =====
<syntaxhighlight lang="bash" line="1">
stopsap R3
dbmcli -d <SID> -u superdba,<PW> -uSQL SAP<SID>,<PW> sql_execute ALTER PASSWORD <PWALT> TO <PWNEW>
xuser -U DEFAULT clear
xuser -U DEFAULT -u SAP<SID>,<PW> -d <SID> -n <HOST> -S SAPR3 -t 0
dbmcli -d $SID -u superdba,<PW> db_clear
startsap R3
</syntaxhighlight>

===== SAPSIDDB (Java oder ADS) =====
<syntaxhighlight lang="bash" line="1">
stopsap J2EE
dbmcli -d <SID> -u superdba,<PW> -uSQL SAPADKDB,<PWALT> sql_execute ALTER PASSWORD <PWALT> TO <PWNEW>
xuser -U DEFAULT clear
xuser -U DEFAULT -u SAP<SID>DB,<PW> -d <SID> -n <HOST> -S SAPR3 -t 0
</syntaxhighlight>
Im Configtool unter Connection Pools "Password" das neue Passwort eintragen
<syntaxhighlight lang="bash" line="1">
cd /usr/sap/<SID>/J00/j2ee/configtool/ && ./configtool.sh
</syntaxhighlight>
<syntaxhighlight lang="bash" line="1">
startsap J2EE
</syntaxhighlight>

=== Update ===

=== Test ===

=== Fehlerbehebung ===
==== Disk autolog voll und AUTOSAVE IS OFF ====
Die Datenbank Logs der MaxDB werden automatisch in das Verzeichnis /sapdb/<SID>/autolog geschrieben. Sofern es Fehler bei der Sicherung dieser Dateien gibt, kann die Platte schnell volllaufen.
===== Lösung =====
<syntaxhighlight lang="bash" line="1">
dbmcli -d <SID> -u superdba,<PW>
dbmcli on IGK>db_state
OK
State
ONLINE

---
dbmcli on IGK>autolog_show
OK
AUTOSAVE IS OFF

---
dbmcli on IGK>medium_getall
OK
IGK_PIPE_AUTOLOG /backup/pipes/IGK_AUTOLOG PIPE LOG 0 8 NO NO 20230714144617 20230714144617 BACK 0
IGK_STAGE_AUTOLOG /sapdb/IGK/autolog/IGK_AUTOLOG FILE AUTO 0 8 NO NO 20230714144855 20230714144855 NONE 0
IGK_PIPE_INC\P01 /backup/pipes/IGK_INC_01,* PIPE PAGES 0 8 NO NO 20230714145322 20230714145322 BACK 0
IGK_PIPE_INC\P02 /backup/pipes/IGK_INC_02,* PIPE PAGES 0 8 NO NO 20230714145322 20230714145322 BACK 0
IGK_PIPE_FULL\P01 /backup/pipes/IGK_FULL_01,* PIPE DATA 0 8 NO NO 20230714145434 20230714145434 BACK 0
IGK_PIPE_FULL\P02 /backup/pipes/IGK_FULL_02,* PIPE DATA 0 8 NO NO 20230714145434 20230714145434 BACK 0
FULL_INITIAL /dev/null FILE DATA 0 8 YES NO 20230714145632 20230714145632 NONE 0

---
dbmcli on IGK>autolog_on IGK_STAGE_AUTOLOG INTERVAL 1800
OK

---
dbmcli on IGK>autolog_show
OK
AUTOSAVE IS ON
</syntaxhighlight>
Anschließend am besten nochmal alle xuser mit xuser_list und "dbmcli -U <XUSER_KEY> db_state" überprüfen
Falls autosave weiterhin off ist zunächst autolog erweitern

=== Codeschnipsel ===

=== Nützliche Links ===

=== Codeschnipsel ===

=== Nützliche Links ===

SAP MaxDB

2026-04-10T14:27:37Z

Hendrik:

=== Beschreibung ===

=== Download ===

=== Installation ===

=== Konfiguration ===

=== Backups ===
==== Fullbackup ====
Als Einzeiler
<syntaxhighlight lang="bash" line="1">
dbmcli -d SID -u superdba,PW -uUTL -c backup_start SID_PIPE_FULL
</syntaxhighlight>
Oder
<syntaxhighlight lang="bash" line="1">
dbmcli -d <SID> -u superdba,<PW>
util_connect
backup_start <SID>_PIPE_FULL
</syntaxhighlight>
==== Logbackups ====
dbmcli -d <SID> -u superdba,<PW>
util_connect
backup_start <SID>_PIPE_FULL
==== Kennwörter ändern ====
===== CONTROL =====
Hinweis: Wenn Monitoring-Checks "lila" sind, ist noch folgendes auszuführen: sudo /opt/nrpe/addxuser
<syntaxhighlight lang="bash" line="1">
dbmcli -d <SID> -u control,<passwort_alt> user_put control password=<passwort_neu>
xuser -U c clear
xuser -U c -u CONTROL,<passwort> -d SID -n Hostname -S INTERNAL
dbmcli -d $SID -u superdba,<PW> db_clear
</syntaxhighlight>

===== SUPERDBA =====
<syntaxhighlight lang="bash" line="1">
dbmcli -d <SID> -u control,<PW> user_put superdba password=<PW>
dbmcli -d <SID> -u control,<PW> load_systab -u superdba,<PW>
xuser -U w clear
xuser -U w -u superdba,<PW> -d <SID> -n <HOST> -S INTERNAL
dbmcli -d $SID -u superdba,<PW> db_clear
</syntaxhighlight>

===== SAPSID =====
<syntaxhighlight lang="bash" line="1">
stopsap R3
dbmcli -d <SID> -u superdba,<PW> -uSQL SAP<SID>,<PW> sql_execute ALTER PASSWORD <PWALT> TO <PWNEW>
xuser -U DEFAULT clear
xuser -U DEFAULT -u SAP<SID>,<PW> -d <SID> -n <HOST> -S SAPR3 -t 0
dbmcli -d $SID -u superdba,<PW> db_clear
startsap R3
</syntaxhighlight>

===== SAPSIDDB (Java oder ADS) =====
<syntaxhighlight lang="bash" line="1">
stopsap J2EE
dbmcli -d <SID> -u superdba,<PW> -uSQL SAPADKDB,<PWALT> sql_execute ALTER PASSWORD <PWALT> TO <PWNEW>
xuser -U DEFAULT clear
xuser -U DEFAULT -u SAP<SID>DB,<PW> -d <SID> -n <HOST> -S SAPR3 -t 0
</syntaxhighlight>
Im Configtool unter Connection Pools "Password" das neue Passwort eintragen
<syntaxhighlight lang="bash" line="1">
cd /usr/sap/<SID>/J00/j2ee/configtool/ && ./configtool.sh
</syntaxhighlight>
<syntaxhighlight lang="bash" line="1">
startsap J2EE
</syntaxhighlight>

=== Update ===

=== Test ===

=== Fehlerbehebung ===
==== Disk autolog voll und AUTOSAVE IS OFF ====
Die Datenbank Logs der MaxDB werden automatisch in das Verzeichnis /sapdb/<SID>/autolog geschrieben. Sofern es Fehler bei der Sicherung dieser Dateien gibt, kann die Platte schnell volllaufen.
===== Lösung =====
<syntaxhighlight lang="bash" line="1">
dbmcli -d <SID> -u superdba,<PW>
dbmcli on IGK>db_state
OK
State
ONLINE

---
dbmcli on IGK>autolog_show
OK
AUTOSAVE IS OFF

---
dbmcli on IGK>medium_getall
OK
IGK_PIPE_AUTOLOG /backup/pipes/IGK_AUTOLOG PIPE LOG 0 8 NO NO 20230714144617 20230714144617 BACK 0
IGK_STAGE_AUTOLOG /sapdb/IGK/autolog/IGK_AUTOLOG FILE AUTO 0 8 NO NO 20230714144855 20230714144855 NONE 0
IGK_PIPE_INC\P01 /backup/pipes/IGK_INC_01,* PIPE PAGES 0 8 NO NO 20230714145322 20230714145322 BACK 0
IGK_PIPE_INC\P02 /backup/pipes/IGK_INC_02,* PIPE PAGES 0 8 NO NO 20230714145322 20230714145322 BACK 0
IGK_PIPE_FULL\P01 /backup/pipes/IGK_FULL_01,* PIPE DATA 0 8 NO NO 20230714145434 20230714145434 BACK 0
IGK_PIPE_FULL\P02 /backup/pipes/IGK_FULL_02,* PIPE DATA 0 8 NO NO 20230714145434 20230714145434 BACK 0
FULL_INITIAL /dev/null FILE DATA 0 8 YES NO 20230714145632 20230714145632 NONE 0

---
dbmcli on IGK>autolog_on IGK_STAGE_AUTOLOG INTERVAL 1800
OK

---
dbmcli on IGK>autolog_show
OK
AUTOSAVE IS ON
</syntaxhighlight>
Anschließend am besten nochmal alle xuser mit xuser_list und "dbmcli -U <XUSER_KEY> db_state" überprüfen
Falls autosave weiterhin off ist zunächst autolog erweitern

=== Codeschnipsel ===

=== Nützliche Links ===

=== Codeschnipsel ===

=== Nützliche Links ===

SAP HANA

2026-03-31T16:22:35Z

Hendrik: /* SAP HANA Update */

=== Beschreibung ===
SAP HANA ist eine In-Memory-Datenbank-Plattform, die von SAP entwickelt wurde. Sie ermöglicht es Unternehmen große Datenmengen in Echtzeit zu verarbeiten und zu analysieren. Die Plattform unterstützt sowohl transaktionale als auch analytische Anwendungen und ermöglicht es den Benutzern auf umfassende Echtzeit-Analysen zuzugreifen.

Die In-Memory-Technologie von SAP HANA sorgt dafür, dass Daten in einem schnellen Zugriffsspeicher gespeichert werden, anstatt auf langsameren Festplatten. Dies führt zu schnelleren Datenzugriffszeiten und ermöglicht es Unternehmen Echtzeit-Transaktionen und Analysen durchzuführen.

SAP HANA bietet auch eine Vielzahl von Tools und Anwendungen um Unternehmen bei der Verwaltung von Daten zu unterstützen. Dazu gehören Tools zur Datenintegration, Datenbereinigung, Datenmodellierung und Analyse. Es unterstützt auch eine Vielzahl von Programmiersprachen und Entwicklungsplattformen, einschließlich SAPUI5, Java, Python uvm.
=== Download ===
[https://me.sap.com/softwarecenter/support/index SAP Downloadportal]

<nowiki>https://launchpad.support.sap.com/#/softwarecenter</nowiki> SUPPORT PACKAGES & PATCHES By Category SAP IN-MEMORY (SAP HANA ) HANA PLATFORM EDITION SAP HANA PLATFORM EDITION SAP HANA PLATFORM EDITION 2.0 SAP HANA CLIENT 2.0

=== Installation ===
hdblcm nach installieren:
https://me.sap.com/notes/2651885<syntaxhighlight lang="bash">
<install_media>/SAP_HANA_DATABASE/hdblcm --sid=<SID> --action=update --components=hdblcm --ignore=check_signature_file --verify_signature=off
</syntaxhighlight>

=== Konfiguration===

===SAP HANA Update===
Aktuelle Version als <SID>ADM mit "HDB version" anzeigen lassen

IMDB_CLIENT<VERSION> und IMDB_SERVER<VERSION> von SAP herunterladen und nach /usr/sap/SUM kopieren
<syntaxhighlight lang="bash">
sudo su - root
cd /usr/sap/SUM
chmod +x SAPCAR && ./SAPCAR -xvf IMDB_CLIENT*
chmod +x SAPCAR && ./SAPCAR -xvf IMDB_SERVER*
cd /hana/shared/*/hdblcm && ./hdblcm --action=update --ignore=check_signature_file --verify_signature=off --lss_trust_unsigned_components
Enter directory root to search for components: /usr/sap/SUM/
</syntaxhighlight>
Mit "1" alles updaten und weiter den Anweisungen folgen.

===SAP HANA Sizing Report===
Quelle: [https://blogs.sap.com/2021/05/06/how-to-install-run-the-abap-on-hana-sizing-report-sap-note-1872170-a-step-by-step-guide/ HANA Sizing Report]
# Note updaten. Prüfen, ob 2810633 einbaubar ist, ansonsten 2504480 einbauen. Wenn beide nicht einbaubar sind ist das System zu alt oder bereits aktuell.
# SA38 aufrufen und /SDF/HDB_SIZING ausführen
# "Number of parallel dialog processes" auf 3 setzen
# Ausführen oder als Hintergrundjob einplanen

=== Kennwort ändern ===
<syntaxhighlight lang="bash">
ALTER USER <USER> PASSWORD "<new_password>";
</syntaxhighlight>
Oder direkt mit dem <SID>adm in der Shell:
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER <USER> PASSWORD <PW>"
</syntaxhighlight>
Hinweise:
* für System@TENANT und SYSTEM@SYSTEMDB muss der HDBUSERSTORE für das Backupscript aktualisiert werden und für die Anmeldung im HDB Studio

Lösungsschritte:

Prüfen, welche Keys angelegt sind:

<syntaxhighlight lang="bash">
hdbuserstore list
</syntaxhighlight>
Alten Key löschen (Beispiel: BACKUPKEY für SystemDB):

<syntaxhighlight lang="bash">
hdbuserstore delete BACKUP
</syntaxhighlight>
Neuen Key anlegen und dabei das aktualisierte SYSTEM‑Passwort verwenden:

<syntaxhighlight lang="bash">
hdbuserstore set BACKUP <FQDN>:30013 SYSTEM "<PW>"
</syntaxhighlight>

====SAPABAP1 Kennwort ändern====

Direkt auf Applikationsserver:
<syntaxhighlight lang="bash">
hdbsql -n <FQDN>:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPABAP1 PASSWORD <PW>"
hdbuserstore set default <FQDN>:30013@<TENANT> SAPABAP1 "<PW>"
</syntaxhighlight>
Oder im HDB Studio:
<syntaxhighlight lang="bash">
ALTER USER SAPABAP1 PASSWORD "<PW>";
</syntaxhighlight>
[https://me.sap.com/notes/0002750829 2750829 - Ändern des SAPABAP1-Schemakennworts in SAP HANA]
Anschließend siehe "SAP HANA HDB Userstore neu setzen"

==== SAPDBCTRL (SAP Host Agent) ====
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d SYSTEMDB -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
</syntaxhighlight>
Dann als root fortfahren.
Für SystemDB, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname SYSTEMDB@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>
Für Tenant, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname IMP@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>

=== Encryption root keys backup password setzen ===
<syntaxhighlight lang="bash">
ALTER SYSTEM SET ENCRYPTION ROOT KEYS BACKUP PASSWORD "<PW>";
</syntaxhighlight>
===SAP HANA HDB Userstore neu setzen===
<syntaxhighlight lang="bash">
hdbuserstore SET DEFAULT <FQDN:30013>@TENANT SAPABAP1 'PASSWORD'
</syntaxhighlight>Hinweis: Auch wenn Kennwörter geändert werden, wie z.B. vom Backupuser, dann muss der hdbuserstore neu gesetzt werden.<syntaxhighlight lang="bash">
hdbuserstore SET BACKUP <FQDN>:30013@SYSTEMDB BACKUP_OPERATOR 'PASSWORD'
</syntaxhighlight>

===Primary/Secondary Schwenk===
ACHTUNG!! NUR CODESCHNIPSEL!!! Anleitung unvollständig
<syntaxhighlight lang="bash">
Primary
hdbnsutil -sr_enable --name=S4PDB1

Secondary
hdbnsutil -sr_unregister
hdbnsutil -sr_register --remoteHost=simas4pdb --remoteInstance=00 --replicationMode=sync --operationMode=logreplay --name=S4PDB2 --force_full_replica --online
HDB start
</syntaxhighlight>

===Test===
<syntaxhighlight lang="bash">
HDB info
HDB version
</syntaxhighlight>

=== SAP User anzeigen lassen ===
<syntaxhighlight lang="sql">
hdbsql -n localhost:30013 -d <SID>-u SAPABAP1 -p '<PW>' "SELECT BNAME, USTYP FROM SAPABAP1.USR02 WHERE MANDT = '000'"
</syntaxhighlight>

=== SAP User kopieren ===
Dieser Befehl kopiert den User von 000 nach 210
<syntaxhighlight lang="sql">
hdbsql -n localhost:30013 -d <SID>-u SAPABAP1 -p '<PW>' "DELETE FROM SAPABAP1.USR02 WHERE MANDT = '210' AND BNAME = '<USER>'"; INSERT INTO SAPABAP1.USR02 (MANDT, BNAME, BCODE, GLTGV, GLTGB, USTYP, CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT, TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2, BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4, BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN, CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE, RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE, PWDSALTEDHASH, SECURITY_POLICY ) SELECT '210', BNAME, BCODE, GLTGV, GLTGB, USTYP, CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT, TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2, BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4, BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN, CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE, RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE, PWDSALTEDHASH, SECURITY_POLICY FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<USER>';"
</syntaxhighlight>

=== Passwort Hash eines einzelnen Users kopieren ===
<syntaxhighlight lang="sql">
UPDATE SAPABAP1.USR02
SET
BCODE = (SELECT BCODE FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<NAME>'),
PASSCODE = (SELECT PASSCODE FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<NAME>'),
PWDSALTEDHASH = (SELECT PWDSALTEDHASH FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<NAME>'),
CODVN = (SELECT CODVN FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<NAME>'),
PWDINITIAL = 0,
UFLAG = 0,
GLTGB = '99991231'
WHERE MANDT = '321'
AND BNAME = '<NAME>';
COMMIT;

</syntaxhighlight>

=== Passwort Hashes kompletter Mandanten kopieren ===
<syntaxhighlight lang="sql">
UPDATE SAPABAP1.USR02 T_ZIEL
SET T_ZIEL.PWDSALTEDHASH = (
SELECT T_QUELLE.PWDSALTEDHASH
FROM SAPABAP1.USR02 T_QUELLE
WHERE T_QUELLE.MANDT = '300'
AND T_QUELLE.BNAME = T_ZIEL.BNAME
)
WHERE T_ZIEL.MANDT = '301'
AND EXISTS (
SELECT 1
FROM SAPABAP1.USR02 T_CHECK
WHERE T_CHECK.MANDT = '300'
AND T_CHECK.BNAME = T_ZIEL.BNAME
);
COMMIT;
</syntaxhighlight>

===Störungen===
====Tabelle BALDAT====
in der SLG2 je Mandant alles löschen und den Job SBAL_DELETE einplanen
<syntaxhighlight lang="bash">
</syntaxhighlight>
====Transparent Huge Pages (THP) are activated or not readable====
HANA Cockpit -> SystemDB -> Database Overview -> Alerting and Diagnostics -> Alert Definitions
ID 116 auswählen Transparent Huge Pages status ( ID 116)
Edit und den Schalter "Schedule Active" auf "no" setzen und "save"
===Codeschnipsel===
SELECT * FROM SYS.M_TABLES where SCHEMA_NAME ='SAPHANADB' AND TABLE_NAME = 'LMSEMAPHORE';

===Nützliche Links===

VLLm

2026-03-21T00:40:24Z

Hendrik:

=== Beschreibung ===
Docker normal installieren

=== Download ===
Normal (ROCm)
<syntaxhighlight lang="bash" line="1">
docker pull rocm/vllm-dev:nightly
</syntaxhighlight>
gfx906
<syntaxhighlight lang="bash" line="1">
docker pull nalanzeyu/vllm-gfx906
</syntaxhighlight>

=== Ausführen ===
==== Variante 1 ====
<syntaxhighlight lang="bash" line="1">
docker run -it --rm --shm-size=8g --device=/dev/kfd --device=/dev/dri \
--group-add video -p 8086:8000 \
-v /mnt/share/models:/models \
nalanzeyu/vllm-gfx906 \
vllm serve /models/Qwen3-Coder-30B-A3B-Instruct-AWQ-4bit --served-model-name Homelab --max-model-len 30000 --enable-auto-tool-choice --tool-call-parser hermes
</syntaxhighlight>Variante 2, getestet 18.12.2025:<syntaxhighlight lang="bash">
sudo docker run -it --rm --network=host --group-add=video --ipc=host --cap-add=SYS_PTRACE --security-opt seccomp=unconfined --device /dev/kfd --device /dev/dri -v /home/hendrik/.lmstudio/models/:/app/models -e HF_HOME="/app/models" -e HF_TOKEN="<TOKEN>" -e NCCL_P2P_DISABLE=1 -e VLLM_CUSTOM_OPS=all -e VLLM_ROCM_USE_AITER=0 -e SAFETENSORS_FAST_GPU=1 -e PYTORCH_TUNABLEOP_ENABLED=1 rocm/vllm-dev:nightly

</syntaxhighlight>
Für gfx1201:<syntaxhighlight lang="bash">
sudo docker run -it --rm --network=host \
--group-add=video --ipc=host --cap-add=SYS_PTRACE \
--security-opt seccomp=unconfined --device /dev/kfd \
--device /dev/dri \
-v /home/hendrik/.lmstudio/models/:/app/models \
-e HF_HOME="/app/models" \
-e HF_TOKEN="<TOKEN>" \
-e NCCL_P2P_DISABLE=1 \
-e VLLM_CUSTOM_OPS=all \
-e VLLM_ROCM_USE_AITER=0 \
-e SAFETENSORS_FAST_GPU=1 \
-e PYTORCH_TUNABLEOP_ENABLED=1
kyuz0/vllm-therock-gfx1201
</syntaxhighlight>

Ohne Tensor Parallism:<syntaxhighlight lang="bash">
vllm serve Qwen/Qwen3-VL-8B-Thinking --served-model-name Homelab --max_model_len 4096 --enable-auto-tool-choice --tool-call-parser hermes --reasoning-parser qwen3
</syntaxhighlight>Mit:<syntaxhighlight lang="bash">
vllm serve Qwen/Qwen3-VL-8B-Thinking --served-model-name Homelab --tp 2 --max_model_len 4096 --enable-auto-tool-choice --tool-call-parser hermes --reasoning-parser qwen3
</syntaxhighlight>
Benchmark:
<syntaxhighlight lang="bash">
vllm bench serve --num-prompts 1 --dataset-name=random --input-len 512 --output-len 128 --model Qwen/Qwen3-4B-Instruct-2507-FP8
</syntaxhighlight>
==== Variante 2 (Pro W7800) ====
<syntaxhighlight lang="bash" line="1">
#!/bin/bash
docker run --rm \
--device /dev/kfd \
--device /dev/dri \
-e HSA_ENABLE_IPC_MODE_LEGACY=0 \
-e HIP_VISIBLE_DEVICES=1 \
-p 8000:8000 \
-v ~/.cache/huggingface:/root/.cache/huggingface \
--ipc=host \
rocm/vllm-dev:nightly_main_20260318 \
vllm serve cyankiwi/Qwen3.5-35B-A3B-AWQ-4bit \
--tensor-parallel-size 1 \
--max-model-len 16000 \
--dtype float16 \
--reasoning-parser qwen3 \
--speculative-config '{"method":"qwen3_next_mtp","num_speculative_tokens":2}' \
--enable-auto-tool-choice \
--tool-call-parser qwen3_coder \
--language-model-only \
--served-model-name Homelab

</syntaxhighlight>
=== Test ===
=== Bekannte Probleme ===
=== Nützliche Links ===
* [https://github.com/ggml-org/llama.cpp Offizielles llama.cpp Repository]
* [https://github.com/ggml-org/llama.cpp/blob/master/examples/server/README.md llama-server Dokumentation]
* [https://github.com/ggml-org/llama.cpp/discussions llama.cpp Discussions]
* [https://huggingface.co/models?library=gguf GGUF Modelle auf Hugging Face]
* [https://docs.rocm.com/ ROCm Dokumentation]
* [https://vulkan.lunarg.com/doc/sdk Vulkan SDK Dokumentation]

DAA Agent

2026-03-06T09:30:39Z

Hendrik:

=== Beschreibung ===
Diese Anleitung beschreibt das manuelle Update der SAP JVM für den Diagnostics Agent (DAA) unter Linux. Da der Agent seine JVM nicht automatisch patchen kann, muss das Update manuell über das Anlegen eines neuen Verzeichnisses und die Anpassung des Instanzprofils erfolgen.

=== Download ===
Die aktuelle SAP JVM 8.1 im `.SAR`-Format muss aus dem SAP Support Portal (Software Center) heruntergeladen werden.
* Dateibeispiel: `SAPJVM8_108-80000202.SAR`

=== Installation ===
''Die Erstinstallation des Diagnostics Agents ist nicht Teil dieses Artikels. Dieser Artikel behandelt ausschließlich das Update der zugrundeliegenden SAP JVM.''

=== Konfiguration ===
Es ist keine spezielle Vorab-Konfiguration auf Betriebssystemebene nötig, außer dass das Tool `SAPCAR` verfügbar sein muss und Root-Rechte für die Verzeichniserstellung benötigt werden.

=== Update ===
Das Update erfolgt durch das Anlegen eines neuen JVM-Verzeichnisses und dem Anpassen der Profil-Datei.

<syntaxhighlight lang="bash" line="1">
# 1. Neues Verzeichnis als root/sudo anlegen und Rechte setzen (Version im Namen anpassen)
mkdir -p /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108
chown daaadm:sapsys /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108

# 2. Zum DAA-User wechseln und in das neue Verzeichnis gehen
su - daaadm
cd /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108

# 3. Das heruntergeladene Archiv entpacken (Pfade anpassen)
/usr/sap/DAA/SYS/exe/uc/linuxx86_64/SAPCAR -xvf /Pfad/zur/Datei/SAPJVM8_108-80000202.SAR
</syntaxhighlight>

Im Anschluss muss das Instanzprofil des Diagnostics Agents angepasst werden:

<syntaxhighlight lang="bash" line="1">
# Profil editieren (Hostname anpassen)
vi /usr/sap/DAA/SYS/profile/DAA_SMDA98_<Hostname>
</syntaxhighlight>

Dort die Variable `SAPJVM_VERSION` auf die neue Version ändern:
<syntaxhighlight lang="ini" line="1">
# Vorher:
SAPJVM_VERSION = 8.1.097

# Nachher:
SAPJVM_VERSION = 8.1.108
</syntaxhighlight>

Den Agenten als `daaadm` neu starten, damit `sapcpe` die neuen Binaries kopiert:
<syntaxhighlight lang="bash" line="1">
stopsap
startsap
</syntaxhighlight>

=== Test ===
Nach dem Neustart prüfen, ob die neue JVM aktiv ist (als User `daaadm`):

<syntaxhighlight lang="bash" line="1">
# 1. Java-Version im Ausführungsverzeichnis prüfen
/usr/sap/DAA/SMDA98/exe/sapjvm_8/bin/java -version

# 2. Prüfen, ob der laufende Agent-Prozess den korrekten Java-Pfad nutzt
ps -ef | grep DAA | grep java

# 3. sapcpe-Log prüfen, ob aus dem neuen Verzeichnis kopiert wurde
grep "source" /usr/sap/DAA/SMDA98/work/sapcpe.log | tail -n 5
</syntaxhighlight>

=== Fehlerbehebung===
====Starten/Stoppen funktioniert nicht====
Wenn `startsap` oder `stopsap` mit Fehlern abbrechen oder der Agent nicht hochfährt:

=====Lösung 1=====
Prüfen, ob die Berechtigungen des neu angelegten Verzeichnisses korrekt sind und ob die Versionsnummer im Profil (`SAPJVM_VERSION`) exakt dem Namen des Verzeichnisses entspricht.
<syntaxhighlight lang="bash" line="1">
# Berechtigungen prüfen (muss daaadm:sapsys sein)
ls -ld /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108

# Falls falsch, als root korrigieren:
chown -R daaadm:sapsys /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108
</syntaxhighlight>

=== Codeschnipsel ===
Entpacken, wenn `SAPCAR` und die `.SAR`-Datei eine Ebene über dem neuen JVM-Verzeichnis liegen:
<syntaxhighlight lang="bash" line="1">
# Als daaadm aus dem neu erstellten sapjvm_8.1.x Ordner heraus ausführen
../SAPCAR -xvf ../SAPJVM8_108-80000202.SAR
</syntaxhighlight>

=== Nützliche Links ===
* [https://me.sap.com/notes/3408212 SAP Note 3408212 - Diagnostics Agent - How To Switch or Update SAP JVM]

DAA Agent

2026-03-06T09:30:11Z

Hendrik: /* Download */

=== Beschreibung ===
Diese Anleitung beschreibt das manuelle Update der SAP JVM für den Diagnostics Agent (DAA) unter Linux. Da der Agent seine JVM nicht automatisch patchen kann, muss das Update manuell über das Anlegen eines neuen Verzeichnisses und die Anpassung des Instanzprofils erfolgen.

=== Download ===
Die aktuelle SAP JVM 8.1 im .SAR-Format muss aus dem SAP Support Portal (Software Center) heruntergeladen werden.
* Dateibeispiel: SAPJVM8_108-80000202.SAR

=== Installation ===
''Die Erstinstallation des Diagnostics Agents ist nicht Teil dieses Artikels. Dieser Artikel behandelt ausschließlich das Update der zugrundeliegenden SAP JVM.''

=== Konfiguration ===
Es ist keine spezielle Vorab-Konfiguration auf Betriebssystemebene nötig, außer dass das Tool SAPCAR verfügbar sein muss und Root-Rechte für die Verzeichniserstellung benötigt werden.

=== Update ===
Das Update erfolgt durch das Anlegen eines neuen JVM-Verzeichnisses und dem Anpassen der Profil-Datei.

<syntaxhighlight lang="bash" line="1">
1. Neues Verzeichnis als root/sudo anlegen und Rechte setzen (Version im Namen anpassen)
mkdir -p /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108
chown daaadm:sapsys /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108

2. Zum DAA-User wechseln und in das neue Verzeichnis gehen
su - daaadm
cd /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108

3. Das heruntergeladene Archiv entpacken (Pfade anpassen)
/usr/sap/DAA/SYS/exe/uc/linuxx86_64/SAPCAR -xvf /Pfad/zur/Datei/SAPJVM8_108-80000202.SAR
</syntaxhighlight>

Im Anschluss muss das Instanzprofil des Diagnostics Agents angepasst werden:

<syntaxhighlight lang="bash" line="1">
Profil editieren (Hostname anpassen)
vi /usr/sap/DAA/SYS/profile/DAA_SMDA98_<Hostname>
</syntaxhighlight>

Dort die Variable SAPJVM_VERSION auf die neue Version ändern:
<syntaxhighlight lang="ini" line="1">
Vorher:
SAPJVM_VERSION = 8.1.097

Nachher:
SAPJVM_VERSION = 8.1.108
</syntaxhighlight>

Den Agenten als daaadm neu starten, damit sapcpe die neuen Binaries kopiert:
<syntaxhighlight lang="bash" line="1">
stopsap
startsap
</syntaxhighlight>

=== Test ===
Nach dem Neustart prüfen, ob die neue JVM aktiv ist (als User daaadm):

<syntaxhighlight lang="bash" line="1">
1. Java-Version im Ausführungsverzeichnis prüfen
/usr/sap/DAA/SMDA98/exe/sapjvm_8/bin/java -version

2. Prüfen, ob der laufende Agent-Prozess den korrekten Java-Pfad nutzt
ps -ef | grep DAA | grep java

3. sapcpe-Log prüfen, ob aus dem neuen Verzeichnis kopiert wurde
grep "source" /usr/sap/DAA/SMDA98/work/sapcpe.log | tail -n 5
</syntaxhighlight>

=== Fehlerbehebung===
====Starten/Stoppen funktioniert nicht====
Wenn startsap oder stopsap mit Fehlern abbrechen oder der Agent nicht hochfährt:

=====Lösung 1=====
Prüfen, ob die Berechtigungen des neu angelegten Verzeichnisses korrekt sind und ob die Versionsnummer im Profil (SAPJVM_VERSION) exakt dem Namen des Verzeichnisses entspricht.
<syntaxhighlight lang="bash" line="1">
Berechtigungen prüfen (muss daaadm:sapsys sein)
ls -ld /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108

Falls falsch, als root korrigieren:
chown -R daaadm:sapsys /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108
</syntaxhighlight>

=== Codeschnipsel ===
Entpacken, wenn SAPCAR und die .SAR-Datei eine Ebene über dem neuen JVM-Verzeichnis liegen:
<syntaxhighlight lang="bash" line="1">
Als daaadm aus dem neu erstellten sapjvm_8.1.x Ordner heraus ausführen
../SAPCAR -xvf ../SAPJVM8_108-80000202.SAR
</syntaxhighlight>

=== Nützliche Links ===
* [https://me.sap.com/notes/3408212 SAP Note 3408212 - Diagnostics Agent - How To Switch or Update SAP JVM]

DAA Agent

2026-03-06T09:29:40Z

Hendrik: /* Beschreibung */

=== Beschreibung ===
Diese Anleitung beschreibt das manuelle Update der SAP JVM für den Diagnostics Agent (DAA) unter Linux. Da der Agent seine JVM nicht automatisch patchen kann, muss das Update manuell über das Anlegen eines neuen Verzeichnisses und die Anpassung des Instanzprofils erfolgen.

=== Download ===
Die aktuelle SAP JVM 8.1 im .SAR-Format muss aus dem SAP Support Portal (Software Center) heruntergeladen werden.
* Dateibeispiel: SAPJVM8_108-80000202.SAR

=== Installation ===
''Die Erstinstallation des Diagnostics Agents ist nicht Teil dieses Artikels. Dieser Artikel behandelt ausschließlich das Update der zugrundeliegenden SAP JVM.''

=== Konfiguration ===
Es ist keine spezielle Vorab-Konfiguration auf Betriebssystemebene nötig, außer dass das Tool SAPCAR verfügbar sein muss und Root-Rechte für die Verzeichniserstellung benötigt werden.

=== Update ===
Das Update erfolgt durch das Anlegen eines neuen JVM-Verzeichnisses und dem Anpassen der Profil-Datei.

<syntaxhighlight lang="bash" line="1">
1. Neues Verzeichnis als root/sudo anlegen und Rechte setzen (Version im Namen anpassen)
mkdir -p /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108
chown daaadm:sapsys /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108

2. Zum DAA-User wechseln und in das neue Verzeichnis gehen
su - daaadm
cd /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108

3. Das heruntergeladene Archiv entpacken (Pfade anpassen)
/usr/sap/DAA/SYS/exe/uc/linuxx86_64/SAPCAR -xvf /Pfad/zur/Datei/SAPJVM8_108-80000202.SAR
</syntaxhighlight>

Im Anschluss muss das Instanzprofil des Diagnostics Agents angepasst werden:

<syntaxhighlight lang="bash" line="1">
Profil editieren (Hostname anpassen)
vi /usr/sap/DAA/SYS/profile/DAA_SMDA98_<Hostname>
</syntaxhighlight>

Dort die Variable SAPJVM_VERSION auf die neue Version ändern:
<syntaxhighlight lang="ini" line="1">
Vorher:
SAPJVM_VERSION = 8.1.097

Nachher:
SAPJVM_VERSION = 8.1.108
</syntaxhighlight>

Den Agenten als daaadm neu starten, damit sapcpe die neuen Binaries kopiert:
<syntaxhighlight lang="bash" line="1">
stopsap
startsap
</syntaxhighlight>

=== Test ===
Nach dem Neustart prüfen, ob die neue JVM aktiv ist (als User daaadm):

<syntaxhighlight lang="bash" line="1">
1. Java-Version im Ausführungsverzeichnis prüfen
/usr/sap/DAA/SMDA98/exe/sapjvm_8/bin/java -version

2. Prüfen, ob der laufende Agent-Prozess den korrekten Java-Pfad nutzt
ps -ef | grep DAA | grep java

3. sapcpe-Log prüfen, ob aus dem neuen Verzeichnis kopiert wurde
grep "source" /usr/sap/DAA/SMDA98/work/sapcpe.log | tail -n 5
</syntaxhighlight>

=== Fehlerbehebung===
====Starten/Stoppen funktioniert nicht====
Wenn startsap oder stopsap mit Fehlern abbrechen oder der Agent nicht hochfährt:

=====Lösung 1=====
Prüfen, ob die Berechtigungen des neu angelegten Verzeichnisses korrekt sind und ob die Versionsnummer im Profil (SAPJVM_VERSION) exakt dem Namen des Verzeichnisses entspricht.
<syntaxhighlight lang="bash" line="1">
Berechtigungen prüfen (muss daaadm:sapsys sein)
ls -ld /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108

Falls falsch, als root korrigieren:
chown -R daaadm:sapsys /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108
</syntaxhighlight>

=== Codeschnipsel ===
Entpacken, wenn SAPCAR und die .SAR-Datei eine Ebene über dem neuen JVM-Verzeichnis liegen:
<syntaxhighlight lang="bash" line="1">
Als daaadm aus dem neu erstellten sapjvm_8.1.x Ordner heraus ausführen
../SAPCAR -xvf ../SAPJVM8_108-80000202.SAR
</syntaxhighlight>

=== Nützliche Links ===
* [https://me.sap.com/notes/3408212 SAP Note 3408212 - Diagnostics Agent - How To Switch or Update SAP JVM]

DAA Agent

2026-03-06T09:29:26Z

Hendrik:

=== Beschreibung ===
Diese Anleitung beschreibt das manuelle Update der SAP JVM für den Diagnostics Agent (DAA) unter Linux. Da der Agent seine JVM nicht automatisch patchen kann, muss das Update manuell über das Anlegen eines neuen Verzeichnisses und die Anpassung des Instanzprofils
erfolgen.

=== Download ===
Die aktuelle SAP JVM 8.1 im .SAR-Format muss aus dem SAP Support Portal (Software Center) heruntergeladen werden.
* Dateibeispiel: SAPJVM8_108-80000202.SAR

=== Installation ===
''Die Erstinstallation des Diagnostics Agents ist nicht Teil dieses Artikels. Dieser Artikel behandelt ausschließlich das Update der zugrundeliegenden SAP JVM.''

=== Konfiguration ===
Es ist keine spezielle Vorab-Konfiguration auf Betriebssystemebene nötig, außer dass das Tool SAPCAR verfügbar sein muss und Root-Rechte für die Verzeichniserstellung benötigt werden.

=== Update ===
Das Update erfolgt durch das Anlegen eines neuen JVM-Verzeichnisses und dem Anpassen der Profil-Datei.

<syntaxhighlight lang="bash" line="1">
1. Neues Verzeichnis als root/sudo anlegen und Rechte setzen (Version im Namen anpassen)
mkdir -p /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108
chown daaadm:sapsys /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108

2. Zum DAA-User wechseln und in das neue Verzeichnis gehen
su - daaadm
cd /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108

3. Das heruntergeladene Archiv entpacken (Pfade anpassen)
/usr/sap/DAA/SYS/exe/uc/linuxx86_64/SAPCAR -xvf /Pfad/zur/Datei/SAPJVM8_108-80000202.SAR
</syntaxhighlight>

Im Anschluss muss das Instanzprofil des Diagnostics Agents angepasst werden:

<syntaxhighlight lang="bash" line="1">
Profil editieren (Hostname anpassen)
vi /usr/sap/DAA/SYS/profile/DAA_SMDA98_<Hostname>
</syntaxhighlight>

Dort die Variable SAPJVM_VERSION auf die neue Version ändern:
<syntaxhighlight lang="ini" line="1">
Vorher:
SAPJVM_VERSION = 8.1.097

Nachher:
SAPJVM_VERSION = 8.1.108
</syntaxhighlight>

Den Agenten als daaadm neu starten, damit sapcpe die neuen Binaries kopiert:
<syntaxhighlight lang="bash" line="1">
stopsap
startsap
</syntaxhighlight>

=== Test ===
Nach dem Neustart prüfen, ob die neue JVM aktiv ist (als User daaadm):

<syntaxhighlight lang="bash" line="1">
1. Java-Version im Ausführungsverzeichnis prüfen
/usr/sap/DAA/SMDA98/exe/sapjvm_8/bin/java -version

2. Prüfen, ob der laufende Agent-Prozess den korrekten Java-Pfad nutzt
ps -ef | grep DAA | grep java

3. sapcpe-Log prüfen, ob aus dem neuen Verzeichnis kopiert wurde
grep "source" /usr/sap/DAA/SMDA98/work/sapcpe.log | tail -n 5
</syntaxhighlight>

=== Fehlerbehebung===
====Starten/Stoppen funktioniert nicht====
Wenn startsap oder stopsap mit Fehlern abbrechen oder der Agent nicht hochfährt:

=====Lösung 1=====
Prüfen, ob die Berechtigungen des neu angelegten Verzeichnisses korrekt sind und ob die Versionsnummer im Profil (SAPJVM_VERSION) exakt dem Namen des Verzeichnisses entspricht.
<syntaxhighlight lang="bash" line="1">
Berechtigungen prüfen (muss daaadm:sapsys sein)
ls -ld /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108

Falls falsch, als root korrigieren:
chown -R daaadm:sapsys /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108
</syntaxhighlight>

=== Codeschnipsel ===
Entpacken, wenn SAPCAR und die .SAR-Datei eine Ebene über dem neuen JVM-Verzeichnis liegen:
<syntaxhighlight lang="bash" line="1">
Als daaadm aus dem neu erstellten sapjvm_8.1.x Ordner heraus ausführen
../SAPCAR -xvf ../SAPJVM8_108-80000202.SAR
</syntaxhighlight>

=== Nützliche Links ===
* [https://me.sap.com/notes/3408212 SAP Note 3408212 - Diagnostics Agent - How To Switch or Update SAP JVM]

DAA Agent

2026-03-06T09:29:00Z

Hendrik:

<pre>
=== Beschreibung ===
Diese Anleitung beschreibt das manuelle Update der SAP JVM für den Diagnostics Agent (DAA) unter Linux. Da der Agent seine JVM nicht automatisch patchen kann, muss das Update manuell über das Anlegen eines neuen Verzeichnisses und die Anpassung des Instanzprofils
erfolgen.

=== Download ===
Die aktuelle SAP JVM 8.1 im .SAR-Format muss aus dem SAP Support Portal (Software Center) heruntergeladen werden.
* Dateibeispiel: SAPJVM8_108-80000202.SAR

=== Installation ===
''Die Erstinstallation des Diagnostics Agents ist nicht Teil dieses Artikels. Dieser Artikel behandelt ausschließlich das Update der zugrundeliegenden SAP JVM.''

=== Konfiguration ===
Es ist keine spezielle Vorab-Konfiguration auf Betriebssystemebene nötig, außer dass das Tool SAPCAR verfügbar sein muss und Root-Rechte für die Verzeichniserstellung benötigt werden.

=== Update ===
Das Update erfolgt durch das Anlegen eines neuen JVM-Verzeichnisses und dem Anpassen der Profil-Datei.

<syntaxhighlight lang="bash" line="1">
1. Neues Verzeichnis als root/sudo anlegen und Rechte setzen (Version im Namen anpassen)
mkdir -p /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108
chown daaadm:sapsys /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108

2. Zum DAA-User wechseln und in das neue Verzeichnis gehen
su - daaadm
cd /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108

3. Das heruntergeladene Archiv entpacken (Pfade anpassen)
/usr/sap/DAA/SYS/exe/uc/linuxx86_64/SAPCAR -xvf /Pfad/zur/Datei/SAPJVM8_108-80000202.SAR
</syntaxhighlight>

Im Anschluss muss das Instanzprofil des Diagnostics Agents angepasst werden:

<syntaxhighlight lang="bash" line="1">
Profil editieren (Hostname anpassen)
vi /usr/sap/DAA/SYS/profile/DAA_SMDA98_<Hostname>
</syntaxhighlight>

Dort die Variable SAPJVM_VERSION auf die neue Version ändern:
<syntaxhighlight lang="ini" line="1">
Vorher:
SAPJVM_VERSION = 8.1.097

Nachher:
SAPJVM_VERSION = 8.1.108
</syntaxhighlight>

Den Agenten als daaadm neu starten, damit sapcpe die neuen Binaries kopiert:
<syntaxhighlight lang="bash" line="1">
stopsap
startsap
</syntaxhighlight>

=== Test ===
Nach dem Neustart prüfen, ob die neue JVM aktiv ist (als User daaadm):

<syntaxhighlight lang="bash" line="1">
1. Java-Version im Ausführungsverzeichnis prüfen
/usr/sap/DAA/SMDA98/exe/sapjvm_8/bin/java -version

2. Prüfen, ob der laufende Agent-Prozess den korrekten Java-Pfad nutzt
ps -ef | grep DAA | grep java

3. sapcpe-Log prüfen, ob aus dem neuen Verzeichnis kopiert wurde
grep "source" /usr/sap/DAA/SMDA98/work/sapcpe.log | tail -n 5
</syntaxhighlight>

=== Fehlerbehebung===
====Starten/Stoppen funktioniert nicht====
Wenn startsap oder stopsap mit Fehlern abbrechen oder der Agent nicht hochfährt:

=====Lösung 1=====
Prüfen, ob die Berechtigungen des neu angelegten Verzeichnisses korrekt sind und ob die Versionsnummer im Profil (SAPJVM_VERSION) exakt dem Namen des Verzeichnisses entspricht.
<syntaxhighlight lang="bash" line="1">
Berechtigungen prüfen (muss daaadm:sapsys sein)
ls -ld /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108

Falls falsch, als root korrigieren:
chown -R daaadm:sapsys /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108
</syntaxhighlight>

=== Codeschnipsel ===
Entpacken, wenn SAPCAR und die .SAR-Datei eine Ebene über dem neuen JVM-Verzeichnis liegen:
<syntaxhighlight lang="bash" line="1">
Als daaadm aus dem neu erstellten sapjvm_8.1.x Ordner heraus ausführen
../SAPCAR -xvf ../SAPJVM8_108-80000202.SAR
</syntaxhighlight>

=== Nützliche Links ===
* [https://me.sap.com/notes/3408212 SAP Note 3408212 - Diagnostics Agent - How To Switch or Update SAP JVM]
</pre>

DAA Agent

2026-03-06T09:27:45Z

Hendrik: Die Seite wurde neu angelegt: „ <pre> = Update der SAP JVM für den SAP Diagnostics Agent (DAA) = Diese Anleitung beschreibt das manuelle Update der SAP JVM für den Diagnostics Agent (DAA) unter Linux. Da der Agent seine JVM nicht automatisch patchen kann, muss das Update manuell über das Anlegen eines neuen Verzeichnisses und die Anpassung des Instanzprofils erfolgen (gemäß SAP Hinweis 3408212). == Voraussetzungen == * Die aktuelle SAP JVM 8.1 im .SAR-Format wurde…“

<pre>
= Update der SAP JVM für den SAP Diagnostics Agent (DAA) =

Diese Anleitung beschreibt das manuelle Update der SAP JVM für den Diagnostics Agent (DAA) unter Linux. Da der Agent seine JVM nicht automatisch patchen kann, muss das Update manuell über das Anlegen eines neuen Verzeichnisses und die Anpassung des Instanzprofils erfolgen
(gemäß SAP Hinweis 3408212).

== Voraussetzungen ==
* Die aktuelle SAP JVM 8.1 im .SAR-Format wurde aus dem SAP Support Portal heruntergeladen (z. B. <code>SAPJVM8_108-80000202.SAR</code>).
* Das Tool <code>SAPCAR</code> ist auf dem Server verfügbar.
* Root-Rechte (bzw. <code>sudo</code>) zum Anlegen der Berechtigungen sowie Zugriff auf den User <code>daaadm</code>.

== 1. Neues Verzeichnis anlegen ==
Zuerst muss ein neues Verzeichnis für die Ziel-Version der JVM erstellt und dem DAA-Admin zugewiesen werden.

Melde dich als <code>root</code> (oder mit <code>sudo</code>) an:
<pre>
mkdir -p /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108
chown daaadm:sapsys /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108
</pre>

== 2. SAP JVM entpacken ==
Das Entpacken erfolgt als User <code>daaadm</code> direkt in das neu erstellte Verzeichnis.

Wechsle zum User <code>daaadm</code>:
<pre>
su - daaadm
cd /usr/sap/DAA/SYS/exe/jvm/linuxx86_64/sapjvm_8.1.108
</pre>

Entpacke das Archiv mit <code>SAPCAR</code> (Pfade ggf. an den Ablageort der Datei anpassen):
<pre>
/usr/sap/DAA/SYS/exe/uc/linuxx86_64/SAPCAR -xvf /Pfad/zur/Datei/SAPJVM8_108-80000202.SAR
</pre>
''Hinweis: Falls SAPCAR und die .SAR-Datei eine Ebene darüber liegen, kann auch <code>../SAPCAR -xvf ../SAPJVM8_108-80000202.SAR</code> genutzt werden.''

== 3. Instanzprofil anpassen ==
Damit das Programm <code>sapcpe</code> beim Starten des Agenten die neuen Binaries in das Ausführungsverzeichnis kopiert, muss die Version im Profil aktualisiert werden.

Öffne das Profil mit einem Texteditor (z. B. <code>vi</code>):
<pre>
vi /usr/sap/DAA/SYS/profile/DAA_SMDA98_<Hostname>
</pre>

Suche nach der Variable <code>SAPJVM_VERSION</code> und ändere den Wert auf die neue Version (hier im Beispiel von <code>8.1.097</code> auf <code>8.1.108</code>):
<pre>
Vorher:
SAPJVM_VERSION = 8.1.097

Nachher:
SAPJVM_VERSION = 8.1.108
</pre>
Speichern und schließen.

== 4. Diagnostics Agent neustarten ==
Damit die Änderungen aktiv und die neuen Dateien per <code>sapcpe</code> in das Ausführungsverzeichnis (<code>/usr/sap/DAA/SMDA98/exe/</code>) kopiert werden, muss der Agent neu gestartet werden.

Als User <code>daaadm</code> ausführen:
<pre>
stopsap
startsap
</pre>

== 5. Überprüfung ==
Nach dem Neustart kann mit folgenden Befehlen (als <code>daaadm</code>) verifiziert werden, ob die neue JVM aktiv ist:

'''1. Java-Version im Ausführungsverzeichnis prüfen:'''
<pre>
/usr/sap/DAA/SMDA98/exe/sapjvm_8/bin/java -version
</pre>
''Erwartete Ausgabe: Die neue Version (z.B. 8.1.108) wird angezeigt.''

'''2. Laufenden Prozess überprüfen:'''
<pre>
ps -ef | grep DAA | grep java
</pre>
''Der Pfad zum laufenden Java-Prozess sollte auf <code>/usr/sap/DAA/SMDA98/exe/sapjvm_8/bin/java</code> zeigen.''

'''3. Copy-Log (sapcpe) kontrollieren:'''
Um sicherzugehen, dass <code>sapcpe</code> das korrekte Quellverzeichnis genutzt hat:
<pre>
grep "source" /usr/sap/DAA/SMDA98/work/sapcpe.log | tail -n 5
</pre>
''Hier sollte das neu angelegte Verzeichnis <code>sapjvm_8.1.108</code> als Quelle (Source) gelistet sein.''
</pre>

Hauptseite

2026-03-06T09:27:26Z

Hendrik:

In diesem Wiki stelle ich meine selbst erstellten Anleitungen, Workitems und Codeschnipsel für meinen Alltag für alle zur Verfügung.

==Inhaltsverzeichnis==

#Virtualisierung
##[[Qemu/KVM|Qemu/KVM mit Libvirt]]
##[[Proxmox]]
###[[Proxmox-GPU-Passthough|GPU-Passthough]]
##[[Hyper-V|Windows Hyper-V]]
##[[Docker]]
#Betriebssysteme
##[[Linux]]
###[[Wireguard einrichten]]
###[[FTP Server]]
###[[Exim4]]
###[[Linux-Befehle|Befehle]]
###[[Netzwerktest]]
###[[NFS Share einrichten]]
###[[Postfix]]
##[[Windows]]
#Monitoring
##[[Icinga]]
##Focused Run
# [[SAP]]
## [[ABAP]]
### [[ZBV]]
## [[SAP Router]]
## [[SAP Hana Cockpit]]
## [[SAP Hostagent]]
## [[SAP Cloud Connector]]
## [[SAP Import All einplanen|Import All einplanen]]
## [[Archivserver]]
### [[SAPERION]]
## [[DAA Agent]]
##Datenbanken
###[[Oracle]]
###[[SAP MaxDB]]
###[[SAP HANA]]
####[[SAP HANA Zero Downtime Patching]]
###[[SAP ASE (Sybase)]]
###[[IBM DB2]]
## [[Adobe Document Server]]
#Smart-Home
##[[OpenHAB]]
#Netzwerk
##[https://schroederdennis.de/tutorial-howto/homelab-trotz-ipv6-cgnat-wireguard-nat-dnat-snat-1euro-vserver-dg-ipv4/ IPv4 mit VPS durch IPv6 tunneln]
##Firewall
###[[OPNSense]]
####[[VPN Gateway]]
#Spiele
##[[WoW]]
###[[AzerothCore]]
#AI / KI
##[[llama.cpp]]
##[[VLLm]]
##[[Ollama]]
## [https://www.reddit.com/r/ROCm/comments/1o99swp/rocm_70_install_for_mi50_32gb_ubuntu_2404_lts/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1re&utm_medium=web3x&utm_name=web3xcss&utm_term=1|Instinct Mi 50 mit ROCm 7]
###[[Modell erstellen]]
#Sonstiges
##[[UrBackup]]
##[[Passwortregeln]]
#[[Template für neue Seiten]]
#[[Codeschnipsel ohne eigene Seite]]
#[[Test]]

Hilfe zur Verwendung und Konfiguration der Wiki-Software findest du im [[mediawikiwiki:Special:MyLanguage/Help:Contents|Benutzerhandbuch]].

==Starthilfen==

* [[mediawikiwiki:Special:MyLanguage/Manual:Configuration_settings|Liste der Konfigurationsparameter]]
*[[mediawikiwiki:Special:MyLanguage/Manual:FAQ|Häufige Fragen zu MediaWiki]]
*[https://lists.wikimedia.org/postorius/lists/mediawiki-announce.lists.wikimedia.org/ Mailingliste zu neuen Versionen von MediaWiki]
*[[mediawikiwiki:Special:MyLanguage/Localisation#Translation_resources|Übersetze MediaWiki für deine Sprache]]
*[[mediawikiwiki:Special:MyLanguage/Manual:Combating_spam|Erfahre, wie du Spam auf deinem Wiki bekämpfen kannst]]

SAP

2026-02-19T16:16:26Z

Hendrik: /* Kernelupdate */

===Beschreibung===
SAP ist ein deutsches Software-Unternehmen, das sich auf die Entwicklung und den Verkauf von Unternehmenssoftware spezialisiert hat. Die Software-Lösungen von SAP decken eine Vielzahl von Geschäftsbereichen ab, wie z.B. Finanzwesen, Personalwesen, Einkauf und Vertrieb. Die Produkte von SAP werden von Unternehmen aller Größenordnungen genutzt, um Geschäftsprozesse zu optimieren, Daten zu verwalten und bessere Entscheidungen zu treffen. Zu den bekanntesten SAP-Produkten gehören SAP ERP, SAP HANA und SAP S/4HANA.
===SWPM starten===
<syntaxhighlight lang="bash">
./sapinst SAPINST_STACK_XML=stack.xml SAPINST_USE_HOSTNAME=
</syntaxhighlight>
===SUM starten===

====Linux====
=====ABAP=====
<syntaxhighlight lang="bash">
cd /usr/sap/$SID/SUM
rm -R ./SUM
SAPCAR -xvf ./SUM11SP05_0-80006800.SAR
cd SUM/abap/
sudo ./SUMSTART confighostagent $SAPSYSTEMNAME
</syntaxhighlight>
=====JAVA=====
<syntaxhighlight lang="bash">
cd /usr/sap/$SID/SUM
rm -R ./SUM
SAPCAR -xvf ./SUM*.SAR
cd SUM
sudo ./STARTUP confighostagent $SAPSYSTEMNAME
</syntaxhighlight>
====Windows====
<syntaxhighlight lang="bash">
STARTUP.BAT confighostagent <SID>
</syntaxhighlight>

===SAP Host Agent===
====Version überprüfen====
[https://me.sap.com/notes/0002032385 Note 0002032385]

Windows:
<syntaxhighlight lang="bash">
%Program Files%\SAP\hostctrl\exe\saphostexec.exe -version
</syntaxhighlight>
Linux:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostexec -version
</syntaxhighlight>

====Aktualisieren====
SAR Archiv herunterladen und nach /usr/sap/Z36/SUM kopieren
als root am System anmelden
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostexec -upgrade -archive /usr/sap/<SID>/SUM/<ARCHIV>.SAR
</syntaxhighlight>
===Kernelupdate===
''Siehe auch Workitem 2L-DB-SAP-00026: SAP Kerneltausch''

Aktuellen Kernel herausfinden
<syntaxhighlight lang="bash">
disp+work -v | grep "patch number"
</syntaxhighlight>
Kernel Backup erstellen und alte Backups löschen
<syntaxhighlight lang="bash">
cd /sapmnt/*/exe/uc/
rm -R linuxx86_64_*
cp -R linuxx86_64 linuxx86_64_<DATUM>
</syntaxhighlight>
[https://me.sap.com/softwarecenter/support/index SAP Downloadportal] aufrufen und nach "SAP KERNEL 7.85 64-BIT UNICODE" suchen
Die Archive herunterladen
Nach /sapmnt/SID/exe/uc/linuxx86_64/ kopieren
Ggf. noch Dateiberechtigungen korrigieren
<syntaxhighlight lang="bash">
chown -R <SIDADM>:sapsys *.SAR && chown -R <SIDADM>:sapsys *.sar
</syntaxhighlight>
In der richtigen Reihenfolge entpacken:
# SAPEXEDB.SAR
# SAPEXE.SAR
# dw_utils
# dw
# R3Trans
# tp
<syntaxhighlight lang="bash">
cd /sapmnt/*/exe/uc/linuxx86_64/
SAPCAR -xvf SAPEXEDB_*.SAR
SAPCAR -xvf SAPEXE_*.SAR
SAPCAR -xvf R3trans_*.SAR
SAPCAR -xvf tp_*.sar
SAPCAR -xvf sapftp_*.sar
SAPCAR -xvf dw_*-*.sar
SAPCAR -xvf dw_utils*.sar
SAPCAR -xvf lib_dbsl*.sar
SAPCAR -xvf sapwebgui*.sar
SAPCAR -xvf saphttp_*.sar
SAPCAR -xvf abap2vcs_*.sar
SAPCAR -xvf SYBCTRL*.SAR
SAPCAR -xvf sapnwrfc_*.sar
SAPCAR -xvf ENSA2_*.SAR
SAPCAR -xvf saplicense_*.sar
SAPCAR -xvf enserver_*.sar
</syntaxhighlight>
Aufräumen und alte SAR Archive löschen:
<syntaxhighlight lang="bash">
cd /sapmnt/*/exe/uc/linuxx86_64/ && rm *.SAR && rm *.sar
</syntaxhighlight>
Oder in Kurzform:
<syntaxhighlight lang="bash">
ls -r | grep -i '.*[0-9].*\.sar$' | xargs -I {} SAPCAR -xvf {}
</syntaxhighlight>Dies entpackt alle sar Archive, die eine Versionsnummer im Namen enthalten.

''Hinweis: vorher Kontrollieren ob mit der Ausgabe ls -r tatsächlich SAPEXEDB '''vor''' SAPEXE ausgegeben wird.''

SAP neustarten:
<syntaxhighlight lang="bash">
sapcontrol -nr 00 -function RestartSystem && watch -n 1 sapcontrol -nr 00 -function GetProcessList
</syntaxhighlight>

====Oracle Nacharbeiten====
Falls als DB die Oracle DB verwendet wird, müssen noch Berechtigungen nachgezogen werden, da es sonst zu Problemen mit BRTOOLS kommen kann.
Als root unter /sapmnt/<SID>/exe/nuc/linuxx86_64:
<syntaxhighlight lang="bash">
chmod -R 4775 brrestore brspace brrecover brconnect brbackup brarchive
chown -R oracle:oinstall brrestore brspace brrecover brconnect brbackup
</syntaxhighlight>

===Memory Parameter prüfen===
<syntaxhighlight lang="bash">
cdpro
sappfpar check pf=<profile_file>
</syntaxhighlight>

===JAVA===
====Sysinfo.xml generieren====
https://me.sap.com/notes/2293050
====Java patchen====
Patches unter /usr/sap/<SID>/SUM/ ablegen

Als <SID>adm anmelden
<syntaxhighlight lang="bash">
telnet localhost 50108
Administrator
<PW>
add deploy
deploy /usr/sap/<SID>/SUM/ version_rule=all
</syntaxhighlight>
Eventuell SAP stoppen und wieder starten

<syntaxhighlight lang="bash">
dura9708:djpadm 73> ls -ltr
total 169596
-rw-r--r-- 1 djpadm sapsys 18675603 Apr 28 10:29 MESSAGING24P_21-80000682.SCA
-rw-r--r-- 1 djpadm sapsys 25564663 Apr 28 10:29 J2EEFRMW24P_4-80000439.SCA
-rw-r--r-- 1 djpadm sapsys 14020864 Apr 28 10:29 ENGINEAPI24P_4-80000618.SCA
-rw-r--r-- 1 djpadm sapsys 115196425 Apr 28 10:29 SERVERCORE24P_15-80000485.SCA
</syntaxhighlight>

===Codeschnipsel===

===Nützliche Links===

SAP

2026-02-19T16:12:43Z

Hendrik: /* Kernelupdate */

===Beschreibung===
SAP ist ein deutsches Software-Unternehmen, das sich auf die Entwicklung und den Verkauf von Unternehmenssoftware spezialisiert hat. Die Software-Lösungen von SAP decken eine Vielzahl von Geschäftsbereichen ab, wie z.B. Finanzwesen, Personalwesen, Einkauf und Vertrieb. Die Produkte von SAP werden von Unternehmen aller Größenordnungen genutzt, um Geschäftsprozesse zu optimieren, Daten zu verwalten und bessere Entscheidungen zu treffen. Zu den bekanntesten SAP-Produkten gehören SAP ERP, SAP HANA und SAP S/4HANA.
===SWPM starten===
<syntaxhighlight lang="bash">
./sapinst SAPINST_STACK_XML=stack.xml SAPINST_USE_HOSTNAME=
</syntaxhighlight>
===SUM starten===

====Linux====
=====ABAP=====
<syntaxhighlight lang="bash">
cd /usr/sap/$SID/SUM
rm -R ./SUM
SAPCAR -xvf ./SUM11SP05_0-80006800.SAR
cd SUM/abap/
sudo ./SUMSTART confighostagent $SAPSYSTEMNAME
</syntaxhighlight>
=====JAVA=====
<syntaxhighlight lang="bash">
cd /usr/sap/$SID/SUM
rm -R ./SUM
SAPCAR -xvf ./SUM*.SAR
cd SUM
sudo ./STARTUP confighostagent $SAPSYSTEMNAME
</syntaxhighlight>
====Windows====
<syntaxhighlight lang="bash">
STARTUP.BAT confighostagent <SID>
</syntaxhighlight>

===SAP Host Agent===
====Version überprüfen====
[https://me.sap.com/notes/0002032385 Note 0002032385]

Windows:
<syntaxhighlight lang="bash">
%Program Files%\SAP\hostctrl\exe\saphostexec.exe -version
</syntaxhighlight>
Linux:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostexec -version
</syntaxhighlight>

====Aktualisieren====
SAR Archiv herunterladen und nach /usr/sap/Z36/SUM kopieren
als root am System anmelden
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostexec -upgrade -archive /usr/sap/<SID>/SUM/<ARCHIV>.SAR
</syntaxhighlight>
===Kernelupdate===
''Siehe auch Workitem 2L-DB-SAP-00026: SAP Kerneltausch''

Aktuellen Kernel herausfinden
<syntaxhighlight lang="bash">
disp+work -v | grep "patch number"
</syntaxhighlight>
Kernel Backup erstellen und alte Backups löschen
<syntaxhighlight lang="bash">
cd /sapmnt/*/exe/uc/
rm -R linuxx86_64_*
cp -R linuxx86_64 linuxx86_64_<DATUM>
</syntaxhighlight>
[https://me.sap.com/softwarecenter/support/index SAP Downloadportal] aufrufen und nach "SAP KERNEL 7.85 64-BIT UNICODE" suchen
Die Archive herunterladen
Nach /sapmnt/SID/exe/uc/linuxx86_64/ kopieren
Ggf. noch Dateiberechtigungen korrigieren
<syntaxhighlight lang="bash">
chown -R <SIDADM>:sapsys *.SAR && chown -R <SIDADM>:sapsys *.sar
</syntaxhighlight>
In der richtigen Reihenfolge entpacken:
# SAPEXEDB.SAR
# SAPEXE.SAR
# dw_utils
# dw
# R3Trans
# tp
<syntaxhighlight lang="bash">
cd /sapmnt/*/exe/uc/linuxx86_64/
SAPCAR -xvf SAPEXEDB_*.SAR
SAPCAR -xvf SAPEXE_*.SAR
SAPCAR -xvf R3trans_*.SAR
SAPCAR -xvf tp_*.sar
SAPCAR -xvf sapftp_*.sar
SAPCAR -xvf dw_*-*.sar
SAPCAR -xvf dw_utils*.sar
SAPCAR -xvf lib_dbsl*.sar
SAPCAR -xvf sapwebgui*.sar
SAPCAR -xvf saphttp_*.sar
SAPCAR -xvf abap2vcs_*.sar
SAPCAR -xvf SYBCTRL*.SAR
</syntaxhighlight>
Aufräumen und alte SAR Archive löschen:
<syntaxhighlight lang="bash">
cd /sapmnt/*/exe/uc/linuxx86_64/ && rm *.SAR && rm *.sar
</syntaxhighlight>
Oder in Kurzform:
<syntaxhighlight lang="bash">
ls -r | grep -i '.*[0-9].*\.sar$' | xargs -I {} SAPCAR -xvf {}
</syntaxhighlight>Dies entpackt alle sar Archive, die eine Versionsnummer im Namen enthalten.

''Hinweis: vorher Kontrollieren ob mit der Ausgabe ls -r tatsächlich SAPEXEDB '''vor''' SAPEXE ausgegeben wird.''

SAP neustarten:
<syntaxhighlight lang="bash">
sapcontrol -nr 00 -function RestartSystem && watch -n 1 sapcontrol -nr 00 -function GetProcessList
</syntaxhighlight>

====Oracle Nacharbeiten====
Falls als DB die Oracle DB verwendet wird, müssen noch Berechtigungen nachgezogen werden, da es sonst zu Problemen mit BRTOOLS kommen kann.
Als root unter /sapmnt/<SID>/exe/nuc/linuxx86_64:
<syntaxhighlight lang="bash">
chmod -R 4775 brrestore brspace brrecover brconnect brbackup brarchive
chown -R oracle:oinstall brrestore brspace brrecover brconnect brbackup
</syntaxhighlight>

===Memory Parameter prüfen===
<syntaxhighlight lang="bash">
cdpro
sappfpar check pf=<profile_file>
</syntaxhighlight>

===JAVA===
====Sysinfo.xml generieren====
https://me.sap.com/notes/2293050
====Java patchen====
Patches unter /usr/sap/<SID>/SUM/ ablegen

Als <SID>adm anmelden
<syntaxhighlight lang="bash">
telnet localhost 50108
Administrator
<PW>
add deploy
deploy /usr/sap/<SID>/SUM/ version_rule=all
</syntaxhighlight>
Eventuell SAP stoppen und wieder starten

<syntaxhighlight lang="bash">
dura9708:djpadm 73> ls -ltr
total 169596
-rw-r--r-- 1 djpadm sapsys 18675603 Apr 28 10:29 MESSAGING24P_21-80000682.SCA
-rw-r--r-- 1 djpadm sapsys 25564663 Apr 28 10:29 J2EEFRMW24P_4-80000439.SCA
-rw-r--r-- 1 djpadm sapsys 14020864 Apr 28 10:29 ENGINEAPI24P_4-80000618.SCA
-rw-r--r-- 1 djpadm sapsys 115196425 Apr 28 10:29 SERVERCORE24P_15-80000485.SCA
</syntaxhighlight>

===Codeschnipsel===

===Nützliche Links===

SAP

2026-02-19T16:11:37Z

Hendrik: /* Kernelupdate */

===Beschreibung===
SAP ist ein deutsches Software-Unternehmen, das sich auf die Entwicklung und den Verkauf von Unternehmenssoftware spezialisiert hat. Die Software-Lösungen von SAP decken eine Vielzahl von Geschäftsbereichen ab, wie z.B. Finanzwesen, Personalwesen, Einkauf und Vertrieb. Die Produkte von SAP werden von Unternehmen aller Größenordnungen genutzt, um Geschäftsprozesse zu optimieren, Daten zu verwalten und bessere Entscheidungen zu treffen. Zu den bekanntesten SAP-Produkten gehören SAP ERP, SAP HANA und SAP S/4HANA.
===SWPM starten===
<syntaxhighlight lang="bash">
./sapinst SAPINST_STACK_XML=stack.xml SAPINST_USE_HOSTNAME=
</syntaxhighlight>
===SUM starten===

====Linux====
=====ABAP=====
<syntaxhighlight lang="bash">
cd /usr/sap/$SID/SUM
rm -R ./SUM
SAPCAR -xvf ./SUM11SP05_0-80006800.SAR
cd SUM/abap/
sudo ./SUMSTART confighostagent $SAPSYSTEMNAME
</syntaxhighlight>
=====JAVA=====
<syntaxhighlight lang="bash">
cd /usr/sap/$SID/SUM
rm -R ./SUM
SAPCAR -xvf ./SUM*.SAR
cd SUM
sudo ./STARTUP confighostagent $SAPSYSTEMNAME
</syntaxhighlight>
====Windows====
<syntaxhighlight lang="bash">
STARTUP.BAT confighostagent <SID>
</syntaxhighlight>

===SAP Host Agent===
====Version überprüfen====
[https://me.sap.com/notes/0002032385 Note 0002032385]

Windows:
<syntaxhighlight lang="bash">
%Program Files%\SAP\hostctrl\exe\saphostexec.exe -version
</syntaxhighlight>
Linux:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostexec -version
</syntaxhighlight>

====Aktualisieren====
SAR Archiv herunterladen und nach /usr/sap/Z36/SUM kopieren
als root am System anmelden
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostexec -upgrade -archive /usr/sap/<SID>/SUM/<ARCHIV>.SAR
</syntaxhighlight>
===Kernelupdate===
''Siehe auch Workitem 2L-DB-SAP-00026: SAP Kerneltausch''

Aktuellen Kernel herausfinden
<syntaxhighlight lang="bash">
disp+work -v | grep "patch number"
</syntaxhighlight>
Kernel Backup erstellen und alte Backups löschen
<syntaxhighlight lang="bash">
cd /sapmnt/*/exe/uc/
rm -R linuxx86_64_*
cp -R linuxx86_64 linuxx86_64_<DATUM>
</syntaxhighlight>
[https://me.sap.com/softwarecenter/support/index SAP Downloadportal] aufrufen und nach "SAP KERNEL 7.85 64-BIT UNICODE" suchen
Die Archive herunterladen
Nach /sapmnt/SID/exe/uc/linuxx86_64/ kopieren
Ggf. noch Dateiberechtigungen korrigieren
<syntaxhighlight lang="bash">
chown -R <SIDADM>:sapsys *.SAR && chown -R <SIDADM>:sapsys *.sar
</syntaxhighlight>
In der richtigen Reihenfolge entpacken:
# SAPEXEDB.SAR
# SAPEXE.SAR
# dw_utils
# dw
# R3Trans
# tp
<syntaxhighlight lang="bash">
cd /sapmnt/*/exe/uc/linuxx86_64/
SAPCAR -xvf SAPEXEDB_*.SAR
SAPCAR -xvf SAPEXE_*.SAR
SAPCAR -xvf R3trans_*.SAR
SAPCAR -xvf tp_*.sar
SAPCAR -xvf sapftp_*.sar
SAPCAR -xvf dw_*-*.sar
SAPCAR -xvf dw_utils*.sar
SAPCAR -xvf lib_dbsl*.sar
SAPCAR -xvf sapwebgui*.sar
SAPCAR -xvf saphttp_*.sar
SAPCAR -xvf abap2vcs_*.sar
SAPCAR -xvf SYBCTRL*.SAR
</syntaxhighlight>
Aufräumen und alte SAR Archive löschen:
<syntaxhighlight lang="bash">
rm /sapmnt/*/exe/uc/linuxx86_64/*.SAR && rm /sapmnt/*/exe/uc/linuxx86_64/*.sar
</syntaxhighlight>
Oder in Kurzform:
<syntaxhighlight lang="bash">
ls -r | grep -i '.*[0-9].*\.sar$' | xargs -I {} SAPCAR -xvf {}
</syntaxhighlight>Dies entpackt alle sar Archive, die eine Versionsnummer im Namen enthalten.

''Hinweis: vorher Kontrollieren ob mit der Ausgabe ls -r tatsächlich SAPEXEDB '''vor''' SAPEXE ausgegeben wird.''

SAP neustarten:
<syntaxhighlight lang="bash">
sapcontrol -nr 00 -function RestartSystem && watch -n 1 sapcontrol -nr 00 -function GetProcessList
</syntaxhighlight>

====Oracle Nacharbeiten====
Falls als DB die Oracle DB verwendet wird, müssen noch Berechtigungen nachgezogen werden, da es sonst zu Problemen mit BRTOOLS kommen kann.
Als root unter /sapmnt/<SID>/exe/nuc/linuxx86_64:
<syntaxhighlight lang="bash">
chmod -R 4775 brrestore brspace brrecover brconnect brbackup brarchive
chown -R oracle:oinstall brrestore brspace brrecover brconnect brbackup
</syntaxhighlight>

===Memory Parameter prüfen===
<syntaxhighlight lang="bash">
cdpro
sappfpar check pf=<profile_file>
</syntaxhighlight>

===JAVA===
====Sysinfo.xml generieren====
https://me.sap.com/notes/2293050
====Java patchen====
Patches unter /usr/sap/<SID>/SUM/ ablegen

Als <SID>adm anmelden
<syntaxhighlight lang="bash">
telnet localhost 50108
Administrator
<PW>
add deploy
deploy /usr/sap/<SID>/SUM/ version_rule=all
</syntaxhighlight>
Eventuell SAP stoppen und wieder starten

<syntaxhighlight lang="bash">
dura9708:djpadm 73> ls -ltr
total 169596
-rw-r--r-- 1 djpadm sapsys 18675603 Apr 28 10:29 MESSAGING24P_21-80000682.SCA
-rw-r--r-- 1 djpadm sapsys 25564663 Apr 28 10:29 J2EEFRMW24P_4-80000439.SCA
-rw-r--r-- 1 djpadm sapsys 14020864 Apr 28 10:29 ENGINEAPI24P_4-80000618.SCA
-rw-r--r-- 1 djpadm sapsys 115196425 Apr 28 10:29 SERVERCORE24P_15-80000485.SCA
</syntaxhighlight>

===Codeschnipsel===

===Nützliche Links===

SAP HANA

2026-02-19T13:17:42Z

Hendrik: /* Passwort Hash eines einzelnen Users kopieren */

=== Beschreibung ===
SAP HANA ist eine In-Memory-Datenbank-Plattform, die von SAP entwickelt wurde. Sie ermöglicht es Unternehmen große Datenmengen in Echtzeit zu verarbeiten und zu analysieren. Die Plattform unterstützt sowohl transaktionale als auch analytische Anwendungen und ermöglicht es den Benutzern auf umfassende Echtzeit-Analysen zuzugreifen.

Die In-Memory-Technologie von SAP HANA sorgt dafür, dass Daten in einem schnellen Zugriffsspeicher gespeichert werden, anstatt auf langsameren Festplatten. Dies führt zu schnelleren Datenzugriffszeiten und ermöglicht es Unternehmen Echtzeit-Transaktionen und Analysen durchzuführen.

SAP HANA bietet auch eine Vielzahl von Tools und Anwendungen um Unternehmen bei der Verwaltung von Daten zu unterstützen. Dazu gehören Tools zur Datenintegration, Datenbereinigung, Datenmodellierung und Analyse. Es unterstützt auch eine Vielzahl von Programmiersprachen und Entwicklungsplattformen, einschließlich SAPUI5, Java, Python uvm.
=== Download ===
[https://me.sap.com/softwarecenter/support/index SAP Downloadportal]

<nowiki>https://launchpad.support.sap.com/#/softwarecenter</nowiki> SUPPORT PACKAGES & PATCHES By Category SAP IN-MEMORY (SAP HANA ) HANA PLATFORM EDITION SAP HANA PLATFORM EDITION SAP HANA PLATFORM EDITION 2.0 SAP HANA CLIENT 2.0

=== Installation ===
hdblcm nach installieren:
https://me.sap.com/notes/2651885<syntaxhighlight lang="bash">
<install_media>/SAP_HANA_DATABASE/hdblcm --sid=<SID> --action=update --components=hdblcm --ignore=check_signature_file --verify_signature=off
</syntaxhighlight>

=== Konfiguration===

===SAP HANA Update===
Aktuelle Version als <SID>ADM mit "HDB version" anzeigen lassen

IMDB_CLIENT<VERSION> und IMDB_SERVER<VERSION> von SAP herunterladen und nach /usr/sap/SUM kopieren
<syntaxhighlight lang="bash">
sudo su - root
cd /usr/sap/SUM
chmod +x SAPCAR && ./SAPCAR -xvf IMDB_CLIENT*
chmod +x SAPCAR && ./SAPCAR -xvf IMDB_SERVER*
cd /hana/shared/*/hdblcm
./hdblcm --action=update --ignore=check_signature_file --verify_signature=off --lss_trust_unsigned_components
Enter directory root to search for components: /usr/sap/SUM/
</syntaxhighlight>
Mit "1" alles updaten und weiter den Anweisungen folgen.

===SAP HANA Sizing Report===
Quelle: [https://blogs.sap.com/2021/05/06/how-to-install-run-the-abap-on-hana-sizing-report-sap-note-1872170-a-step-by-step-guide/ HANA Sizing Report]
# Note updaten. Prüfen, ob 2810633 einbaubar ist, ansonsten 2504480 einbauen. Wenn beide nicht einbaubar sind ist das System zu alt oder bereits aktuell.
# SA38 aufrufen und /SDF/HDB_SIZING ausführen
# "Number of parallel dialog processes" auf 3 setzen
# Ausführen oder als Hintergrundjob einplanen

=== Kennwort ändern ===
<syntaxhighlight lang="bash">
ALTER USER <USER> PASSWORD "<new_password>";
</syntaxhighlight>
Oder direkt mit dem <SID>adm in der Shell:
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER <USER> PASSWORD <PW>"
</syntaxhighlight>
Hinweise:
* für System@TENANT und SYSTEM@SYSTEMDB muss der HDBUSERSTORE für das Backupscript aktualisiert werden und für die Anmeldung im HDB Studio

Lösungsschritte:

Prüfen, welche Keys angelegt sind:

<syntaxhighlight lang="bash">
hdbuserstore list
</syntaxhighlight>
Alten Key löschen (Beispiel: BACKUPKEY für SystemDB):

<syntaxhighlight lang="bash">
hdbuserstore delete BACKUP
</syntaxhighlight>
Neuen Key anlegen und dabei das aktualisierte SYSTEM‑Passwort verwenden:

<syntaxhighlight lang="bash">
hdbuserstore set BACKUP <FQDN>:30013 SYSTEM "<PW>"
</syntaxhighlight>

====SAPABAP1 Kennwort ändern====

Direkt auf Applikationsserver:
<syntaxhighlight lang="bash">
hdbsql -n <FQDN>:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPABAP1 PASSWORD <PW>"
hdbuserstore set default <FQDN>:30013@<TENANT> SAPABAP1 "<PW>"
</syntaxhighlight>
Oder im HDB Studio:
<syntaxhighlight lang="bash">
ALTER USER SAPABAP1 PASSWORD "<PW>";
</syntaxhighlight>
[https://me.sap.com/notes/0002750829 2750829 - Ändern des SAPABAP1-Schemakennworts in SAP HANA]
Anschließend siehe "SAP HANA HDB Userstore neu setzen"

==== SAPDBCTRL (SAP Host Agent) ====
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d SYSTEMDB -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
</syntaxhighlight>
Dann als root fortfahren.
Für SystemDB, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname SYSTEMDB@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>
Für Tenant, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname IMP@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>

=== Encryption root keys backup password setzen ===
<syntaxhighlight lang="bash">
ALTER SYSTEM SET ENCRYPTION ROOT KEYS BACKUP PASSWORD "<PW>";
</syntaxhighlight>
===SAP HANA HDB Userstore neu setzen===
<syntaxhighlight lang="bash">
hdbuserstore SET DEFAULT <FQDN:30013>@TENANT SAPABAP1 'PASSWORD'
</syntaxhighlight>Hinweis: Auch wenn Kennwörter geändert werden, wie z.B. vom Backupuser, dann muss der hdbuserstore neu gesetzt werden.<syntaxhighlight lang="bash">
hdbuserstore SET BACKUP <FQDN>:30013@SYSTEMDB BACKUP_OPERATOR 'PASSWORD'
</syntaxhighlight>

===Primary/Secondary Schwenk===
ACHTUNG!! NUR CODESCHNIPSEL!!! Anleitung unvollständig
<syntaxhighlight lang="bash">
Primary
hdbnsutil -sr_enable --name=S4PDB1

Secondary
hdbnsutil -sr_unregister
hdbnsutil -sr_register --remoteHost=simas4pdb --remoteInstance=00 --replicationMode=sync --operationMode=logreplay --name=S4PDB2 --force_full_replica --online
HDB start
</syntaxhighlight>

===Test===
<syntaxhighlight lang="bash">
HDB info
HDB version
</syntaxhighlight>

=== SAP User anzeigen lassen ===
<syntaxhighlight lang="sql">
hdbsql -n localhost:30013 -d <SID>-u SAPABAP1 -p '<PW>' "SELECT BNAME, USTYP FROM SAPABAP1.USR02 WHERE MANDT = '000'"
</syntaxhighlight>

=== SAP User kopieren ===
Dieser Befehl kopiert den User von 000 nach 210
<syntaxhighlight lang="sql">
hdbsql -n localhost:30013 -d <SID>-u SAPABAP1 -p '<PW>' "DELETE FROM SAPABAP1.USR02 WHERE MANDT = '210' AND BNAME = '<USER>'"; INSERT INTO SAPABAP1.USR02 (MANDT, BNAME, BCODE, GLTGV, GLTGB, USTYP, CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT, TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2, BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4, BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN, CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE, RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE, PWDSALTEDHASH, SECURITY_POLICY ) SELECT '210', BNAME, BCODE, GLTGV, GLTGB, USTYP, CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT, TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2, BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4, BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN, CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE, RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE, PWDSALTEDHASH, SECURITY_POLICY FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<USER>';"
</syntaxhighlight>

=== Passwort Hash eines einzelnen Users kopieren ===
<syntaxhighlight lang="sql">
UPDATE SAPABAP1.USR02
SET
BCODE = (SELECT BCODE FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<NAME>'),
PASSCODE = (SELECT PASSCODE FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<NAME>'),
PWDSALTEDHASH = (SELECT PWDSALTEDHASH FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<NAME>'),
CODVN = (SELECT CODVN FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<NAME>'),
PWDINITIAL = 0,
UFLAG = 0,
GLTGB = '99991231'
WHERE MANDT = '321'
AND BNAME = '<NAME>';
COMMIT;

</syntaxhighlight>

=== Passwort Hashes kompletter Mandanten kopieren ===
<syntaxhighlight lang="sql">
UPDATE SAPABAP1.USR02 T_ZIEL
SET T_ZIEL.PWDSALTEDHASH = (
SELECT T_QUELLE.PWDSALTEDHASH
FROM SAPABAP1.USR02 T_QUELLE
WHERE T_QUELLE.MANDT = '300'
AND T_QUELLE.BNAME = T_ZIEL.BNAME
)
WHERE T_ZIEL.MANDT = '301'
AND EXISTS (
SELECT 1
FROM SAPABAP1.USR02 T_CHECK
WHERE T_CHECK.MANDT = '300'
AND T_CHECK.BNAME = T_ZIEL.BNAME
);
COMMIT;
</syntaxhighlight>

===Störungen===
====Tabelle BALDAT====
in der SLG2 je Mandant alles löschen und den Job SBAL_DELETE einplanen
<syntaxhighlight lang="bash">
</syntaxhighlight>
====Transparent Huge Pages (THP) are activated or not readable====
HANA Cockpit -> SystemDB -> Database Overview -> Alerting and Diagnostics -> Alert Definitions
ID 116 auswählen Transparent Huge Pages status ( ID 116)
Edit und den Schalter "Schedule Active" auf "no" setzen und "save"
===Codeschnipsel===
SELECT * FROM SYS.M_TABLES where SCHEMA_NAME ='SAPHANADB' AND TABLE_NAME = 'LMSEMAPHORE';

===Nützliche Links===

SAP HANA

2026-02-19T13:17:17Z

Hendrik: /* Passwort Hash eines einzelnen Users kopieren */

=== Beschreibung ===
SAP HANA ist eine In-Memory-Datenbank-Plattform, die von SAP entwickelt wurde. Sie ermöglicht es Unternehmen große Datenmengen in Echtzeit zu verarbeiten und zu analysieren. Die Plattform unterstützt sowohl transaktionale als auch analytische Anwendungen und ermöglicht es den Benutzern auf umfassende Echtzeit-Analysen zuzugreifen.

Die In-Memory-Technologie von SAP HANA sorgt dafür, dass Daten in einem schnellen Zugriffsspeicher gespeichert werden, anstatt auf langsameren Festplatten. Dies führt zu schnelleren Datenzugriffszeiten und ermöglicht es Unternehmen Echtzeit-Transaktionen und Analysen durchzuführen.

SAP HANA bietet auch eine Vielzahl von Tools und Anwendungen um Unternehmen bei der Verwaltung von Daten zu unterstützen. Dazu gehören Tools zur Datenintegration, Datenbereinigung, Datenmodellierung und Analyse. Es unterstützt auch eine Vielzahl von Programmiersprachen und Entwicklungsplattformen, einschließlich SAPUI5, Java, Python uvm.
=== Download ===
[https://me.sap.com/softwarecenter/support/index SAP Downloadportal]

<nowiki>https://launchpad.support.sap.com/#/softwarecenter</nowiki> SUPPORT PACKAGES & PATCHES By Category SAP IN-MEMORY (SAP HANA ) HANA PLATFORM EDITION SAP HANA PLATFORM EDITION SAP HANA PLATFORM EDITION 2.0 SAP HANA CLIENT 2.0

=== Installation ===
hdblcm nach installieren:
https://me.sap.com/notes/2651885<syntaxhighlight lang="bash">
<install_media>/SAP_HANA_DATABASE/hdblcm --sid=<SID> --action=update --components=hdblcm --ignore=check_signature_file --verify_signature=off
</syntaxhighlight>

=== Konfiguration===

===SAP HANA Update===
Aktuelle Version als <SID>ADM mit "HDB version" anzeigen lassen

IMDB_CLIENT<VERSION> und IMDB_SERVER<VERSION> von SAP herunterladen und nach /usr/sap/SUM kopieren
<syntaxhighlight lang="bash">
sudo su - root
cd /usr/sap/SUM
chmod +x SAPCAR && ./SAPCAR -xvf IMDB_CLIENT*
chmod +x SAPCAR && ./SAPCAR -xvf IMDB_SERVER*
cd /hana/shared/*/hdblcm
./hdblcm --action=update --ignore=check_signature_file --verify_signature=off --lss_trust_unsigned_components
Enter directory root to search for components: /usr/sap/SUM/
</syntaxhighlight>
Mit "1" alles updaten und weiter den Anweisungen folgen.

===SAP HANA Sizing Report===
Quelle: [https://blogs.sap.com/2021/05/06/how-to-install-run-the-abap-on-hana-sizing-report-sap-note-1872170-a-step-by-step-guide/ HANA Sizing Report]
# Note updaten. Prüfen, ob 2810633 einbaubar ist, ansonsten 2504480 einbauen. Wenn beide nicht einbaubar sind ist das System zu alt oder bereits aktuell.
# SA38 aufrufen und /SDF/HDB_SIZING ausführen
# "Number of parallel dialog processes" auf 3 setzen
# Ausführen oder als Hintergrundjob einplanen

=== Kennwort ändern ===
<syntaxhighlight lang="bash">
ALTER USER <USER> PASSWORD "<new_password>";
</syntaxhighlight>
Oder direkt mit dem <SID>adm in der Shell:
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER <USER> PASSWORD <PW>"
</syntaxhighlight>
Hinweise:
* für System@TENANT und SYSTEM@SYSTEMDB muss der HDBUSERSTORE für das Backupscript aktualisiert werden und für die Anmeldung im HDB Studio

Lösungsschritte:

Prüfen, welche Keys angelegt sind:

<syntaxhighlight lang="bash">
hdbuserstore list
</syntaxhighlight>
Alten Key löschen (Beispiel: BACKUPKEY für SystemDB):

<syntaxhighlight lang="bash">
hdbuserstore delete BACKUP
</syntaxhighlight>
Neuen Key anlegen und dabei das aktualisierte SYSTEM‑Passwort verwenden:

<syntaxhighlight lang="bash">
hdbuserstore set BACKUP <FQDN>:30013 SYSTEM "<PW>"
</syntaxhighlight>

====SAPABAP1 Kennwort ändern====

Direkt auf Applikationsserver:
<syntaxhighlight lang="bash">
hdbsql -n <FQDN>:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPABAP1 PASSWORD <PW>"
hdbuserstore set default <FQDN>:30013@<TENANT> SAPABAP1 "<PW>"
</syntaxhighlight>
Oder im HDB Studio:
<syntaxhighlight lang="bash">
ALTER USER SAPABAP1 PASSWORD "<PW>";
</syntaxhighlight>
[https://me.sap.com/notes/0002750829 2750829 - Ändern des SAPABAP1-Schemakennworts in SAP HANA]
Anschließend siehe "SAP HANA HDB Userstore neu setzen"

==== SAPDBCTRL (SAP Host Agent) ====
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d SYSTEMDB -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
</syntaxhighlight>
Dann als root fortfahren.
Für SystemDB, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname SYSTEMDB@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>
Für Tenant, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname IMP@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>

=== Encryption root keys backup password setzen ===
<syntaxhighlight lang="bash">
ALTER SYSTEM SET ENCRYPTION ROOT KEYS BACKUP PASSWORD "<PW>";
</syntaxhighlight>
===SAP HANA HDB Userstore neu setzen===
<syntaxhighlight lang="bash">
hdbuserstore SET DEFAULT <FQDN:30013>@TENANT SAPABAP1 'PASSWORD'
</syntaxhighlight>Hinweis: Auch wenn Kennwörter geändert werden, wie z.B. vom Backupuser, dann muss der hdbuserstore neu gesetzt werden.<syntaxhighlight lang="bash">
hdbuserstore SET BACKUP <FQDN>:30013@SYSTEMDB BACKUP_OPERATOR 'PASSWORD'
</syntaxhighlight>

===Primary/Secondary Schwenk===
ACHTUNG!! NUR CODESCHNIPSEL!!! Anleitung unvollständig
<syntaxhighlight lang="bash">
Primary
hdbnsutil -sr_enable --name=S4PDB1

Secondary
hdbnsutil -sr_unregister
hdbnsutil -sr_register --remoteHost=simas4pdb --remoteInstance=00 --replicationMode=sync --operationMode=logreplay --name=S4PDB2 --force_full_replica --online
HDB start
</syntaxhighlight>

===Test===
<syntaxhighlight lang="bash">
HDB info
HDB version
</syntaxhighlight>

=== SAP User anzeigen lassen ===
<syntaxhighlight lang="sql">
hdbsql -n localhost:30013 -d <SID>-u SAPABAP1 -p '<PW>' "SELECT BNAME, USTYP FROM SAPABAP1.USR02 WHERE MANDT = '000'"
</syntaxhighlight>

=== SAP User kopieren ===
Dieser Befehl kopiert den User von 000 nach 210
<syntaxhighlight lang="sql">
hdbsql -n localhost:30013 -d <SID>-u SAPABAP1 -p '<PW>' "DELETE FROM SAPABAP1.USR02 WHERE MANDT = '210' AND BNAME = '<USER>'"; INSERT INTO SAPABAP1.USR02 (MANDT, BNAME, BCODE, GLTGV, GLTGB, USTYP, CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT, TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2, BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4, BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN, CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE, RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE, PWDSALTEDHASH, SECURITY_POLICY ) SELECT '210', BNAME, BCODE, GLTGV, GLTGB, USTYP, CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT, TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2, BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4, BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN, CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE, RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE, PWDSALTEDHASH, SECURITY_POLICY FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<USER>';"
</syntaxhighlight>

=== Passwort Hash eines einzelnen Users kopieren ===
<syntaxhighlight lang="sql">
UPDATE SAPABAP1.USR02
SET
BCODE = (SELECT BCODE FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = 'BTC-SUPPORT'),
PASSCODE = (SELECT PASSCODE FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = 'BTC-SUPPORT'),
PWDSALTEDHASH = (SELECT PWDSALTEDHASH FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = 'BTC-SUPPORT'),
CODVN = (SELECT CODVN FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = 'BTC-SUPPORT'),
PWDINITIAL = 0,
UFLAG = 0,
GLTGB = '99991231'
WHERE MANDT = '321'
AND BNAME = 'BTC-SUPPORT';
COMMIT;

</syntaxhighlight>

=== Passwort Hashes kompletter Mandanten kopieren ===
<syntaxhighlight lang="sql">
UPDATE SAPABAP1.USR02 T_ZIEL
SET T_ZIEL.PWDSALTEDHASH = (
SELECT T_QUELLE.PWDSALTEDHASH
FROM SAPABAP1.USR02 T_QUELLE
WHERE T_QUELLE.MANDT = '300'
AND T_QUELLE.BNAME = T_ZIEL.BNAME
)
WHERE T_ZIEL.MANDT = '301'
AND EXISTS (
SELECT 1
FROM SAPABAP1.USR02 T_CHECK
WHERE T_CHECK.MANDT = '300'
AND T_CHECK.BNAME = T_ZIEL.BNAME
);
COMMIT;
</syntaxhighlight>

===Störungen===
====Tabelle BALDAT====
in der SLG2 je Mandant alles löschen und den Job SBAL_DELETE einplanen
<syntaxhighlight lang="bash">
</syntaxhighlight>
====Transparent Huge Pages (THP) are activated or not readable====
HANA Cockpit -> SystemDB -> Database Overview -> Alerting and Diagnostics -> Alert Definitions
ID 116 auswählen Transparent Huge Pages status ( ID 116)
Edit und den Schalter "Schedule Active" auf "no" setzen und "save"
===Codeschnipsel===
SELECT * FROM SYS.M_TABLES where SCHEMA_NAME ='SAPHANADB' AND TABLE_NAME = 'LMSEMAPHORE';

===Nützliche Links===

Llama.cpp

2026-02-18T10:36:51Z

Hendrik: /* ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 */

=== Beschreibung ===
llama.cpp ist eine C/C++ Implementierung für Inference von Large Language Models (LLMs). Es unterstützt verschiedene Backends (CPU, Vulkan, ROCm, CUDA) und ermöglicht das Ausführen von quantisierten Modellen im GGUF-Format.

=== Download ===
<syntaxhighlight lang="bash" line="1">
git clone https://github.com/ggml-org/llama.cpp
cd llama.cpp
</syntaxhighlight>

=== Installation ===
==== Vulkan Build ====
<syntaxhighlight lang="bash" line="1">
sudo dnf install libcurl-devel -y
</syntaxhighlight>
[https://vulkan.lunarg.com/sdk/home Vulkan SDK] herunterladen
entpacken, ins Verzeichnis wechseln und source setup-env.sh ausführen. Dann ins Verzeichnis wechseln wo llama.cpp installiert werden soll.
Dort dann
<syntaxhighlight lang="bash" line="1">
source /home/hendrik/Programme/Vulkan_SDK/1.4.335.0/setup-env.sh
cd /home/hendrik/Programme/llama.cpp/
rm -R build-vulkan
cmake -B build-vulkan -DGGML_VULKAN=1
cmake --build build-vulkan --config Release -- -j $(nproc)
</syntaxhighlight>

==== ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 ====
gfx906 = Mi 50 Support 
gfx1100 = 7900 XTX Support

<syntaxhighlight lang="bash" line="1">
Copy & paste all the commands from the quick install https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/quick-start.html
Before rebooting to complete the install, download the 6.4 rocblas from the AUR: https://archlinux.org/packages/extra/x86_64/rocblas/
Extract it
Copy all tensor files that contain gfx906 in rocblas-6.4.3-3-x86_64.pkg/opt/rocm/lib/rocblas/library to /opt/rocm/lib/rocblas/library
Reboot
Check if it worked by running sudo update-alternatives --display rocm
</syntaxhighlight>
# ROCm Umgebung einrichten (optional falls Fehler auftreten)
<syntaxhighlight lang="bash" line="1">
echo 'export LD_LIBRARY_PATH=/opt/rocm/lib:$LD_LIBRARY_PATH' >> ~/.bashrc
echo 'export HSA_OVERRIDE_GFX_VERSION=9.0.6' >> ~/.bashrc # Für MI50/MI60
source ~/.bashrc
</syntaxhighlight>
<syntaxhighlight lang="bash" line="1">
sudo dnf install rocwmma-devel -y
</syntaxhighlight>
# llama.cpp kompilieren
<syntaxhighlight lang="bash" line="1">
cd /home/hendrik/Programme/llama.cpp/ && git pull && rm -R build-rocm-old && cp -R build-rocm build-rocm-old
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" cmake -B build-rocm -DGGML_HIP=ON -DAMDGPU_TARGETS=gfx1100 -DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 16
</syntaxhighlight>

Für mehrere GPU-Architekturen gleichzeitig:
<syntaxhighlight lang="bash" line="1">
# Für MI50 (gfx906) und RX 7900 XTX (gfx1100)
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
cmake -S . -B build -DGGML_HIP=ON -DAMDGPU_TARGETS="gfx906;gfx1100" \
-DCMAKE_BUILD_TYPE=Release && \
cmake --build build --config Release -- -j 8
</syntaxhighlight>

=== ik_llama.cpp ===
==== Vulkan Build ====
<syntaxhighlight lang="bash" line="1">
sudo dnf install libcurl-devel -y
</syntaxhighlight>
[https://vulkan.lunarg.com/sdk/home Vulkan SDK] herunterladen
entpacken, ins Verzeichnis wechseln und source setup-env.sh ausführen. Dann ins Verzeichnis wechseln wo llama.cpp installiert werden soll.
Dort dann
<syntaxhighlight lang="bash" line="1">
source /home/hendrik/Programme/Vulkan_SDK/1.4.335.0/setup-env.sh
cd /home/hendrik/Programme/llama.cpp/
rm -R build-vulkan
cmake -B build-vulkan -DLLAMA_VULKAN=on
cmake --build build-vulkan --config Release -- -j $(nproc)
</syntaxhighlight>

==== ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 ====
gfx906 = Mi 50 Support 
gfx1100 = 7900 XTX Support
# llama.cpp kompilieren
<syntaxhighlight lang="bash" line="1">
cd /home/hendrik/Programme/llama.cpp/ && rm -R build-rocm-old && cp -R build-rocm build-rocm-old
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" cmake -B build-rocm -DLLAMA_HIPBLAS=on -DAMDGPU_TARGETS=gfx1100 -DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 16
</syntaxhighlight>

=== Konfiguration ===

==== llama-server als systemd Service einrichten ====

'''Service-Datei erstellen:'''
<syntaxhighlight lang="bash" line="1">
sudo nano /etc/systemd/system/llama-server.service
</syntaxhighlight>

'''Service-Datei Inhalt (Multi-GPU mit ROCm):'''
<syntaxhighlight lang="ini" line="1">
[Unit]
Description=Llama.cpp ROCm Multi-GPU Server
After=network.target

[Service]
Type=simple
User=username
Group=username
WorkingDirectory=/home/username/llama.cpp/build/bin

# Multi-GPU Konfiguration
Environment="HIP_VISIBLE_DEVICES=0,1"
Environment="HSA_OVERRIDE_GFX_VERSION=9.0.6"
Environment="PATH=/opt/rocm/bin:/usr/local/bin:/usr/bin:/bin"
Environment="LD_LIBRARY_PATH=/opt/rocm/lib"

# Server mit optimalen Multi-GPU Einstellungen
ExecStart=/home/username/llama.cpp/build/bin/llama-server \
-m /home/username/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja

Restart=always
RestartSec=10
LimitNOFILE=65535
LimitMEMLOCK=infinity
StandardOutput=journal
StandardError=journal
SyslogIdentifier=llama-server

[Install]
WantedBy=multi-user.target
</syntaxhighlight>

'''Service aktivieren und starten:'''
<syntaxhighlight lang="bash" line="1">
# Service neu laden
sudo systemctl daemon-reload

# Service aktivieren (Auto-Start beim Boot)
sudo systemctl enable llama-server

# Service starten
sudo systemctl start llama-server

# Status prüfen
sudo systemctl status llama-server

# Logs anzeigen
sudo journalctl -u llama-server -f
</syntaxhighlight>

'''Service Management:'''
<syntaxhighlight lang="bash" line="1">
# Service stoppen
sudo systemctl stop llama-server

# Service neustarten
sudo systemctl restart llama-server

# Service deaktivieren
sudo systemctl disable llama-server

==== Manuelle Server-Starts ====
'''Multi-GPU (ROCm) - Optimal:'''
<syntaxhighlight lang="bash" line="1">
HIP_VISIBLE_DEVICES=0,1 ./llama-server \
-m ~/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja
</syntaxhighlight>

=== Update ===
<syntaxhighlight lang="bash" line="1">
cd ~/llama.cpp
git pull
cmake --build build --config Release -- -j $(nproc)

# Service neu starten falls aktiv
sudo systemctl restart llama-server
</syntaxhighlight>
=== Befehlszeilenargumente ===
llama-server
<syntaxhighlight lang="bash" line="1">
-h, --help, --usage print usage and exit
--version show version and build info
-cl, --cache-list show list of models in cache
--completion-bash print source-able bash completion script for llama.cpp
--verbose-prompt print a verbose prompt before generation (default: false)
-t, --threads N number of CPU threads to use during generation (default: -1)
(env: LLAMA_ARG_THREADS)
-tb, --threads-batch N number of threads to use during batch and prompt processing (default:
same as --threads)
-C, --cpu-mask M CPU affinity mask: arbitrarily long hex. Complements cpu-range
(default: "")
-Cr, --cpu-range lo-hi range of CPUs for affinity. Complements --cpu-mask
--cpu-strict <0|1> use strict CPU placement (default: 0)
--prio N set process/thread priority : low(-1), normal(0), medium(1), high(2),
realtime(3) (default: 0)
--poll <0...100> use polling level to wait for work (0 - no polling, default: 50)
-Cb, --cpu-mask-batch M CPU affinity mask: arbitrarily long hex. Complements cpu-range-batch
(default: same as --cpu-mask)
-Crb, --cpu-range-batch lo-hi ranges of CPUs for affinity. Complements --cpu-mask-batch
--cpu-strict-batch <0|1> use strict CPU placement (default: same as --cpu-strict)
--prio-batch N set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime
(default: 0)
--poll-batch <0|1> use polling to wait for work (default: same as --poll)
-c, --ctx-size N size of the prompt context (default: 4096, 0 = loaded from model)
(env: LLAMA_ARG_CTX_SIZE)
-n, --predict, --n-predict N number of tokens to predict (default: -1, -1 = infinity)
(env: LLAMA_ARG_N_PREDICT)
-b, --batch-size N logical maximum batch size (default: 2048)
(env: LLAMA_ARG_BATCH)
-ub, --ubatch-size N physical maximum batch size (default: 512)
(env: LLAMA_ARG_UBATCH)
--keep N number of tokens to keep from the initial prompt (default: 0, -1 =
all)
--swa-full use full-size SWA cache (default: false)
[(more
info)](https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
(env: LLAMA_ARG_SWA_FULL)
--kv-unified, -kvu use single unified KV buffer for the KV cache of all sequences
(default: false)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/14363)
(env: LLAMA_ARG_KV_SPLIT)
-fa, --flash-attn [on|off|auto] set Flash Attention use ('on', 'off', or 'auto', default: 'auto')
(env: LLAMA_ARG_FLASH_ATTN)
--no-perf disable internal libllama performance timings (default: false)
(env: LLAMA_ARG_NO_PERF)
-e, --escape process escapes sequences (\n, \r, \t, \', \", \\) (default: true)
--no-escape do not process escape sequences
--rope-scaling {none,linear,yarn} RoPE frequency scaling method, defaults to linear unless specified by
the model
(env: LLAMA_ARG_ROPE_SCALING_TYPE)
--rope-scale N RoPE context scaling factor, expands context by a factor of N
(env: LLAMA_ARG_ROPE_SCALE)
--rope-freq-base N RoPE base frequency, used by NTK-aware scaling (default: loaded from
model)
(env: LLAMA_ARG_ROPE_FREQ_BASE)
--rope-freq-scale N RoPE frequency scaling factor, expands context by a factor of 1/N
(env: LLAMA_ARG_ROPE_FREQ_SCALE)
--yarn-orig-ctx N YaRN: original context size of model (default: 0 = model training
context size)
(env: LLAMA_ARG_YARN_ORIG_CTX)
--yarn-ext-factor N YaRN: extrapolation mix factor (default: -1.0, 0.0 = full
interpolation)
(env: LLAMA_ARG_YARN_EXT_FACTOR)
--yarn-attn-factor N YaRN: scale sqrt(t) or attention magnitude (default: -1.0)
(env: LLAMA_ARG_YARN_ATTN_FACTOR)
--yarn-beta-slow N YaRN: high correction dim or alpha (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_SLOW)
--yarn-beta-fast N YaRN: low correction dim or beta (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_FAST)
-nkvo, --no-kv-offload disable KV offload
(env: LLAMA_ARG_NO_KV_OFFLOAD)
-nr, --no-repack disable weight repacking
(env: LLAMA_ARG_NO_REPACK)
--no-host bypass host buffer allowing extra buffers to be used
(env: LLAMA_ARG_NO_HOST)
-ctk, --cache-type-k TYPE KV cache data type for K
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K)
-ctv, --cache-type-v TYPE KV cache data type for V
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V)
-dt, --defrag-thold N KV cache defragmentation threshold (DEPRECATED)
(env: LLAMA_ARG_DEFRAG_THOLD)
-np, --parallel N number of parallel sequences to decode (default: 1)
(env: LLAMA_ARG_N_PARALLEL)
--mlock force system to keep model in RAM rather than swapping or compressing
(env: LLAMA_ARG_MLOCK)
--no-mmap do not memory-map model (slower load but may reduce pageouts if not
using mlock)
(env: LLAMA_ARG_NO_MMAP)
--numa TYPE attempt optimizations that help on some NUMA systems
- distribute: spread execution evenly over all nodes
- isolate: only spawn threads on CPUs on the node that execution
started on
- numactl: use the CPU map provided by numactl
if run without this previously, it is recommended to drop the system
page cache before using this
see https://github.com/ggml-org/llama.cpp/issues/1437
(env: LLAMA_ARG_NUMA)
-dev, --device <dev1,dev2,..> comma-separated list of devices to use for offloading (none = don't
offload)
use --list-devices to see a list of available devices
(env: LLAMA_ARG_DEVICE)
--list-devices print list of available devices and exit
--override-tensor, -ot <tensor name pattern>=<buffer type>,...
override tensor buffer type
--cpu-moe, -cmoe keep all Mixture of Experts (MoE) weights in the CPU
(env: LLAMA_ARG_CPU_MOE)
--n-cpu-moe, -ncmoe N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU
(env: LLAMA_ARG_N_CPU_MOE)
-ngl, --gpu-layers, --n-gpu-layers N max. number of layers to store in VRAM (default: -1)
(env: LLAMA_ARG_N_GPU_LAYERS)
-sm, --split-mode {none,layer,row} how to split the model across multiple GPUs, one of:
- none: use one GPU only
- layer (default): split layers and KV across GPUs
- row: split rows across GPUs
(env: LLAMA_ARG_SPLIT_MODE)
-ts, --tensor-split N0,N1,N2,... fraction of the model to offload to each GPU, comma-separated list of
proportions, e.g. 3,1
(env: LLAMA_ARG_TENSOR_SPLIT)
-mg, --main-gpu INDEX the GPU to use for the model (with split-mode = none), or for
intermediate results and KV (with split-mode = row) (default: 0)
(env: LLAMA_ARG_MAIN_GPU)
--check-tensors check model tensor data for invalid values (default: false)
--override-kv KEY=TYPE:VALUE advanced option to override model metadata by key. may be specified
multiple times.
types: int, float, bool, str. example: --override-kv
tokenizer.ggml.add_bos_token=bool:false
--no-op-offload disable offloading host tensor operations to device (default: false)
--lora FNAME path to LoRA adapter (can be repeated to use multiple adapters)
--lora-scaled FNAME SCALE path to LoRA adapter with user defined scaling (can be repeated to use
multiple adapters)
--control-vector FNAME add a control vector
note: this argument can be repeated to add multiple control vectors
--control-vector-scaled FNAME SCALE add a control vector with user defined scaling SCALE
note: this argument can be repeated to add multiple scaled control
vectors
--control-vector-layer-range START END
layer range to apply the control vector(s) to, start and end inclusive
-m, --model FNAME model path (default: `models/$filename` with filename from `--hf-file`
or `--model-url` if set, otherwise models/7B/ggml-model-f16.gguf)
(env: LLAMA_ARG_MODEL)
-mu, --model-url MODEL_URL model download url (default: unused)
(env: LLAMA_ARG_MODEL_URL)
-dr, --docker-repo [<repo>/]<model>[:quant]
Docker Hub model repository. repo is optional, default to ai/. quant
is optional, default to :latest.
example: gemma3
(default: unused)
(env: LLAMA_ARG_DOCKER_REPO)
-hf, -hfr, --hf-repo <user>/<model>[:quant]
Hugging Face model repository; quant is optional, case-insensitive,
default to Q4_K_M, or falls back to the first file in the repo if
Q4_K_M doesn't exist.
mmproj is also downloaded automatically if available. to disable, add
--no-mmproj
example: unsloth/phi-4-GGUF:q4_k_m
(default: unused)
(env: LLAMA_ARG_HF_REPO)
-hfd, -hfrd, --hf-repo-draft <user>/<model>[:quant]
Same as --hf-repo, but for the draft model (default: unused)
(env: LLAMA_ARG_HFD_REPO)
-hff, --hf-file FILE Hugging Face model file. If specified, it will override the quant in
--hf-repo (default: unused)
(env: LLAMA_ARG_HF_FILE)
-hfv, -hfrv, --hf-repo-v <user>/<model>[:quant]
Hugging Face model repository for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_REPO_V)
-hffv, --hf-file-v FILE Hugging Face model file for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_FILE_V)
-hft, --hf-token TOKEN Hugging Face access token (default: value from HF_TOKEN environment
variable)
(env: HF_TOKEN)
--log-disable Log disable
--log-file FNAME Log to file
--log-colors [on|off|auto] Set colored logging ('on', 'off', or 'auto', default: 'auto')
'auto' enables colors when output is to a terminal
(env: LLAMA_LOG_COLORS)
-v, --verbose, --log-verbose Set verbosity level to infinity (i.e. log all messages, useful for
debugging)
--offline Offline mode: forces use of cache, prevents network access
(env: LLAMA_OFFLINE)
-lv, --verbosity, --log-verbosity N Set the verbosity threshold. Messages with a higher verbosity will be
ignored.
(env: LLAMA_LOG_VERBOSITY)
--log-prefix Enable prefix in log messages
(env: LLAMA_LOG_PREFIX)
--log-timestamps Enable timestamps in log messages
(env: LLAMA_LOG_TIMESTAMPS)
-ctkd, --cache-type-k-draft TYPE KV cache data type for K for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K_DRAFT)
-ctvd, --cache-type-v-draft TYPE KV cache data type for V for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V_DRAFT)

----- sampling params -----

--samplers SAMPLERS samplers that will be used for generation in the order, separated by
';'
(default:
penalties;dry;top_n_sigma;top_k;typ_p;top_p;min_p;xtc;temperature)
-s, --seed SEED RNG seed (default: -1, use random seed for -1)
--sampling-seq, --sampler-seq SEQUENCE
simplified sequence for samplers that will be used (default:
edskypmxt)
--ignore-eos ignore end of stream token and continue generating (implies
--logit-bias EOS-inf)
--temp N temperature (default: 0.8)
--top-k N top-k sampling (default: 40, 0 = disabled)
--top-p N top-p sampling (default: 0.9, 1.0 = disabled)
--min-p N min-p sampling (default: 0.1, 0.0 = disabled)
--top-nsigma N top-n-sigma sampling (default: -1.0, -1.0 = disabled)
--xtc-probability N xtc probability (default: 0.0, 0.0 = disabled)
--xtc-threshold N xtc threshold (default: 0.1, 1.0 = disabled)
--typical N locally typical sampling, parameter p (default: 1.0, 1.0 = disabled)
--repeat-last-n N last n tokens to consider for penalize (default: 64, 0 = disabled, -1
= ctx_size)
--repeat-penalty N penalize repeat sequence of tokens (default: 1.0, 1.0 = disabled)
--presence-penalty N repeat alpha presence penalty (default: 0.0, 0.0 = disabled)
--frequency-penalty N repeat alpha frequency penalty (default: 0.0, 0.0 = disabled)
--dry-multiplier N set DRY sampling multiplier (default: 0.0, 0.0 = disabled)
--dry-base N set DRY sampling base value (default: 1.75)
--dry-allowed-length N set allowed length for DRY sampling (default: 2)
--dry-penalty-last-n N set DRY penalty for the last n tokens (default: -1, 0 = disable, -1 =
context size)
--dry-sequence-breaker STRING add sequence breaker for DRY sampling, clearing out default breakers
('\n', ':', '"', '*') in the process; use "none" to not use any
sequence breakers
--dynatemp-range N dynamic temperature range (default: 0.0, 0.0 = disabled)
--dynatemp-exp N dynamic temperature exponent (default: 1.0)
--mirostat N use Mirostat sampling.
Top K, Nucleus and Locally Typical samplers are ignored if used.
(default: 0, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)
--mirostat-lr N Mirostat learning rate, parameter eta (default: 0.1)
--mirostat-ent N Mirostat target entropy, parameter tau (default: 5.0)
-l, --logit-bias TOKEN_ID(+/-)BIAS modifies the likelihood of token appearing in the completion,
i.e. `--logit-bias 15043+1` to increase likelihood of token ' Hello',
or `--logit-bias 15043-1` to decrease likelihood of token ' Hello'
--grammar GRAMMAR BNF-like grammar to constrain generations (see samples in grammars/
dir) (default: '')
--grammar-file FNAME file to read grammar from
-j, --json-schema SCHEMA JSON schema to constrain generations (https://json-schema.org/), e.g.
`{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead
-jf, --json-schema-file FILE File containing a JSON schema to constrain generations
(https://json-schema.org/), e.g. `{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead

----- example-specific params -----

--ctx-checkpoints, --swa-checkpoints N
max number of context checkpoints to create per slot (default: 8)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/15293)
(env: LLAMA_ARG_CTX_CHECKPOINTS)
--cache-ram, -cram N set the maximum cache size in MiB (default: 8192, -1 - no limit, 0 -
disable)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/16391)
(env: LLAMA_ARG_CACHE_RAM)
--no-context-shift disables context shift on infinite text generation (default: enabled)
(env: LLAMA_ARG_NO_CONTEXT_SHIFT)
--context-shift enables context shift on infinite text generation (default: disabled)
(env: LLAMA_ARG_CONTEXT_SHIFT)
-r, --reverse-prompt PROMPT halt generation at PROMPT, return control in interactive mode
-sp, --special special tokens output enabled (default: false)
--no-warmup skip warming up the model with an empty run
--spm-infill use Suffix/Prefix/Middle pattern for infill (instead of
Prefix/Suffix/Middle) as some models prefer this. (default: disabled)
--pooling {none,mean,cls,last,rank} pooling type for embeddings, use model default if unspecified
(env: LLAMA_ARG_POOLING)
-cb, --cont-batching enable continuous batching (a.k.a dynamic batching) (default: enabled)
(env: LLAMA_ARG_CONT_BATCHING)
-nocb, --no-cont-batching disable continuous batching
(env: LLAMA_ARG_NO_CONT_BATCHING)
--mmproj FILE path to a multimodal projector file. see tools/mtmd/README.md
note: if -hf is used, this argument can be omitted
(env: LLAMA_ARG_MMPROJ)
--mmproj-url URL URL to a multimodal projector file. see tools/mtmd/README.md
(env: LLAMA_ARG_MMPROJ_URL)
--no-mmproj explicitly disable multimodal projector, useful when using -hf
(env: LLAMA_ARG_NO_MMPROJ)
--no-mmproj-offload do not offload multimodal projector to GPU
(env: LLAMA_ARG_NO_MMPROJ_OFFLOAD)
--image-min-tokens N minimum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MIN_TOKENS)
--image-max-tokens N maximum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MAX_TOKENS)
--override-tensor-draft, -otd <tensor name pattern>=<buffer type>,...
override tensor buffer type for draft model
--cpu-moe-draft, -cmoed keep all Mixture of Experts (MoE) weights in the CPU for the draft
model
(env: LLAMA_ARG_CPU_MOE_DRAFT)
--n-cpu-moe-draft, -ncmoed N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU for the draft model
(env: LLAMA_ARG_N_CPU_MOE_DRAFT)
-a, --alias STRING set alias for model name (to be used by REST API)
(env: LLAMA_ARG_ALIAS)
--host HOST ip address to listen, or bind to an UNIX socket if the address ends
with .sock (default: 127.0.0.1)
(env: LLAMA_ARG_HOST)
--port PORT port to listen (default: 8080)
(env: LLAMA_ARG_PORT)
--path PATH path to serve static files from (default: )
(env: LLAMA_ARG_STATIC_PATH)
--api-prefix PREFIX prefix path the server serves from, without the trailing slash
(default: )
(env: LLAMA_ARG_API_PREFIX)
--no-webui Disable the Web UI (default: enabled)
(env: LLAMA_ARG_NO_WEBUI)
--embedding, --embeddings restrict to only support embedding use case; use only with dedicated
embedding models (default: disabled)
(env: LLAMA_ARG_EMBEDDINGS)
--reranking, --rerank enable reranking endpoint on server (default: disabled)
(env: LLAMA_ARG_RERANKING)
--api-key KEY API key to use for authentication (default: none)
(env: LLAMA_API_KEY)
--api-key-file FNAME path to file containing API keys (default: none)
--ssl-key-file FNAME path to file a PEM-encoded SSL private key
(env: LLAMA_ARG_SSL_KEY_FILE)
--ssl-cert-file FNAME path to file a PEM-encoded SSL certificate
(env: LLAMA_ARG_SSL_CERT_FILE)
--chat-template-kwargs STRING sets additional params for the json template parser
(env: LLAMA_CHAT_TEMPLATE_KWARGS)
-to, --timeout N server read/write timeout in seconds (default: 600)
(env: LLAMA_ARG_TIMEOUT)
--threads-http N number of threads used to process HTTP requests (default: -1)
(env: LLAMA_ARG_THREADS_HTTP)
--cache-reuse N min chunk size to attempt reusing from the cache via KV shifting
(default: 0)
[(card)](https://ggml.ai/f0.png)
(env: LLAMA_ARG_CACHE_REUSE)
--metrics enable prometheus compatible metrics endpoint (default: disabled)
(env: LLAMA_ARG_ENDPOINT_METRICS)
--props enable changing global properties via POST /props (default: disabled)
(env: LLAMA_ARG_ENDPOINT_PROPS)
--slots enable slots monitoring endpoint (default: enabled)
(env: LLAMA_ARG_ENDPOINT_SLOTS)
--no-slots disables slots monitoring endpoint
(env: LLAMA_ARG_NO_ENDPOINT_SLOTS)
--slot-save-path PATH path to save slot kv cache (default: disabled)
--jinja use jinja template for chat (default: disabled)
(env: LLAMA_ARG_JINJA)
--reasoning-format FORMAT controls whether thought tags are allowed and/or extracted from the
response, and in which format they're returned; one of:
- none: leaves thoughts unparsed in `message.content`
- deepseek: puts thoughts in `message.reasoning_content`
- deepseek-legacy: keeps `<think>` tags in `message.content` while
also populating `message.reasoning_content`
(default: auto)
(env: LLAMA_ARG_THINK)
--reasoning-budget N controls the amount of thinking allowed; currently only one of: -1 for
unrestricted thinking budget, or 0 to disable thinking (default: -1)
(env: LLAMA_ARG_THINK_BUDGET)
--chat-template JINJA_TEMPLATE set custom jinja chat template (default: template taken from model's
metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE)
--chat-template-file JINJA_TEMPLATE_FILE
set custom jinja chat template file (default: template taken from
model's metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE_FILE)
--no-prefill-assistant whether to prefill the assistant's response if the last message is an
assistant message (default: prefill enabled)
when this flag is set, if the last message is an assistant message
then it will be treated as a full message and not prefilled

(env: LLAMA_ARG_NO_PREFILL_ASSISTANT)
-sps, --slot-prompt-similarity SIMILARITY
how much the prompt of a request must match the prompt of a slot in
order to use that slot (default: 0.10, 0.0 = disabled)
--lora-init-without-apply load LoRA adapters without applying them (apply later via POST
/lora-adapters) (default: disabled)
-td, --threads-draft N number of threads to use during generation (default: same as
--threads)
-tbd, --threads-batch-draft N number of threads to use during batch and prompt processing (default:
same as --threads-draft)
--draft-max, --draft, --draft-n N number of tokens to draft for speculative decoding (default: 16)
(env: LLAMA_ARG_DRAFT_MAX)
--draft-min, --draft-n-min N minimum number of draft tokens to use for speculative decoding
(default: 0)
(env: LLAMA_ARG_DRAFT_MIN)
--draft-p-min P minimum speculative decoding probability (greedy) (default: 0.8)
(env: LLAMA_ARG_DRAFT_P_MIN)
-cd, --ctx-size-draft N size of the prompt context for the draft model (default: 0, 0 = loaded
from model)
(env: LLAMA_ARG_CTX_SIZE_DRAFT)
-devd, --device-draft <dev1,dev2,..> comma-separated list of devices to use for offloading the draft model
(none = don't offload)
use --list-devices to see a list of available devices
-ngld, --gpu-layers-draft, --n-gpu-layers-draft N
number of layers to store in VRAM for the draft model
(env: LLAMA_ARG_N_GPU_LAYERS_DRAFT)
-md, --model-draft FNAME draft model for speculative decoding (default: unused)
(env: LLAMA_ARG_MODEL_DRAFT)
--spec-replace TARGET DRAFT translate the string in TARGET into DRAFT if the draft model and main
model are not compatible
-mv, --model-vocoder FNAME vocoder model for audio generation (default: unused)
--tts-use-guide-tokens Use guide tokens to improve TTS word recall
--embd-gemma-default use default EmbeddingGemma model (note: can download weights from the
internet)
--fim-qwen-1.5b-default use default Qwen 2.5 Coder 1.5B (note: can download weights from the
internet)
--fim-qwen-3b-default use default Qwen 2.5 Coder 3B (note: can download weights from the
internet)
--fim-qwen-7b-default use default Qwen 2.5 Coder 7B (note: can download weights from the
internet)
--fim-qwen-7b-spec use Qwen 2.5 Coder 7B + 0.5B draft for speculative decoding (note: can
download weights from the internet)
--fim-qwen-14b-spec use Qwen 2.5 Coder 14B + 0.5B draft for speculative decoding (note:
can download weights from the internet)
--fim-qwen-30b-default use default Qwen 3 Coder 30B A3B Instruct (note: can download weights
from the internet)
--gpt-oss-20b-default use gpt-oss-20b (note: can download weights from the internet)
--gpt-oss-120b-default use gpt-oss-120b (note: can download weights from the internet)
--vision-gemma-4b-default use Gemma 3 4B QAT (note: can download weights from the internet)
--vision-gemma-12b-default use Gemma 3 12B QAT (note: can download weights from the internet)
</syntaxhighlight>
=== Test ===
=== Bekannte Probleme ===
<syntaxhighlight lang="bash" line="1">
Error 1 [ 34%] Linking HIP shared library ../../../bin/libggml-hip.so [ 34%] Built target ggml-hip gmake[1]: *** [CMakeFiles/Makefile2:1775: ggml/src/CMakeFiles/ggml-cpu.dir/all] Error 2 gmake: *** [Makefile:146: all] Error 2
</syntaxhighlight>
Lösung:
<syntaxhighlight lang="bash" line="1">
GCC 13.3.0 konnte damit nicht umgehen, aber GCC 14.2.0 hat bessere C++-Standard-Compliance und kann diesen Header-Konflikt korrekt auflösen – deshalb funktioniert die Kompilierung jetzt mit der neueren Compiler-Version.
sudo update-alternatives --remove gcc /usr/bin/gcc-13
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-14 100
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-13 50
sudo update-alternatives --config gcc
# Wähle gcc-14

sudo update-alternatives --remove g++ /usr/bin/g++-13
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-14 100
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-13 50
sudo update-alternatives --config g++
# Wähle g++-14
</syntaxhighlight>
=== Nützliche Links ===
* [https://github.com/ggml-org/llama.cpp Offizielles llama.cpp Repository]
* [https://github.com/ggml-org/llama.cpp/blob/master/examples/server/README.md llama-server Dokumentation]
* [https://github.com/ggml-org/llama.cpp/discussions llama.cpp Discussions]
* [https://huggingface.co/models?library=gguf GGUF Modelle auf Hugging Face]
* [https://docs.rocm.com/ ROCm Dokumentation]
* [https://vulkan.lunarg.com/doc/sdk Vulkan SDK Dokumentation]

Llama.cpp

2026-02-13T21:42:51Z

Hendrik:

=== Beschreibung ===
llama.cpp ist eine C/C++ Implementierung für Inference von Large Language Models (LLMs). Es unterstützt verschiedene Backends (CPU, Vulkan, ROCm, CUDA) und ermöglicht das Ausführen von quantisierten Modellen im GGUF-Format.

=== Download ===
<syntaxhighlight lang="bash" line="1">
git clone https://github.com/ggml-org/llama.cpp
cd llama.cpp
</syntaxhighlight>

=== Installation ===
==== Vulkan Build ====
<syntaxhighlight lang="bash" line="1">
sudo dnf install libcurl-devel -y
</syntaxhighlight>
[https://vulkan.lunarg.com/sdk/home Vulkan SDK] herunterladen
entpacken, ins Verzeichnis wechseln und source setup-env.sh ausführen. Dann ins Verzeichnis wechseln wo llama.cpp installiert werden soll.
Dort dann
<syntaxhighlight lang="bash" line="1">
source /home/hendrik/Programme/Vulkan_SDK/1.4.335.0/setup-env.sh
cd /home/hendrik/Programme/llama.cpp/
rm -R build-vulkan
cmake -B build-vulkan -DGGML_VULKAN=1
cmake --build build-vulkan --config Release -- -j $(nproc)
</syntaxhighlight>

==== ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 ====
gfx906 = Mi 50 Support 
gfx1100 = 7900 XTX Support

<syntaxhighlight lang="bash" line="1">
Copy & paste all the commands from the quick install https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/quick-start.html
Before rebooting to complete the install, download the 6.4 rocblas from the AUR: https://archlinux.org/packages/extra/x86_64/rocblas/
Extract it
Copy all tensor files that contain gfx906 in rocblas-6.4.3-3-x86_64.pkg/opt/rocm/lib/rocblas/library to /opt/rocm/lib/rocblas/library
Reboot
Check if it worked by running sudo update-alternatives --display rocm
</syntaxhighlight>
# ROCm Umgebung einrichten (optional falls Fehler auftreten)
<syntaxhighlight lang="bash" line="1">
echo 'export LD_LIBRARY_PATH=/opt/rocm/lib:$LD_LIBRARY_PATH' >> ~/.bashrc
echo 'export HSA_OVERRIDE_GFX_VERSION=9.0.6' >> ~/.bashrc # Für MI50/MI60
source ~/.bashrc
</syntaxhighlight>
<syntaxhighlight lang="bash" line="1">
sudo dnf install rocwmma-devel -y
</syntaxhighlight>
# llama.cpp kompilieren
<syntaxhighlight lang="bash" line="1">
cd /home/hendrik/Programme/llama.cpp/ && rm -R build-rocm-old && cp -R build-rocm build-rocm-old
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" cmake -B build-rocm -DGGML_HIP=ON -DAMDGPU_TARGETS=gfx1100 -DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 16
</syntaxhighlight>

Für mehrere GPU-Architekturen gleichzeitig:
<syntaxhighlight lang="bash" line="1">
# Für MI50 (gfx906) und RX 7900 XTX (gfx1100)
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
cmake -S . -B build -DGGML_HIP=ON -DAMDGPU_TARGETS="gfx906;gfx1100" \
-DCMAKE_BUILD_TYPE=Release && \
cmake --build build --config Release -- -j 8
</syntaxhighlight>

=== ik_llama.cpp ===
==== Vulkan Build ====
<syntaxhighlight lang="bash" line="1">
sudo dnf install libcurl-devel -y
</syntaxhighlight>
[https://vulkan.lunarg.com/sdk/home Vulkan SDK] herunterladen
entpacken, ins Verzeichnis wechseln und source setup-env.sh ausführen. Dann ins Verzeichnis wechseln wo llama.cpp installiert werden soll.
Dort dann
<syntaxhighlight lang="bash" line="1">
source /home/hendrik/Programme/Vulkan_SDK/1.4.335.0/setup-env.sh
cd /home/hendrik/Programme/llama.cpp/
rm -R build-vulkan
cmake -B build-vulkan -DLLAMA_VULKAN=on
cmake --build build-vulkan --config Release -- -j $(nproc)
</syntaxhighlight>

==== ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 ====
gfx906 = Mi 50 Support 
gfx1100 = 7900 XTX Support
# llama.cpp kompilieren
<syntaxhighlight lang="bash" line="1">
cd /home/hendrik/Programme/llama.cpp/ && rm -R build-rocm-old && cp -R build-rocm build-rocm-old
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" cmake -B build-rocm -DLLAMA_HIPBLAS=on -DAMDGPU_TARGETS=gfx1100 -DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 16
</syntaxhighlight>

=== Konfiguration ===

==== llama-server als systemd Service einrichten ====

'''Service-Datei erstellen:'''
<syntaxhighlight lang="bash" line="1">
sudo nano /etc/systemd/system/llama-server.service
</syntaxhighlight>

'''Service-Datei Inhalt (Multi-GPU mit ROCm):'''
<syntaxhighlight lang="ini" line="1">
[Unit]
Description=Llama.cpp ROCm Multi-GPU Server
After=network.target

[Service]
Type=simple
User=username
Group=username
WorkingDirectory=/home/username/llama.cpp/build/bin

# Multi-GPU Konfiguration
Environment="HIP_VISIBLE_DEVICES=0,1"
Environment="HSA_OVERRIDE_GFX_VERSION=9.0.6"
Environment="PATH=/opt/rocm/bin:/usr/local/bin:/usr/bin:/bin"
Environment="LD_LIBRARY_PATH=/opt/rocm/lib"

# Server mit optimalen Multi-GPU Einstellungen
ExecStart=/home/username/llama.cpp/build/bin/llama-server \
-m /home/username/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja

Restart=always
RestartSec=10
LimitNOFILE=65535
LimitMEMLOCK=infinity
StandardOutput=journal
StandardError=journal
SyslogIdentifier=llama-server

[Install]
WantedBy=multi-user.target
</syntaxhighlight>

'''Service aktivieren und starten:'''
<syntaxhighlight lang="bash" line="1">
# Service neu laden
sudo systemctl daemon-reload

# Service aktivieren (Auto-Start beim Boot)
sudo systemctl enable llama-server

# Service starten
sudo systemctl start llama-server

# Status prüfen
sudo systemctl status llama-server

# Logs anzeigen
sudo journalctl -u llama-server -f
</syntaxhighlight>

'''Service Management:'''
<syntaxhighlight lang="bash" line="1">
# Service stoppen
sudo systemctl stop llama-server

# Service neustarten
sudo systemctl restart llama-server

# Service deaktivieren
sudo systemctl disable llama-server

==== Manuelle Server-Starts ====
'''Multi-GPU (ROCm) - Optimal:'''
<syntaxhighlight lang="bash" line="1">
HIP_VISIBLE_DEVICES=0,1 ./llama-server \
-m ~/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja
</syntaxhighlight>

=== Update ===
<syntaxhighlight lang="bash" line="1">
cd ~/llama.cpp
git pull
cmake --build build --config Release -- -j $(nproc)

# Service neu starten falls aktiv
sudo systemctl restart llama-server
</syntaxhighlight>
=== Befehlszeilenargumente ===
llama-server
<syntaxhighlight lang="bash" line="1">
-h, --help, --usage print usage and exit
--version show version and build info
-cl, --cache-list show list of models in cache
--completion-bash print source-able bash completion script for llama.cpp
--verbose-prompt print a verbose prompt before generation (default: false)
-t, --threads N number of CPU threads to use during generation (default: -1)
(env: LLAMA_ARG_THREADS)
-tb, --threads-batch N number of threads to use during batch and prompt processing (default:
same as --threads)
-C, --cpu-mask M CPU affinity mask: arbitrarily long hex. Complements cpu-range
(default: "")
-Cr, --cpu-range lo-hi range of CPUs for affinity. Complements --cpu-mask
--cpu-strict <0|1> use strict CPU placement (default: 0)
--prio N set process/thread priority : low(-1), normal(0), medium(1), high(2),
realtime(3) (default: 0)
--poll <0...100> use polling level to wait for work (0 - no polling, default: 50)
-Cb, --cpu-mask-batch M CPU affinity mask: arbitrarily long hex. Complements cpu-range-batch
(default: same as --cpu-mask)
-Crb, --cpu-range-batch lo-hi ranges of CPUs for affinity. Complements --cpu-mask-batch
--cpu-strict-batch <0|1> use strict CPU placement (default: same as --cpu-strict)
--prio-batch N set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime
(default: 0)
--poll-batch <0|1> use polling to wait for work (default: same as --poll)
-c, --ctx-size N size of the prompt context (default: 4096, 0 = loaded from model)
(env: LLAMA_ARG_CTX_SIZE)
-n, --predict, --n-predict N number of tokens to predict (default: -1, -1 = infinity)
(env: LLAMA_ARG_N_PREDICT)
-b, --batch-size N logical maximum batch size (default: 2048)
(env: LLAMA_ARG_BATCH)
-ub, --ubatch-size N physical maximum batch size (default: 512)
(env: LLAMA_ARG_UBATCH)
--keep N number of tokens to keep from the initial prompt (default: 0, -1 =
all)
--swa-full use full-size SWA cache (default: false)
[(more
info)](https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
(env: LLAMA_ARG_SWA_FULL)
--kv-unified, -kvu use single unified KV buffer for the KV cache of all sequences
(default: false)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/14363)
(env: LLAMA_ARG_KV_SPLIT)
-fa, --flash-attn [on|off|auto] set Flash Attention use ('on', 'off', or 'auto', default: 'auto')
(env: LLAMA_ARG_FLASH_ATTN)
--no-perf disable internal libllama performance timings (default: false)
(env: LLAMA_ARG_NO_PERF)
-e, --escape process escapes sequences (\n, \r, \t, \', \", \\) (default: true)
--no-escape do not process escape sequences
--rope-scaling {none,linear,yarn} RoPE frequency scaling method, defaults to linear unless specified by
the model
(env: LLAMA_ARG_ROPE_SCALING_TYPE)
--rope-scale N RoPE context scaling factor, expands context by a factor of N
(env: LLAMA_ARG_ROPE_SCALE)
--rope-freq-base N RoPE base frequency, used by NTK-aware scaling (default: loaded from
model)
(env: LLAMA_ARG_ROPE_FREQ_BASE)
--rope-freq-scale N RoPE frequency scaling factor, expands context by a factor of 1/N
(env: LLAMA_ARG_ROPE_FREQ_SCALE)
--yarn-orig-ctx N YaRN: original context size of model (default: 0 = model training
context size)
(env: LLAMA_ARG_YARN_ORIG_CTX)
--yarn-ext-factor N YaRN: extrapolation mix factor (default: -1.0, 0.0 = full
interpolation)
(env: LLAMA_ARG_YARN_EXT_FACTOR)
--yarn-attn-factor N YaRN: scale sqrt(t) or attention magnitude (default: -1.0)
(env: LLAMA_ARG_YARN_ATTN_FACTOR)
--yarn-beta-slow N YaRN: high correction dim or alpha (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_SLOW)
--yarn-beta-fast N YaRN: low correction dim or beta (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_FAST)
-nkvo, --no-kv-offload disable KV offload
(env: LLAMA_ARG_NO_KV_OFFLOAD)
-nr, --no-repack disable weight repacking
(env: LLAMA_ARG_NO_REPACK)
--no-host bypass host buffer allowing extra buffers to be used
(env: LLAMA_ARG_NO_HOST)
-ctk, --cache-type-k TYPE KV cache data type for K
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K)
-ctv, --cache-type-v TYPE KV cache data type for V
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V)
-dt, --defrag-thold N KV cache defragmentation threshold (DEPRECATED)
(env: LLAMA_ARG_DEFRAG_THOLD)
-np, --parallel N number of parallel sequences to decode (default: 1)
(env: LLAMA_ARG_N_PARALLEL)
--mlock force system to keep model in RAM rather than swapping or compressing
(env: LLAMA_ARG_MLOCK)
--no-mmap do not memory-map model (slower load but may reduce pageouts if not
using mlock)
(env: LLAMA_ARG_NO_MMAP)
--numa TYPE attempt optimizations that help on some NUMA systems
- distribute: spread execution evenly over all nodes
- isolate: only spawn threads on CPUs on the node that execution
started on
- numactl: use the CPU map provided by numactl
if run without this previously, it is recommended to drop the system
page cache before using this
see https://github.com/ggml-org/llama.cpp/issues/1437
(env: LLAMA_ARG_NUMA)
-dev, --device <dev1,dev2,..> comma-separated list of devices to use for offloading (none = don't
offload)
use --list-devices to see a list of available devices
(env: LLAMA_ARG_DEVICE)
--list-devices print list of available devices and exit
--override-tensor, -ot <tensor name pattern>=<buffer type>,...
override tensor buffer type
--cpu-moe, -cmoe keep all Mixture of Experts (MoE) weights in the CPU
(env: LLAMA_ARG_CPU_MOE)
--n-cpu-moe, -ncmoe N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU
(env: LLAMA_ARG_N_CPU_MOE)
-ngl, --gpu-layers, --n-gpu-layers N max. number of layers to store in VRAM (default: -1)
(env: LLAMA_ARG_N_GPU_LAYERS)
-sm, --split-mode {none,layer,row} how to split the model across multiple GPUs, one of:
- none: use one GPU only
- layer (default): split layers and KV across GPUs
- row: split rows across GPUs
(env: LLAMA_ARG_SPLIT_MODE)
-ts, --tensor-split N0,N1,N2,... fraction of the model to offload to each GPU, comma-separated list of
proportions, e.g. 3,1
(env: LLAMA_ARG_TENSOR_SPLIT)
-mg, --main-gpu INDEX the GPU to use for the model (with split-mode = none), or for
intermediate results and KV (with split-mode = row) (default: 0)
(env: LLAMA_ARG_MAIN_GPU)
--check-tensors check model tensor data for invalid values (default: false)
--override-kv KEY=TYPE:VALUE advanced option to override model metadata by key. may be specified
multiple times.
types: int, float, bool, str. example: --override-kv
tokenizer.ggml.add_bos_token=bool:false
--no-op-offload disable offloading host tensor operations to device (default: false)
--lora FNAME path to LoRA adapter (can be repeated to use multiple adapters)
--lora-scaled FNAME SCALE path to LoRA adapter with user defined scaling (can be repeated to use
multiple adapters)
--control-vector FNAME add a control vector
note: this argument can be repeated to add multiple control vectors
--control-vector-scaled FNAME SCALE add a control vector with user defined scaling SCALE
note: this argument can be repeated to add multiple scaled control
vectors
--control-vector-layer-range START END
layer range to apply the control vector(s) to, start and end inclusive
-m, --model FNAME model path (default: `models/$filename` with filename from `--hf-file`
or `--model-url` if set, otherwise models/7B/ggml-model-f16.gguf)
(env: LLAMA_ARG_MODEL)
-mu, --model-url MODEL_URL model download url (default: unused)
(env: LLAMA_ARG_MODEL_URL)
-dr, --docker-repo [<repo>/]<model>[:quant]
Docker Hub model repository. repo is optional, default to ai/. quant
is optional, default to :latest.
example: gemma3
(default: unused)
(env: LLAMA_ARG_DOCKER_REPO)
-hf, -hfr, --hf-repo <user>/<model>[:quant]
Hugging Face model repository; quant is optional, case-insensitive,
default to Q4_K_M, or falls back to the first file in the repo if
Q4_K_M doesn't exist.
mmproj is also downloaded automatically if available. to disable, add
--no-mmproj
example: unsloth/phi-4-GGUF:q4_k_m
(default: unused)
(env: LLAMA_ARG_HF_REPO)
-hfd, -hfrd, --hf-repo-draft <user>/<model>[:quant]
Same as --hf-repo, but for the draft model (default: unused)
(env: LLAMA_ARG_HFD_REPO)
-hff, --hf-file FILE Hugging Face model file. If specified, it will override the quant in
--hf-repo (default: unused)
(env: LLAMA_ARG_HF_FILE)
-hfv, -hfrv, --hf-repo-v <user>/<model>[:quant]
Hugging Face model repository for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_REPO_V)
-hffv, --hf-file-v FILE Hugging Face model file for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_FILE_V)
-hft, --hf-token TOKEN Hugging Face access token (default: value from HF_TOKEN environment
variable)
(env: HF_TOKEN)
--log-disable Log disable
--log-file FNAME Log to file
--log-colors [on|off|auto] Set colored logging ('on', 'off', or 'auto', default: 'auto')
'auto' enables colors when output is to a terminal
(env: LLAMA_LOG_COLORS)
-v, --verbose, --log-verbose Set verbosity level to infinity (i.e. log all messages, useful for
debugging)
--offline Offline mode: forces use of cache, prevents network access
(env: LLAMA_OFFLINE)
-lv, --verbosity, --log-verbosity N Set the verbosity threshold. Messages with a higher verbosity will be
ignored.
(env: LLAMA_LOG_VERBOSITY)
--log-prefix Enable prefix in log messages
(env: LLAMA_LOG_PREFIX)
--log-timestamps Enable timestamps in log messages
(env: LLAMA_LOG_TIMESTAMPS)
-ctkd, --cache-type-k-draft TYPE KV cache data type for K for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K_DRAFT)
-ctvd, --cache-type-v-draft TYPE KV cache data type for V for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V_DRAFT)

----- sampling params -----

--samplers SAMPLERS samplers that will be used for generation in the order, separated by
';'
(default:
penalties;dry;top_n_sigma;top_k;typ_p;top_p;min_p;xtc;temperature)
-s, --seed SEED RNG seed (default: -1, use random seed for -1)
--sampling-seq, --sampler-seq SEQUENCE
simplified sequence for samplers that will be used (default:
edskypmxt)
--ignore-eos ignore end of stream token and continue generating (implies
--logit-bias EOS-inf)
--temp N temperature (default: 0.8)
--top-k N top-k sampling (default: 40, 0 = disabled)
--top-p N top-p sampling (default: 0.9, 1.0 = disabled)
--min-p N min-p sampling (default: 0.1, 0.0 = disabled)
--top-nsigma N top-n-sigma sampling (default: -1.0, -1.0 = disabled)
--xtc-probability N xtc probability (default: 0.0, 0.0 = disabled)
--xtc-threshold N xtc threshold (default: 0.1, 1.0 = disabled)
--typical N locally typical sampling, parameter p (default: 1.0, 1.0 = disabled)
--repeat-last-n N last n tokens to consider for penalize (default: 64, 0 = disabled, -1
= ctx_size)
--repeat-penalty N penalize repeat sequence of tokens (default: 1.0, 1.0 = disabled)
--presence-penalty N repeat alpha presence penalty (default: 0.0, 0.0 = disabled)
--frequency-penalty N repeat alpha frequency penalty (default: 0.0, 0.0 = disabled)
--dry-multiplier N set DRY sampling multiplier (default: 0.0, 0.0 = disabled)
--dry-base N set DRY sampling base value (default: 1.75)
--dry-allowed-length N set allowed length for DRY sampling (default: 2)
--dry-penalty-last-n N set DRY penalty for the last n tokens (default: -1, 0 = disable, -1 =
context size)
--dry-sequence-breaker STRING add sequence breaker for DRY sampling, clearing out default breakers
('\n', ':', '"', '*') in the process; use "none" to not use any
sequence breakers
--dynatemp-range N dynamic temperature range (default: 0.0, 0.0 = disabled)
--dynatemp-exp N dynamic temperature exponent (default: 1.0)
--mirostat N use Mirostat sampling.
Top K, Nucleus and Locally Typical samplers are ignored if used.
(default: 0, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)
--mirostat-lr N Mirostat learning rate, parameter eta (default: 0.1)
--mirostat-ent N Mirostat target entropy, parameter tau (default: 5.0)
-l, --logit-bias TOKEN_ID(+/-)BIAS modifies the likelihood of token appearing in the completion,
i.e. `--logit-bias 15043+1` to increase likelihood of token ' Hello',
or `--logit-bias 15043-1` to decrease likelihood of token ' Hello'
--grammar GRAMMAR BNF-like grammar to constrain generations (see samples in grammars/
dir) (default: '')
--grammar-file FNAME file to read grammar from
-j, --json-schema SCHEMA JSON schema to constrain generations (https://json-schema.org/), e.g.
`{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead
-jf, --json-schema-file FILE File containing a JSON schema to constrain generations
(https://json-schema.org/), e.g. `{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead

----- example-specific params -----

--ctx-checkpoints, --swa-checkpoints N
max number of context checkpoints to create per slot (default: 8)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/15293)
(env: LLAMA_ARG_CTX_CHECKPOINTS)
--cache-ram, -cram N set the maximum cache size in MiB (default: 8192, -1 - no limit, 0 -
disable)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/16391)
(env: LLAMA_ARG_CACHE_RAM)
--no-context-shift disables context shift on infinite text generation (default: enabled)
(env: LLAMA_ARG_NO_CONTEXT_SHIFT)
--context-shift enables context shift on infinite text generation (default: disabled)
(env: LLAMA_ARG_CONTEXT_SHIFT)
-r, --reverse-prompt PROMPT halt generation at PROMPT, return control in interactive mode
-sp, --special special tokens output enabled (default: false)
--no-warmup skip warming up the model with an empty run
--spm-infill use Suffix/Prefix/Middle pattern for infill (instead of
Prefix/Suffix/Middle) as some models prefer this. (default: disabled)
--pooling {none,mean,cls,last,rank} pooling type for embeddings, use model default if unspecified
(env: LLAMA_ARG_POOLING)
-cb, --cont-batching enable continuous batching (a.k.a dynamic batching) (default: enabled)
(env: LLAMA_ARG_CONT_BATCHING)
-nocb, --no-cont-batching disable continuous batching
(env: LLAMA_ARG_NO_CONT_BATCHING)
--mmproj FILE path to a multimodal projector file. see tools/mtmd/README.md
note: if -hf is used, this argument can be omitted
(env: LLAMA_ARG_MMPROJ)
--mmproj-url URL URL to a multimodal projector file. see tools/mtmd/README.md
(env: LLAMA_ARG_MMPROJ_URL)
--no-mmproj explicitly disable multimodal projector, useful when using -hf
(env: LLAMA_ARG_NO_MMPROJ)
--no-mmproj-offload do not offload multimodal projector to GPU
(env: LLAMA_ARG_NO_MMPROJ_OFFLOAD)
--image-min-tokens N minimum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MIN_TOKENS)
--image-max-tokens N maximum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MAX_TOKENS)
--override-tensor-draft, -otd <tensor name pattern>=<buffer type>,...
override tensor buffer type for draft model
--cpu-moe-draft, -cmoed keep all Mixture of Experts (MoE) weights in the CPU for the draft
model
(env: LLAMA_ARG_CPU_MOE_DRAFT)
--n-cpu-moe-draft, -ncmoed N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU for the draft model
(env: LLAMA_ARG_N_CPU_MOE_DRAFT)
-a, --alias STRING set alias for model name (to be used by REST API)
(env: LLAMA_ARG_ALIAS)
--host HOST ip address to listen, or bind to an UNIX socket if the address ends
with .sock (default: 127.0.0.1)
(env: LLAMA_ARG_HOST)
--port PORT port to listen (default: 8080)
(env: LLAMA_ARG_PORT)
--path PATH path to serve static files from (default: )
(env: LLAMA_ARG_STATIC_PATH)
--api-prefix PREFIX prefix path the server serves from, without the trailing slash
(default: )
(env: LLAMA_ARG_API_PREFIX)
--no-webui Disable the Web UI (default: enabled)
(env: LLAMA_ARG_NO_WEBUI)
--embedding, --embeddings restrict to only support embedding use case; use only with dedicated
embedding models (default: disabled)
(env: LLAMA_ARG_EMBEDDINGS)
--reranking, --rerank enable reranking endpoint on server (default: disabled)
(env: LLAMA_ARG_RERANKING)
--api-key KEY API key to use for authentication (default: none)
(env: LLAMA_API_KEY)
--api-key-file FNAME path to file containing API keys (default: none)
--ssl-key-file FNAME path to file a PEM-encoded SSL private key
(env: LLAMA_ARG_SSL_KEY_FILE)
--ssl-cert-file FNAME path to file a PEM-encoded SSL certificate
(env: LLAMA_ARG_SSL_CERT_FILE)
--chat-template-kwargs STRING sets additional params for the json template parser
(env: LLAMA_CHAT_TEMPLATE_KWARGS)
-to, --timeout N server read/write timeout in seconds (default: 600)
(env: LLAMA_ARG_TIMEOUT)
--threads-http N number of threads used to process HTTP requests (default: -1)
(env: LLAMA_ARG_THREADS_HTTP)
--cache-reuse N min chunk size to attempt reusing from the cache via KV shifting
(default: 0)
[(card)](https://ggml.ai/f0.png)
(env: LLAMA_ARG_CACHE_REUSE)
--metrics enable prometheus compatible metrics endpoint (default: disabled)
(env: LLAMA_ARG_ENDPOINT_METRICS)
--props enable changing global properties via POST /props (default: disabled)
(env: LLAMA_ARG_ENDPOINT_PROPS)
--slots enable slots monitoring endpoint (default: enabled)
(env: LLAMA_ARG_ENDPOINT_SLOTS)
--no-slots disables slots monitoring endpoint
(env: LLAMA_ARG_NO_ENDPOINT_SLOTS)
--slot-save-path PATH path to save slot kv cache (default: disabled)
--jinja use jinja template for chat (default: disabled)
(env: LLAMA_ARG_JINJA)
--reasoning-format FORMAT controls whether thought tags are allowed and/or extracted from the
response, and in which format they're returned; one of:
- none: leaves thoughts unparsed in `message.content`
- deepseek: puts thoughts in `message.reasoning_content`
- deepseek-legacy: keeps `<think>` tags in `message.content` while
also populating `message.reasoning_content`
(default: auto)
(env: LLAMA_ARG_THINK)
--reasoning-budget N controls the amount of thinking allowed; currently only one of: -1 for
unrestricted thinking budget, or 0 to disable thinking (default: -1)
(env: LLAMA_ARG_THINK_BUDGET)
--chat-template JINJA_TEMPLATE set custom jinja chat template (default: template taken from model's
metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE)
--chat-template-file JINJA_TEMPLATE_FILE
set custom jinja chat template file (default: template taken from
model's metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE_FILE)
--no-prefill-assistant whether to prefill the assistant's response if the last message is an
assistant message (default: prefill enabled)
when this flag is set, if the last message is an assistant message
then it will be treated as a full message and not prefilled

(env: LLAMA_ARG_NO_PREFILL_ASSISTANT)
-sps, --slot-prompt-similarity SIMILARITY
how much the prompt of a request must match the prompt of a slot in
order to use that slot (default: 0.10, 0.0 = disabled)
--lora-init-without-apply load LoRA adapters without applying them (apply later via POST
/lora-adapters) (default: disabled)
-td, --threads-draft N number of threads to use during generation (default: same as
--threads)
-tbd, --threads-batch-draft N number of threads to use during batch and prompt processing (default:
same as --threads-draft)
--draft-max, --draft, --draft-n N number of tokens to draft for speculative decoding (default: 16)
(env: LLAMA_ARG_DRAFT_MAX)
--draft-min, --draft-n-min N minimum number of draft tokens to use for speculative decoding
(default: 0)
(env: LLAMA_ARG_DRAFT_MIN)
--draft-p-min P minimum speculative decoding probability (greedy) (default: 0.8)
(env: LLAMA_ARG_DRAFT_P_MIN)
-cd, --ctx-size-draft N size of the prompt context for the draft model (default: 0, 0 = loaded
from model)
(env: LLAMA_ARG_CTX_SIZE_DRAFT)
-devd, --device-draft <dev1,dev2,..> comma-separated list of devices to use for offloading the draft model
(none = don't offload)
use --list-devices to see a list of available devices
-ngld, --gpu-layers-draft, --n-gpu-layers-draft N
number of layers to store in VRAM for the draft model
(env: LLAMA_ARG_N_GPU_LAYERS_DRAFT)
-md, --model-draft FNAME draft model for speculative decoding (default: unused)
(env: LLAMA_ARG_MODEL_DRAFT)
--spec-replace TARGET DRAFT translate the string in TARGET into DRAFT if the draft model and main
model are not compatible
-mv, --model-vocoder FNAME vocoder model for audio generation (default: unused)
--tts-use-guide-tokens Use guide tokens to improve TTS word recall
--embd-gemma-default use default EmbeddingGemma model (note: can download weights from the
internet)
--fim-qwen-1.5b-default use default Qwen 2.5 Coder 1.5B (note: can download weights from the
internet)
--fim-qwen-3b-default use default Qwen 2.5 Coder 3B (note: can download weights from the
internet)
--fim-qwen-7b-default use default Qwen 2.5 Coder 7B (note: can download weights from the
internet)
--fim-qwen-7b-spec use Qwen 2.5 Coder 7B + 0.5B draft for speculative decoding (note: can
download weights from the internet)
--fim-qwen-14b-spec use Qwen 2.5 Coder 14B + 0.5B draft for speculative decoding (note:
can download weights from the internet)
--fim-qwen-30b-default use default Qwen 3 Coder 30B A3B Instruct (note: can download weights
from the internet)
--gpt-oss-20b-default use gpt-oss-20b (note: can download weights from the internet)
--gpt-oss-120b-default use gpt-oss-120b (note: can download weights from the internet)
--vision-gemma-4b-default use Gemma 3 4B QAT (note: can download weights from the internet)
--vision-gemma-12b-default use Gemma 3 12B QAT (note: can download weights from the internet)
</syntaxhighlight>
=== Test ===
=== Bekannte Probleme ===
<syntaxhighlight lang="bash" line="1">
Error 1 [ 34%] Linking HIP shared library ../../../bin/libggml-hip.so [ 34%] Built target ggml-hip gmake[1]: *** [CMakeFiles/Makefile2:1775: ggml/src/CMakeFiles/ggml-cpu.dir/all] Error 2 gmake: *** [Makefile:146: all] Error 2
</syntaxhighlight>
Lösung:
<syntaxhighlight lang="bash" line="1">
GCC 13.3.0 konnte damit nicht umgehen, aber GCC 14.2.0 hat bessere C++-Standard-Compliance und kann diesen Header-Konflikt korrekt auflösen – deshalb funktioniert die Kompilierung jetzt mit der neueren Compiler-Version.
sudo update-alternatives --remove gcc /usr/bin/gcc-13
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-14 100
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-13 50
sudo update-alternatives --config gcc
# Wähle gcc-14

sudo update-alternatives --remove g++ /usr/bin/g++-13
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-14 100
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-13 50
sudo update-alternatives --config g++
# Wähle g++-14
</syntaxhighlight>
=== Nützliche Links ===
* [https://github.com/ggml-org/llama.cpp Offizielles llama.cpp Repository]
* [https://github.com/ggml-org/llama.cpp/blob/master/examples/server/README.md llama-server Dokumentation]
* [https://github.com/ggml-org/llama.cpp/discussions llama.cpp Discussions]
* [https://huggingface.co/models?library=gguf GGUF Modelle auf Hugging Face]
* [https://docs.rocm.com/ ROCm Dokumentation]
* [https://vulkan.lunarg.com/doc/sdk Vulkan SDK Dokumentation]

SAP HANA

2026-02-12T21:49:20Z

Hendrik:

=== Beschreibung ===
SAP HANA ist eine In-Memory-Datenbank-Plattform, die von SAP entwickelt wurde. Sie ermöglicht es Unternehmen große Datenmengen in Echtzeit zu verarbeiten und zu analysieren. Die Plattform unterstützt sowohl transaktionale als auch analytische Anwendungen und ermöglicht es den Benutzern auf umfassende Echtzeit-Analysen zuzugreifen.

Die In-Memory-Technologie von SAP HANA sorgt dafür, dass Daten in einem schnellen Zugriffsspeicher gespeichert werden, anstatt auf langsameren Festplatten. Dies führt zu schnelleren Datenzugriffszeiten und ermöglicht es Unternehmen Echtzeit-Transaktionen und Analysen durchzuführen.

SAP HANA bietet auch eine Vielzahl von Tools und Anwendungen um Unternehmen bei der Verwaltung von Daten zu unterstützen. Dazu gehören Tools zur Datenintegration, Datenbereinigung, Datenmodellierung und Analyse. Es unterstützt auch eine Vielzahl von Programmiersprachen und Entwicklungsplattformen, einschließlich SAPUI5, Java, Python uvm.
=== Download ===
[https://me.sap.com/softwarecenter/support/index SAP Downloadportal]

<nowiki>https://launchpad.support.sap.com/#/softwarecenter</nowiki> SUPPORT PACKAGES & PATCHES By Category SAP IN-MEMORY (SAP HANA ) HANA PLATFORM EDITION SAP HANA PLATFORM EDITION SAP HANA PLATFORM EDITION 2.0 SAP HANA CLIENT 2.0

=== Installation ===
hdblcm nach installieren:
https://me.sap.com/notes/2651885<syntaxhighlight lang="bash">
<install_media>/SAP_HANA_DATABASE/hdblcm --sid=<SID> --action=update --components=hdblcm --ignore=check_signature_file --verify_signature=off
</syntaxhighlight>

=== Konfiguration===

===SAP HANA Update===
Aktuelle Version als <SID>ADM mit "HDB version" anzeigen lassen

IMDB_CLIENT<VERSION> und IMDB_SERVER<VERSION> von SAP herunterladen und nach /usr/sap/SUM kopieren
<syntaxhighlight lang="bash">
sudo su - root
cd /usr/sap/SUM
chmod +x SAPCAR && ./SAPCAR -xvf IMDB_CLIENT*
chmod +x SAPCAR && ./SAPCAR -xvf IMDB_SERVER*
cd /hana/shared/*/hdblcm
./hdblcm --action=update --ignore=check_signature_file --verify_signature=off --lss_trust_unsigned_components
Enter directory root to search for components: /usr/sap/SUM/
</syntaxhighlight>
Mit "1" alles updaten und weiter den Anweisungen folgen.

===SAP HANA Sizing Report===
Quelle: [https://blogs.sap.com/2021/05/06/how-to-install-run-the-abap-on-hana-sizing-report-sap-note-1872170-a-step-by-step-guide/ HANA Sizing Report]
# Note updaten. Prüfen, ob 2810633 einbaubar ist, ansonsten 2504480 einbauen. Wenn beide nicht einbaubar sind ist das System zu alt oder bereits aktuell.
# SA38 aufrufen und /SDF/HDB_SIZING ausführen
# "Number of parallel dialog processes" auf 3 setzen
# Ausführen oder als Hintergrundjob einplanen

=== Kennwort ändern ===
<syntaxhighlight lang="bash">
ALTER USER <USER> PASSWORD "<new_password>";
</syntaxhighlight>
Oder direkt mit dem <SID>adm in der Shell:
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER <USER> PASSWORD <PW>"
</syntaxhighlight>
Hinweise:
* für System@TENANT und SYSTEM@SYSTEMDB muss der HDBUSERSTORE für das Backupscript aktualisiert werden und für die Anmeldung im HDB Studio

Lösungsschritte:

Prüfen, welche Keys angelegt sind:

<syntaxhighlight lang="bash">
hdbuserstore list
</syntaxhighlight>
Alten Key löschen (Beispiel: BACKUPKEY für SystemDB):

<syntaxhighlight lang="bash">
hdbuserstore delete BACKUP
</syntaxhighlight>
Neuen Key anlegen und dabei das aktualisierte SYSTEM‑Passwort verwenden:

<syntaxhighlight lang="bash">
hdbuserstore set BACKUP <FQDN>:30013 SYSTEM "<PW>"
</syntaxhighlight>

====SAPABAP1 Kennwort ändern====

Direkt auf Applikationsserver:
<syntaxhighlight lang="bash">
hdbsql -n <FQDN>:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPABAP1 PASSWORD <PW>"
hdbuserstore set default <FQDN>:30013@<TENANT> SAPABAP1 "<PW>"
</syntaxhighlight>
Oder im HDB Studio:
<syntaxhighlight lang="bash">
ALTER USER SAPABAP1 PASSWORD "<PW>";
</syntaxhighlight>
[https://me.sap.com/notes/0002750829 2750829 - Ändern des SAPABAP1-Schemakennworts in SAP HANA]
Anschließend siehe "SAP HANA HDB Userstore neu setzen"

==== SAPDBCTRL (SAP Host Agent) ====
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d SYSTEMDB -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
</syntaxhighlight>
Dann als root fortfahren.
Für SystemDB, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname SYSTEMDB@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>
Für Tenant, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname IMP@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>

=== Encryption root keys backup password setzen ===
<syntaxhighlight lang="bash">
ALTER SYSTEM SET ENCRYPTION ROOT KEYS BACKUP PASSWORD "<PW>";
</syntaxhighlight>
===SAP HANA HDB Userstore neu setzen===
<syntaxhighlight lang="bash">
hdbuserstore SET DEFAULT <FQDN:30013>@TENANT SAPABAP1 'PASSWORD'
</syntaxhighlight>Hinweis: Auch wenn Kennwörter geändert werden, wie z.B. vom Backupuser, dann muss der hdbuserstore neu gesetzt werden.<syntaxhighlight lang="bash">
hdbuserstore SET BACKUP <FQDN>:30013@SYSTEMDB BACKUP_OPERATOR 'PASSWORD'
</syntaxhighlight>

===Primary/Secondary Schwenk===
ACHTUNG!! NUR CODESCHNIPSEL!!! Anleitung unvollständig
<syntaxhighlight lang="bash">
Primary
hdbnsutil -sr_enable --name=S4PDB1

Secondary
hdbnsutil -sr_unregister
hdbnsutil -sr_register --remoteHost=simas4pdb --remoteInstance=00 --replicationMode=sync --operationMode=logreplay --name=S4PDB2 --force_full_replica --online
HDB start
</syntaxhighlight>

===Test===
<syntaxhighlight lang="bash">
HDB info
HDB version
</syntaxhighlight>

=== SAP User anzeigen lassen ===
<syntaxhighlight lang="sql">
hdbsql -n localhost:30013 -d <SID>-u SAPABAP1 -p '<PW>' "SELECT BNAME, USTYP FROM SAPABAP1.USR02 WHERE MANDT = '000'"
</syntaxhighlight>

=== SAP User kopieren ===
Dieser Befehl kopiert den User von 000 nach 210
<syntaxhighlight lang="sql">
hdbsql -n localhost:30013 -d <SID>-u SAPABAP1 -p '<PW>' "DELETE FROM SAPABAP1.USR02 WHERE MANDT = '210' AND BNAME = '<USER>'"; INSERT INTO SAPABAP1.USR02 (MANDT, BNAME, BCODE, GLTGV, GLTGB, USTYP, CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT, TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2, BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4, BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN, CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE, RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE, PWDSALTEDHASH, SECURITY_POLICY ) SELECT '210', BNAME, BCODE, GLTGV, GLTGB, USTYP, CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT, TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2, BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4, BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN, CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE, RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE, PWDSALTEDHASH, SECURITY_POLICY FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<USER>';"
</syntaxhighlight>

=== Passwort Hash eines einzelnen Users kopieren ===
<syntaxhighlight lang="sql">
UPDATE SAPABAP1.USR02
SET PWDSALTEDHASH = (SELECT PWDSALTEDHASH FROM SAPABAP1.USR02 WHERE MANDT = '330' AND BNAME = 'NAME'),
PASSCODE = (SELECT PASSCODE FROM SAPABAP1.USR02 WHERE MANDT = '330' AND BNAME = 'NAME')
WHERE MANDT = '331'
AND BNAME = 'NAME';
COMMIT;
</syntaxhighlight>

=== Passwort Hashes kompletter Mandanten kopieren ===
<syntaxhighlight lang="sql">
UPDATE SAPABAP1.USR02 T_ZIEL
SET T_ZIEL.PWDSALTEDHASH = (
SELECT T_QUELLE.PWDSALTEDHASH
FROM SAPABAP1.USR02 T_QUELLE
WHERE T_QUELLE.MANDT = '300'
AND T_QUELLE.BNAME = T_ZIEL.BNAME
)
WHERE T_ZIEL.MANDT = '301'
AND EXISTS (
SELECT 1
FROM SAPABAP1.USR02 T_CHECK
WHERE T_CHECK.MANDT = '300'
AND T_CHECK.BNAME = T_ZIEL.BNAME
);
COMMIT;
</syntaxhighlight>

===Störungen===
====Tabelle BALDAT====
in der SLG2 je Mandant alles löschen und den Job SBAL_DELETE einplanen
<syntaxhighlight lang="bash">
</syntaxhighlight>
====Transparent Huge Pages (THP) are activated or not readable====
HANA Cockpit -> SystemDB -> Database Overview -> Alerting and Diagnostics -> Alert Definitions
ID 116 auswählen Transparent Huge Pages status ( ID 116)
Edit und den Schalter "Schedule Active" auf "no" setzen und "save"
===Codeschnipsel===
SELECT * FROM SYS.M_TABLES where SCHEMA_NAME ='SAPHANADB' AND TABLE_NAME = 'LMSEMAPHORE';

===Nützliche Links===

VLLm

2026-02-09T20:02:15Z

Hendrik: /* Ausführen */

=== Beschreibung ===
Docker normal installieren

=== Download ===
Normal (ROCm)
<syntaxhighlight lang="bash" line="1">
docker pull rocm/vllm-dev:nightly
</syntaxhighlight>
gfx906
<syntaxhighlight lang="bash" line="1">
docker pull nalanzeyu/vllm-gfx906
</syntaxhighlight>

=== Ausführen ===
Variante 1:<syntaxhighlight lang="bash" line="1">
docker run -it --rm --shm-size=8g --device=/dev/kfd --device=/dev/dri \
--group-add video -p 8086:8000 \
-v /mnt/share/models:/models \
nalanzeyu/vllm-gfx906 \
vllm serve /models/Qwen3-Coder-30B-A3B-Instruct-AWQ-4bit --served-model-name Homelab --max-model-len 30000 --enable-auto-tool-choice --tool-call-parser hermes
</syntaxhighlight>Variante 2, getestet 18.12.2025:<syntaxhighlight lang="bash">
sudo docker run -it --rm --network=host --group-add=video --ipc=host --cap-add=SYS_PTRACE --security-opt seccomp=unconfined --device /dev/kfd --device /dev/dri -v /home/hendrik/.lmstudio/models/:/app/models -e HF_HOME="/app/models" -e HF_TOKEN="<TOKEN>" -e NCCL_P2P_DISABLE=1 -e VLLM_CUSTOM_OPS=all -e VLLM_ROCM_USE_AITER=0 -e SAFETENSORS_FAST_GPU=1 -e PYTORCH_TUNABLEOP_ENABLED=1 rocm/vllm-dev:nightly

</syntaxhighlight>Für gfx1201:<syntaxhighlight lang="bash">
sudo docker run -it --rm --network=host \
--group-add=video --ipc=host --cap-add=SYS_PTRACE \
--security-opt seccomp=unconfined --device /dev/kfd \
--device /dev/dri \
-v /home/hendrik/.lmstudio/models/:/app/models \
-e HF_HOME="/app/models" \
-e HF_TOKEN="<TOKEN>" \
-e NCCL_P2P_DISABLE=1 \
-e VLLM_CUSTOM_OPS=all \
-e VLLM_ROCM_USE_AITER=0 \
-e SAFETENSORS_FAST_GPU=1 \
-e PYTORCH_TUNABLEOP_ENABLED=1
kyuz0/vllm-therock-gfx1201
</syntaxhighlight>

Ohne Tensor Parallism:<syntaxhighlight lang="bash">
vllm serve Qwen/Qwen3-VL-8B-Thinking --served-model-name Homelab --max_model_len 4096 --enable-auto-tool-choice --tool-call-parser hermes --reasoning-parser qwen3
</syntaxhighlight>Mit:<syntaxhighlight lang="bash">
vllm serve Qwen/Qwen3-VL-8B-Thinking --served-model-name Homelab --tp 2 --max_model_len 4096 --enable-auto-tool-choice --tool-call-parser hermes --reasoning-parser qwen3
</syntaxhighlight>
Benchmark:
<syntaxhighlight lang="bash">
vllm bench serve --num-prompts 1 --dataset-name=random --input-len 512 --output-len 128 --model Qwen/Qwen3-4B-Instruct-2507-FP8
</syntaxhighlight>

=== Test ===
=== Bekannte Probleme ===
=== Nützliche Links ===
* [https://github.com/ggml-org/llama.cpp Offizielles llama.cpp Repository]
* [https://github.com/ggml-org/llama.cpp/blob/master/examples/server/README.md llama-server Dokumentation]
* [https://github.com/ggml-org/llama.cpp/discussions llama.cpp Discussions]
* [https://huggingface.co/models?library=gguf GGUF Modelle auf Hugging Face]
* [https://docs.rocm.com/ ROCm Dokumentation]
* [https://vulkan.lunarg.com/doc/sdk Vulkan SDK Dokumentation]

SAP HANA

2026-02-05T16:05:03Z

Hendrik: /* SAP User kopieren */

=== Beschreibung ===
SAP HANA ist eine In-Memory-Datenbank-Plattform, die von SAP entwickelt wurde. Sie ermöglicht es Unternehmen große Datenmengen in Echtzeit zu verarbeiten und zu analysieren. Die Plattform unterstützt sowohl transaktionale als auch analytische Anwendungen und ermöglicht es den Benutzern auf umfassende Echtzeit-Analysen zuzugreifen.

Die In-Memory-Technologie von SAP HANA sorgt dafür, dass Daten in einem schnellen Zugriffsspeicher gespeichert werden, anstatt auf langsameren Festplatten. Dies führt zu schnelleren Datenzugriffszeiten und ermöglicht es Unternehmen Echtzeit-Transaktionen und Analysen durchzuführen.

SAP HANA bietet auch eine Vielzahl von Tools und Anwendungen um Unternehmen bei der Verwaltung von Daten zu unterstützen. Dazu gehören Tools zur Datenintegration, Datenbereinigung, Datenmodellierung und Analyse. Es unterstützt auch eine Vielzahl von Programmiersprachen und Entwicklungsplattformen, einschließlich SAPUI5, Java, Python uvm.
=== Download ===
[https://me.sap.com/softwarecenter/support/index SAP Downloadportal]

<nowiki>https://launchpad.support.sap.com/#/softwarecenter</nowiki> SUPPORT PACKAGES & PATCHES By Category SAP IN-MEMORY (SAP HANA ) HANA PLATFORM EDITION SAP HANA PLATFORM EDITION SAP HANA PLATFORM EDITION 2.0 SAP HANA CLIENT 2.0

=== Installation ===
hdblcm nach installieren:
https://me.sap.com/notes/2651885<syntaxhighlight lang="bash">
<install_media>/SAP_HANA_DATABASE/hdblcm --sid=<SID> --action=update --components=hdblcm --ignore=check_signature_file --verify_signature=off
</syntaxhighlight>

=== Konfiguration===

===SAP HANA Update===
Aktuelle Version als <SID>ADM mit "HDB version" anzeigen lassen

IMDB_CLIENT<VERSION> und IMDB_SERVER<VERSION> von SAP herunterladen und nach /usr/sap/SUM kopieren
<syntaxhighlight lang="bash">
sudo su - root
cd /usr/sap/SUM
chmod +x SAPCAR && ./SAPCAR -xvf IMDB_CLIENT*
chmod +x SAPCAR && ./SAPCAR -xvf IMDB_SERVER*
cd /hana/shared/*/hdblcm
./hdblcm --action=update --ignore=check_signature_file --verify_signature=off --lss_trust_unsigned_components
Enter directory root to search for components: /usr/sap/SUM/
</syntaxhighlight>
Mit "1" alles updaten und weiter den Anweisungen folgen.

===SAP HANA Sizing Report===
Quelle: [https://blogs.sap.com/2021/05/06/how-to-install-run-the-abap-on-hana-sizing-report-sap-note-1872170-a-step-by-step-guide/ HANA Sizing Report]
# Note updaten. Prüfen, ob 2810633 einbaubar ist, ansonsten 2504480 einbauen. Wenn beide nicht einbaubar sind ist das System zu alt oder bereits aktuell.
# SA38 aufrufen und /SDF/HDB_SIZING ausführen
# "Number of parallel dialog processes" auf 3 setzen
# Ausführen oder als Hintergrundjob einplanen

=== Kennwort ändern ===
<syntaxhighlight lang="bash">
ALTER USER <USER> PASSWORD "<new_password>";
</syntaxhighlight>
Oder direkt mit dem <SID>adm in der Shell:
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER <USER> PASSWORD <PW>"
</syntaxhighlight>
Hinweise:
* für System@TENANT und SYSTEM@SYSTEMDB muss der HDBUSERSTORE für das Backupscript aktualisiert werden und für die Anmeldung im HDB Studio

Lösungsschritte:

Prüfen, welche Keys angelegt sind:

<syntaxhighlight lang="bash">
hdbuserstore list
</syntaxhighlight>
Alten Key löschen (Beispiel: BACKUPKEY für SystemDB):

<syntaxhighlight lang="bash">
hdbuserstore delete BACKUP
</syntaxhighlight>
Neuen Key anlegen und dabei das aktualisierte SYSTEM‑Passwort verwenden:

<syntaxhighlight lang="bash">
hdbuserstore set BACKUP <FQDN>:30013 SYSTEM "<PW>"
</syntaxhighlight>

====SAPABAP1 Kennwort ändern====

Direkt auf Applikationsserver:
<syntaxhighlight lang="bash">
hdbsql -n <FQDN>:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPABAP1 PASSWORD <PW>"
hdbuserstore set default <FQDN>:30013@<TENANT> SAPABAP1 "<PW>"
</syntaxhighlight>
Oder im HDB Studio:
<syntaxhighlight lang="bash">
ALTER USER SAPABAP1 PASSWORD "<PW>";
</syntaxhighlight>
[https://me.sap.com/notes/0002750829 2750829 - Ändern des SAPABAP1-Schemakennworts in SAP HANA]
Anschließend siehe "SAP HANA HDB Userstore neu setzen"

==== SAPDBCTRL (SAP Host Agent) ====
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d SYSTEMDB -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
</syntaxhighlight>
Dann als root fortfahren.
Für SystemDB, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname SYSTEMDB@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>
Für Tenant, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname IMP@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>

=== Encryption root keys backup password setzen ===
<syntaxhighlight lang="bash">
ALTER SYSTEM SET ENCRYPTION ROOT KEYS BACKUP PASSWORD "<PW>";
</syntaxhighlight>
===SAP HANA HDB Userstore neu setzen===
<syntaxhighlight lang="bash">
hdbuserstore SET DEFAULT <FQDN:30013>@TENANT SAPABAP1 'PASSWORD'
</syntaxhighlight>Hinweis: Auch wenn Kennwörter geändert werden, wie z.B. vom Backupuser, dann muss der hdbuserstore neu gesetzt werden.<syntaxhighlight lang="bash">
hdbuserstore SET BACKUP <FQDN>:30013@SYSTEMDB BACKUP_OPERATOR 'PASSWORD'
</syntaxhighlight>

===Primary/Secondary Schwenk===
ACHTUNG!! NUR CODESCHNIPSEL!!! Anleitung unvollständig
<syntaxhighlight lang="bash">
Primary
hdbnsutil -sr_enable --name=S4PDB1

Secondary
hdbnsutil -sr_unregister
hdbnsutil -sr_register --remoteHost=simas4pdb --remoteInstance=00 --replicationMode=sync --operationMode=logreplay --name=S4PDB2 --force_full_replica --online
HDB start
</syntaxhighlight>

===Test===
<syntaxhighlight lang="bash">
HDB info
HDB version
</syntaxhighlight>

=== SAP User anzeigen lassen ===
<syntaxhighlight lang="sql">
hdbsql -n localhost:30013 -d <SID>-u SAPABAP1 -p '<PW>' "SELECT BNAME, USTYP FROM SAPABAP1.USR02 WHERE MANDT = '000'"
</syntaxhighlight>

=== SAP User kopieren ===
Dieser Befehl kopiert den User von 000 nach 210
<syntaxhighlight lang="sql">
hdbsql -n localhost:30013 -d <SID>-u SAPABAP1 -p '<PW>' "DELETE FROM SAPABAP1.USR02 WHERE MANDT = '210' AND BNAME = '<USER>'"; INSERT INTO SAPABAP1.USR02 (MANDT, BNAME, BCODE, GLTGV, GLTGB, USTYP, CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT, TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2, BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4, BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN, CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE, RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE, PWDSALTEDHASH, SECURITY_POLICY ) SELECT '210', BNAME, BCODE, GLTGV, GLTGB, USTYP, CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT, TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2, BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4, BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN, CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE, RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE, PWDSALTEDHASH, SECURITY_POLICY FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<USER>';"
</syntaxhighlight>

===Störungen===
====Tabelle BALDAT====
in der SLG2 je Mandant alles löschen und den Job SBAL_DELETE einplanen
<syntaxhighlight lang="bash">
</syntaxhighlight>
====Transparent Huge Pages (THP) are activated or not readable====
HANA Cockpit -> SystemDB -> Database Overview -> Alerting and Diagnostics -> Alert Definitions
ID 116 auswählen Transparent Huge Pages status ( ID 116)
Edit und den Schalter "Schedule Active" auf "no" setzen und "save"
===Codeschnipsel===
SELECT * FROM SYS.M_TABLES where SCHEMA_NAME ='SAPHANADB' AND TABLE_NAME = 'LMSEMAPHORE';

===Nützliche Links===

SAP HANA

2026-02-04T18:54:44Z

Hendrik: /* SAP HANA Update */

=== Beschreibung ===
SAP HANA ist eine In-Memory-Datenbank-Plattform, die von SAP entwickelt wurde. Sie ermöglicht es Unternehmen große Datenmengen in Echtzeit zu verarbeiten und zu analysieren. Die Plattform unterstützt sowohl transaktionale als auch analytische Anwendungen und ermöglicht es den Benutzern auf umfassende Echtzeit-Analysen zuzugreifen.

Die In-Memory-Technologie von SAP HANA sorgt dafür, dass Daten in einem schnellen Zugriffsspeicher gespeichert werden, anstatt auf langsameren Festplatten. Dies führt zu schnelleren Datenzugriffszeiten und ermöglicht es Unternehmen Echtzeit-Transaktionen und Analysen durchzuführen.

SAP HANA bietet auch eine Vielzahl von Tools und Anwendungen um Unternehmen bei der Verwaltung von Daten zu unterstützen. Dazu gehören Tools zur Datenintegration, Datenbereinigung, Datenmodellierung und Analyse. Es unterstützt auch eine Vielzahl von Programmiersprachen und Entwicklungsplattformen, einschließlich SAPUI5, Java, Python uvm.
=== Download ===
[https://me.sap.com/softwarecenter/support/index SAP Downloadportal]

<nowiki>https://launchpad.support.sap.com/#/softwarecenter</nowiki> SUPPORT PACKAGES & PATCHES By Category SAP IN-MEMORY (SAP HANA ) HANA PLATFORM EDITION SAP HANA PLATFORM EDITION SAP HANA PLATFORM EDITION 2.0 SAP HANA CLIENT 2.0

=== Installation ===
hdblcm nach installieren:
https://me.sap.com/notes/2651885<syntaxhighlight lang="bash">
<install_media>/SAP_HANA_DATABASE/hdblcm --sid=<SID> --action=update --components=hdblcm --ignore=check_signature_file --verify_signature=off
</syntaxhighlight>

=== Konfiguration===

===SAP HANA Update===
Aktuelle Version als <SID>ADM mit "HDB version" anzeigen lassen

IMDB_CLIENT<VERSION> und IMDB_SERVER<VERSION> von SAP herunterladen und nach /usr/sap/SUM kopieren
<syntaxhighlight lang="bash">
sudo su - root
cd /usr/sap/SUM
chmod +x SAPCAR && ./SAPCAR -xvf IMDB_CLIENT*
chmod +x SAPCAR && ./SAPCAR -xvf IMDB_SERVER*
cd /hana/shared/*/hdblcm
./hdblcm --action=update --ignore=check_signature_file --verify_signature=off --lss_trust_unsigned_components
Enter directory root to search for components: /usr/sap/SUM/
</syntaxhighlight>
Mit "1" alles updaten und weiter den Anweisungen folgen.

===SAP HANA Sizing Report===
Quelle: [https://blogs.sap.com/2021/05/06/how-to-install-run-the-abap-on-hana-sizing-report-sap-note-1872170-a-step-by-step-guide/ HANA Sizing Report]
# Note updaten. Prüfen, ob 2810633 einbaubar ist, ansonsten 2504480 einbauen. Wenn beide nicht einbaubar sind ist das System zu alt oder bereits aktuell.
# SA38 aufrufen und /SDF/HDB_SIZING ausführen
# "Number of parallel dialog processes" auf 3 setzen
# Ausführen oder als Hintergrundjob einplanen

=== Kennwort ändern ===
<syntaxhighlight lang="bash">
ALTER USER <USER> PASSWORD "<new_password>";
</syntaxhighlight>
Oder direkt mit dem <SID>adm in der Shell:
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER <USER> PASSWORD <PW>"
</syntaxhighlight>
Hinweise:
* für System@TENANT und SYSTEM@SYSTEMDB muss der HDBUSERSTORE für das Backupscript aktualisiert werden und für die Anmeldung im HDB Studio

Lösungsschritte:

Prüfen, welche Keys angelegt sind:

<syntaxhighlight lang="bash">
hdbuserstore list
</syntaxhighlight>
Alten Key löschen (Beispiel: BACKUPKEY für SystemDB):

<syntaxhighlight lang="bash">
hdbuserstore delete BACKUP
</syntaxhighlight>
Neuen Key anlegen und dabei das aktualisierte SYSTEM‑Passwort verwenden:

<syntaxhighlight lang="bash">
hdbuserstore set BACKUP <FQDN>:30013 SYSTEM "<PW>"
</syntaxhighlight>

====SAPABAP1 Kennwort ändern====

Direkt auf Applikationsserver:
<syntaxhighlight lang="bash">
hdbsql -n <FQDN>:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPABAP1 PASSWORD <PW>"
hdbuserstore set default <FQDN>:30013@<TENANT> SAPABAP1 "<PW>"
</syntaxhighlight>
Oder im HDB Studio:
<syntaxhighlight lang="bash">
ALTER USER SAPABAP1 PASSWORD "<PW>";
</syntaxhighlight>
[https://me.sap.com/notes/0002750829 2750829 - Ändern des SAPABAP1-Schemakennworts in SAP HANA]
Anschließend siehe "SAP HANA HDB Userstore neu setzen"

==== SAPDBCTRL (SAP Host Agent) ====
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d SYSTEMDB -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
</syntaxhighlight>
Dann als root fortfahren.
Für SystemDB, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname SYSTEMDB@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>
Für Tenant, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname IMP@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>

=== Encryption root keys backup password setzen ===
<syntaxhighlight lang="bash">
ALTER SYSTEM SET ENCRYPTION ROOT KEYS BACKUP PASSWORD "<PW>";
</syntaxhighlight>
===SAP HANA HDB Userstore neu setzen===
<syntaxhighlight lang="bash">
hdbuserstore SET DEFAULT <FQDN:30013>@TENANT SAPABAP1 'PASSWORD'
</syntaxhighlight>Hinweis: Auch wenn Kennwörter geändert werden, wie z.B. vom Backupuser, dann muss der hdbuserstore neu gesetzt werden.<syntaxhighlight lang="bash">
hdbuserstore SET BACKUP <FQDN>:30013@SYSTEMDB BACKUP_OPERATOR 'PASSWORD'
</syntaxhighlight>

===Primary/Secondary Schwenk===
ACHTUNG!! NUR CODESCHNIPSEL!!! Anleitung unvollständig
<syntaxhighlight lang="bash">
Primary
hdbnsutil -sr_enable --name=S4PDB1

Secondary
hdbnsutil -sr_unregister
hdbnsutil -sr_register --remoteHost=simas4pdb --remoteInstance=00 --replicationMode=sync --operationMode=logreplay --name=S4PDB2 --force_full_replica --online
HDB start
</syntaxhighlight>

===Test===
<syntaxhighlight lang="bash">
HDB info
HDB version
</syntaxhighlight>

=== SAP User anzeigen lassen ===
<syntaxhighlight lang="sql">
hdbsql -n localhost:30013 -d <SID>-u SAPABAP1 -p '<PW>' "SELECT BNAME, USTYP FROM SAPABAP1.USR02 WHERE MANDT = '000'"
</syntaxhighlight>

=== SAP User kopieren ===
Dieser Befehl kopiert den User von 000 nach 210
<syntaxhighlight lang="sql">
hdbsql -n localhost:30013 -d <SID>-u SAPABAP1 -p '<PW>' "DELETE FROM SAPABAP1.USR02 WHERE MANDT = '210' AND BNAME = '<USER>'; INSERT INTO SAPABAP1.USR02 (MANDT, BNAME, BCODE, GLTGV, GLTGB, USTYP, CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT, TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2, BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4, BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN, CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE, RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE, PWDSALTEDHASH, SECURITY_POLICY ) SELECT '210', BNAME, BCODE, GLTGV, GLTGB, USTYP, CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT, TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2, BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4, BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN, CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE, RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE, PWDSALTEDHASH, SECURITY_POLICY FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<USER>';"
</syntaxhighlight>

===Störungen===
====Tabelle BALDAT====
in der SLG2 je Mandant alles löschen und den Job SBAL_DELETE einplanen
<syntaxhighlight lang="bash">
</syntaxhighlight>
====Transparent Huge Pages (THP) are activated or not readable====
HANA Cockpit -> SystemDB -> Database Overview -> Alerting and Diagnostics -> Alert Definitions
ID 116 auswählen Transparent Huge Pages status ( ID 116)
Edit und den Schalter "Schedule Active" auf "no" setzen und "save"
===Codeschnipsel===
SELECT * FROM SYS.M_TABLES where SCHEMA_NAME ='SAPHANADB' AND TABLE_NAME = 'LMSEMAPHORE';

===Nützliche Links===

SAP HANA

2026-01-21T09:29:24Z

Hendrik:

=== Beschreibung ===
SAP HANA ist eine In-Memory-Datenbank-Plattform, die von SAP entwickelt wurde. Sie ermöglicht es Unternehmen große Datenmengen in Echtzeit zu verarbeiten und zu analysieren. Die Plattform unterstützt sowohl transaktionale als auch analytische Anwendungen und ermöglicht es den Benutzern auf umfassende Echtzeit-Analysen zuzugreifen.

Die In-Memory-Technologie von SAP HANA sorgt dafür, dass Daten in einem schnellen Zugriffsspeicher gespeichert werden, anstatt auf langsameren Festplatten. Dies führt zu schnelleren Datenzugriffszeiten und ermöglicht es Unternehmen Echtzeit-Transaktionen und Analysen durchzuführen.

SAP HANA bietet auch eine Vielzahl von Tools und Anwendungen um Unternehmen bei der Verwaltung von Daten zu unterstützen. Dazu gehören Tools zur Datenintegration, Datenbereinigung, Datenmodellierung und Analyse. Es unterstützt auch eine Vielzahl von Programmiersprachen und Entwicklungsplattformen, einschließlich SAPUI5, Java, Python uvm.
=== Download ===
[https://me.sap.com/softwarecenter/support/index SAP Downloadportal]

<nowiki>https://launchpad.support.sap.com/#/softwarecenter</nowiki> SUPPORT PACKAGES & PATCHES By Category SAP IN-MEMORY (SAP HANA ) HANA PLATFORM EDITION SAP HANA PLATFORM EDITION SAP HANA PLATFORM EDITION 2.0 SAP HANA CLIENT 2.0

=== Installation ===
hdblcm nach installieren:
https://me.sap.com/notes/2651885<syntaxhighlight lang="bash">
<install_media>/SAP_HANA_DATABASE/hdblcm --sid=<SID> --action=update --components=hdblcm --ignore=check_signature_file --verify_signature=off
</syntaxhighlight>

=== Konfiguration===

===SAP HANA Update===
Aktuelle Version als <SID>ADM mit "HDB version" anzeigen lassen

IMDB_CLIENT<VERSION> und IMDB_SERVER<VERSION> von SAP herunterladen und nach /usr/sap/SUM kopieren
<syntaxhighlight lang="bash">
sudo su - root
cd /usr/sap/SUM
SAPCAR -xvf IMDB_CLIENT*
./SAPCAR -xvf IMDB_SERVER*
cd /hana/shared/*/hdblcm
./hdblcm --action=update --ignore=check_signature_file --verify_signature=off --lss_trust_unsigned_components
Enter directory root to search for components: /usr/sap/SUM/
</syntaxhighlight>
Mit "1" alles updaten und weiter den Anweisungen folgen.

===SAP HANA Sizing Report===
Quelle: [https://blogs.sap.com/2021/05/06/how-to-install-run-the-abap-on-hana-sizing-report-sap-note-1872170-a-step-by-step-guide/ HANA Sizing Report]
# Note updaten. Prüfen, ob 2810633 einbaubar ist, ansonsten 2504480 einbauen. Wenn beide nicht einbaubar sind ist das System zu alt oder bereits aktuell.
# SA38 aufrufen und /SDF/HDB_SIZING ausführen
# "Number of parallel dialog processes" auf 3 setzen
# Ausführen oder als Hintergrundjob einplanen

=== Kennwort ändern ===
<syntaxhighlight lang="bash">
ALTER USER <USER> PASSWORD "<new_password>";
</syntaxhighlight>
Oder direkt mit dem <SID>adm in der Shell:
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER <USER> PASSWORD <PW>"
</syntaxhighlight>
Hinweise:
* für System@TENANT und SYSTEM@SYSTEMDB muss der HDBUSERSTORE für das Backupscript aktualisiert werden und für die Anmeldung im HDB Studio

Lösungsschritte:

Prüfen, welche Keys angelegt sind:

<syntaxhighlight lang="bash">
hdbuserstore list
</syntaxhighlight>
Alten Key löschen (Beispiel: BACKUPKEY für SystemDB):

<syntaxhighlight lang="bash">
hdbuserstore delete BACKUP
</syntaxhighlight>
Neuen Key anlegen und dabei das aktualisierte SYSTEM‑Passwort verwenden:

<syntaxhighlight lang="bash">
hdbuserstore set BACKUP <FQDN>:30013 SYSTEM "<PW>"
</syntaxhighlight>

====SAPABAP1 Kennwort ändern====

Direkt auf Applikationsserver:
<syntaxhighlight lang="bash">
hdbsql -n <FQDN>:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPABAP1 PASSWORD <PW>"
hdbuserstore set default <FQDN>:30013@<TENANT> SAPABAP1 "<PW>"
</syntaxhighlight>
Oder im HDB Studio:
<syntaxhighlight lang="bash">
ALTER USER SAPABAP1 PASSWORD "<PW>";
</syntaxhighlight>
[https://me.sap.com/notes/0002750829 2750829 - Ändern des SAPABAP1-Schemakennworts in SAP HANA]
Anschließend siehe "SAP HANA HDB Userstore neu setzen"

==== SAPDBCTRL (SAP Host Agent) ====
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d SYSTEMDB -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
</syntaxhighlight>
Dann als root fortfahren.
Für SystemDB, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname SYSTEMDB@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>
Für Tenant, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname IMP@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>

=== Encryption root keys backup password setzen ===
<syntaxhighlight lang="bash">
ALTER SYSTEM SET ENCRYPTION ROOT KEYS BACKUP PASSWORD "<PW>";
</syntaxhighlight>
===SAP HANA HDB Userstore neu setzen===
<syntaxhighlight lang="bash">
hdbuserstore SET DEFAULT <FQDN:30013>@TENANT SAPABAP1 'PASSWORD'
</syntaxhighlight>Hinweis: Auch wenn Kennwörter geändert werden, wie z.B. vom Backupuser, dann muss der hdbuserstore neu gesetzt werden.<syntaxhighlight lang="bash">
hdbuserstore SET BACKUP <FQDN>:30013@SYSTEMDB BACKUP_OPERATOR 'PASSWORD'
</syntaxhighlight>

===Primary/Secondary Schwenk===
ACHTUNG!! NUR CODESCHNIPSEL!!! Anleitung unvollständig
<syntaxhighlight lang="bash">
Primary
hdbnsutil -sr_enable --name=S4PDB1

Secondary
hdbnsutil -sr_unregister
hdbnsutil -sr_register --remoteHost=simas4pdb --remoteInstance=00 --replicationMode=sync --operationMode=logreplay --name=S4PDB2 --force_full_replica --online
HDB start
</syntaxhighlight>

===Test===
<syntaxhighlight lang="bash">
HDB info
HDB version
</syntaxhighlight>

=== SAP User anzeigen lassen ===
<syntaxhighlight lang="sql">
hdbsql -n localhost:30013 -d <SID>-u SAPABAP1 -p '<PW>' "SELECT BNAME, USTYP FROM SAPABAP1.USR02 WHERE MANDT = '000'"
</syntaxhighlight>

=== SAP User kopieren ===
Dieser Befehl kopiert den User von 000 nach 210
<syntaxhighlight lang="sql">
hdbsql -n localhost:30013 -d <SID>-u SAPABAP1 -p '<PW>' "DELETE FROM SAPABAP1.USR02 WHERE MANDT = '210' AND BNAME = '<USER>'; INSERT INTO SAPABAP1.USR02 (MANDT, BNAME, BCODE, GLTGV, GLTGB, USTYP, CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT, TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2, BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4, BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN, CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE, RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE, PWDSALTEDHASH, SECURITY_POLICY ) SELECT '210', BNAME, BCODE, GLTGV, GLTGB, USTYP, CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT, TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2, BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4, BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN, CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE, RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE, PWDSALTEDHASH, SECURITY_POLICY FROM SAPABAP1.USR02 WHERE MANDT = '000' AND BNAME = '<USER>';"
</syntaxhighlight>

===Störungen===
====Tabelle BALDAT====
in der SLG2 je Mandant alles löschen und den Job SBAL_DELETE einplanen
<syntaxhighlight lang="bash">
</syntaxhighlight>
====Transparent Huge Pages (THP) are activated or not readable====
HANA Cockpit -> SystemDB -> Database Overview -> Alerting and Diagnostics -> Alert Definitions
ID 116 auswählen Transparent Huge Pages status ( ID 116)
Edit und den Schalter "Schedule Active" auf "no" setzen und "save"
===Codeschnipsel===
SELECT * FROM SYS.M_TABLES where SCHEMA_NAME ='SAPHANADB' AND TABLE_NAME = 'LMSEMAPHORE';

===Nützliche Links===

SAP HANA

2026-01-21T08:46:40Z

Hendrik: /* SAP User anzeigen lassen */

=== Beschreibung ===
SAP HANA ist eine In-Memory-Datenbank-Plattform, die von SAP entwickelt wurde. Sie ermöglicht es Unternehmen große Datenmengen in Echtzeit zu verarbeiten und zu analysieren. Die Plattform unterstützt sowohl transaktionale als auch analytische Anwendungen und ermöglicht es den Benutzern auf umfassende Echtzeit-Analysen zuzugreifen.

Die In-Memory-Technologie von SAP HANA sorgt dafür, dass Daten in einem schnellen Zugriffsspeicher gespeichert werden, anstatt auf langsameren Festplatten. Dies führt zu schnelleren Datenzugriffszeiten und ermöglicht es Unternehmen Echtzeit-Transaktionen und Analysen durchzuführen.

SAP HANA bietet auch eine Vielzahl von Tools und Anwendungen um Unternehmen bei der Verwaltung von Daten zu unterstützen. Dazu gehören Tools zur Datenintegration, Datenbereinigung, Datenmodellierung und Analyse. Es unterstützt auch eine Vielzahl von Programmiersprachen und Entwicklungsplattformen, einschließlich SAPUI5, Java, Python uvm.
=== Download ===
[https://me.sap.com/softwarecenter/support/index SAP Downloadportal]

<nowiki>https://launchpad.support.sap.com/#/softwarecenter</nowiki> SUPPORT PACKAGES & PATCHES By Category SAP IN-MEMORY (SAP HANA ) HANA PLATFORM EDITION SAP HANA PLATFORM EDITION SAP HANA PLATFORM EDITION 2.0 SAP HANA CLIENT 2.0

=== Installation ===
hdblcm nach installieren:
https://me.sap.com/notes/2651885<syntaxhighlight lang="bash">
<install_media>/SAP_HANA_DATABASE/hdblcm --sid=<SID> --action=update --components=hdblcm --ignore=check_signature_file --verify_signature=off
</syntaxhighlight>

=== Konfiguration===

===SAP HANA Update===
Aktuelle Version als <SID>ADM mit "HDB version" anzeigen lassen

IMDB_CLIENT<VERSION> und IMDB_SERVER<VERSION> von SAP herunterladen und nach /usr/sap/SUM kopieren
<syntaxhighlight lang="bash">
sudo su - root
cd /usr/sap/SUM
SAPCAR -xvf IMDB_CLIENT*
./SAPCAR -xvf IMDB_SERVER*
cd /hana/shared/*/hdblcm
./hdblcm --action=update --ignore=check_signature_file --verify_signature=off --lss_trust_unsigned_components
Enter directory root to search for components: /usr/sap/SUM/
</syntaxhighlight>
Mit "1" alles updaten und weiter den Anweisungen folgen.

===SAP HANA Sizing Report===
Quelle: [https://blogs.sap.com/2021/05/06/how-to-install-run-the-abap-on-hana-sizing-report-sap-note-1872170-a-step-by-step-guide/ HANA Sizing Report]
# Note updaten. Prüfen, ob 2810633 einbaubar ist, ansonsten 2504480 einbauen. Wenn beide nicht einbaubar sind ist das System zu alt oder bereits aktuell.
# SA38 aufrufen und /SDF/HDB_SIZING ausführen
# "Number of parallel dialog processes" auf 3 setzen
# Ausführen oder als Hintergrundjob einplanen

=== Kennwort ändern ===
<syntaxhighlight lang="bash">
ALTER USER <USER> PASSWORD "<new_password>";
</syntaxhighlight>
Oder direkt mit dem <SID>adm in der Shell:
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER <USER> PASSWORD <PW>"
</syntaxhighlight>
Hinweise:
* für System@TENANT und SYSTEM@SYSTEMDB muss der HDBUSERSTORE für das Backupscript aktualisiert werden und für die Anmeldung im HDB Studio

Lösungsschritte:

Prüfen, welche Keys angelegt sind:

<syntaxhighlight lang="bash">
hdbuserstore list
</syntaxhighlight>
Alten Key löschen (Beispiel: BACKUPKEY für SystemDB):

<syntaxhighlight lang="bash">
hdbuserstore delete BACKUP
</syntaxhighlight>
Neuen Key anlegen und dabei das aktualisierte SYSTEM‑Passwort verwenden:

<syntaxhighlight lang="bash">
hdbuserstore set BACKUP <FQDN>:30013 SYSTEM "<PW>"
</syntaxhighlight>

====SAPABAP1 Kennwort ändern====

Direkt auf Applikationsserver:
<syntaxhighlight lang="bash">
hdbsql -n <FQDN>:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPABAP1 PASSWORD <PW>"
hdbuserstore set default <FQDN>:30013@<TENANT> SAPABAP1 "<PW>"
</syntaxhighlight>
Oder im HDB Studio:
<syntaxhighlight lang="bash">
ALTER USER SAPABAP1 PASSWORD "<PW>";
</syntaxhighlight>
[https://me.sap.com/notes/0002750829 2750829 - Ändern des SAPABAP1-Schemakennworts in SAP HANA]
Anschließend siehe "SAP HANA HDB Userstore neu setzen"

==== SAPDBCTRL (SAP Host Agent) ====
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d SYSTEMDB -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
</syntaxhighlight>
Dann als root fortfahren.
Für SystemDB, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname SYSTEMDB@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>
Für Tenant, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname IMP@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>

=== Encryption root keys backup password setzen ===
<syntaxhighlight lang="bash">
ALTER SYSTEM SET ENCRYPTION ROOT KEYS BACKUP PASSWORD "<PW>";
</syntaxhighlight>
===SAP HANA HDB Userstore neu setzen===
<syntaxhighlight lang="bash">
hdbuserstore SET DEFAULT <FQDN:30013>@TENANT SAPABAP1 'PASSWORD'
</syntaxhighlight>Hinweis: Auch wenn Kennwörter geändert werden, wie z.B. vom Backupuser, dann muss der hdbuserstore neu gesetzt werden.<syntaxhighlight lang="bash">
hdbuserstore SET BACKUP <FQDN>:30013@SYSTEMDB BACKUP_OPERATOR 'PASSWORD'
</syntaxhighlight>

===Primary/Secondary Schwenk===
ACHTUNG!! NUR CODESCHNIPSEL!!! Anleitung unvollständig
<syntaxhighlight lang="bash">
Primary
hdbnsutil -sr_enable --name=S4PDB1

Secondary
hdbnsutil -sr_unregister
hdbnsutil -sr_register --remoteHost=simas4pdb --remoteInstance=00 --replicationMode=sync --operationMode=logreplay --name=S4PDB2 --force_full_replica --online
HDB start
</syntaxhighlight>

===Test===
<syntaxhighlight lang="bash">
HDB info
HDB version
</syntaxhighlight>

=== SAP User anzeigen lassen ===
<syntaxhighlight lang="sql">
hdbsql -n localhost:30013 -d <SID>-u SAPABAP1 -p '<PW>' "SELECT BNAME, USTYP FROM SAPABAP1.USR02 WHERE MANDT = '000'"
</syntaxhighlight>

===Störungen===
====Tabelle BALDAT====
in der SLG2 je Mandant alles löschen und den Job SBAL_DELETE einplanen
<syntaxhighlight lang="bash">
</syntaxhighlight>
====Transparent Huge Pages (THP) are activated or not readable====
HANA Cockpit -> SystemDB -> Database Overview -> Alerting and Diagnostics -> Alert Definitions
ID 116 auswählen Transparent Huge Pages status ( ID 116)
Edit und den Schalter "Schedule Active" auf "no" setzen und "save"
===Codeschnipsel===
SELECT * FROM SYS.M_TABLES where SCHEMA_NAME ='SAPHANADB' AND TABLE_NAME = 'LMSEMAPHORE';

===Nützliche Links===

SAP HANA

2026-01-21T08:46:22Z

Hendrik:

=== Beschreibung ===
SAP HANA ist eine In-Memory-Datenbank-Plattform, die von SAP entwickelt wurde. Sie ermöglicht es Unternehmen große Datenmengen in Echtzeit zu verarbeiten und zu analysieren. Die Plattform unterstützt sowohl transaktionale als auch analytische Anwendungen und ermöglicht es den Benutzern auf umfassende Echtzeit-Analysen zuzugreifen.

Die In-Memory-Technologie von SAP HANA sorgt dafür, dass Daten in einem schnellen Zugriffsspeicher gespeichert werden, anstatt auf langsameren Festplatten. Dies führt zu schnelleren Datenzugriffszeiten und ermöglicht es Unternehmen Echtzeit-Transaktionen und Analysen durchzuführen.

SAP HANA bietet auch eine Vielzahl von Tools und Anwendungen um Unternehmen bei der Verwaltung von Daten zu unterstützen. Dazu gehören Tools zur Datenintegration, Datenbereinigung, Datenmodellierung und Analyse. Es unterstützt auch eine Vielzahl von Programmiersprachen und Entwicklungsplattformen, einschließlich SAPUI5, Java, Python uvm.
=== Download ===
[https://me.sap.com/softwarecenter/support/index SAP Downloadportal]

<nowiki>https://launchpad.support.sap.com/#/softwarecenter</nowiki> SUPPORT PACKAGES & PATCHES By Category SAP IN-MEMORY (SAP HANA ) HANA PLATFORM EDITION SAP HANA PLATFORM EDITION SAP HANA PLATFORM EDITION 2.0 SAP HANA CLIENT 2.0

=== Installation ===
hdblcm nach installieren:
https://me.sap.com/notes/2651885<syntaxhighlight lang="bash">
<install_media>/SAP_HANA_DATABASE/hdblcm --sid=<SID> --action=update --components=hdblcm --ignore=check_signature_file --verify_signature=off
</syntaxhighlight>

=== Konfiguration===

===SAP HANA Update===
Aktuelle Version als <SID>ADM mit "HDB version" anzeigen lassen

IMDB_CLIENT<VERSION> und IMDB_SERVER<VERSION> von SAP herunterladen und nach /usr/sap/SUM kopieren
<syntaxhighlight lang="bash">
sudo su - root
cd /usr/sap/SUM
SAPCAR -xvf IMDB_CLIENT*
./SAPCAR -xvf IMDB_SERVER*
cd /hana/shared/*/hdblcm
./hdblcm --action=update --ignore=check_signature_file --verify_signature=off --lss_trust_unsigned_components
Enter directory root to search for components: /usr/sap/SUM/
</syntaxhighlight>
Mit "1" alles updaten und weiter den Anweisungen folgen.

===SAP HANA Sizing Report===
Quelle: [https://blogs.sap.com/2021/05/06/how-to-install-run-the-abap-on-hana-sizing-report-sap-note-1872170-a-step-by-step-guide/ HANA Sizing Report]
# Note updaten. Prüfen, ob 2810633 einbaubar ist, ansonsten 2504480 einbauen. Wenn beide nicht einbaubar sind ist das System zu alt oder bereits aktuell.
# SA38 aufrufen und /SDF/HDB_SIZING ausführen
# "Number of parallel dialog processes" auf 3 setzen
# Ausführen oder als Hintergrundjob einplanen

=== Kennwort ändern ===
<syntaxhighlight lang="bash">
ALTER USER <USER> PASSWORD "<new_password>";
</syntaxhighlight>
Oder direkt mit dem <SID>adm in der Shell:
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER <USER> PASSWORD <PW>"
</syntaxhighlight>
Hinweise:
* für System@TENANT und SYSTEM@SYSTEMDB muss der HDBUSERSTORE für das Backupscript aktualisiert werden und für die Anmeldung im HDB Studio

Lösungsschritte:

Prüfen, welche Keys angelegt sind:

<syntaxhighlight lang="bash">
hdbuserstore list
</syntaxhighlight>
Alten Key löschen (Beispiel: BACKUPKEY für SystemDB):

<syntaxhighlight lang="bash">
hdbuserstore delete BACKUP
</syntaxhighlight>
Neuen Key anlegen und dabei das aktualisierte SYSTEM‑Passwort verwenden:

<syntaxhighlight lang="bash">
hdbuserstore set BACKUP <FQDN>:30013 SYSTEM "<PW>"
</syntaxhighlight>

====SAPABAP1 Kennwort ändern====

Direkt auf Applikationsserver:
<syntaxhighlight lang="bash">
hdbsql -n <FQDN>:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPABAP1 PASSWORD <PW>"
hdbuserstore set default <FQDN>:30013@<TENANT> SAPABAP1 "<PW>"
</syntaxhighlight>
Oder im HDB Studio:
<syntaxhighlight lang="bash">
ALTER USER SAPABAP1 PASSWORD "<PW>";
</syntaxhighlight>
[https://me.sap.com/notes/0002750829 2750829 - Ändern des SAPABAP1-Schemakennworts in SAP HANA]
Anschließend siehe "SAP HANA HDB Userstore neu setzen"

==== SAPDBCTRL (SAP Host Agent) ====
<syntaxhighlight lang="bash">
hdbsql -n localhost:30013 -d SYSTEMDB -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
hdbsql -n localhost:30013 -d <TENANT> -u SYSTEM -p '<PW>' "ALTER USER SAPDBCTRL PASSWORD <PW>"
</syntaxhighlight>
Dann als root fortfahren.
Für SystemDB, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname SYSTEMDB@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>
Für Tenant, bei dbpass Abfrage das Passwort vom SYSTEM-User eingeben:
<syntaxhighlight lang="bash">
/usr/sap/hostctrl/exe/saphostctrl -function SetDatabaseProperty -dbname IMP@IMP -dbtype hdb DBCredentials=SET -dbuser system -dbpass - -dboption Password=<SAPDBCTRLPW>
</syntaxhighlight>

=== Encryption root keys backup password setzen ===
<syntaxhighlight lang="bash">
ALTER SYSTEM SET ENCRYPTION ROOT KEYS BACKUP PASSWORD "<PW>";
</syntaxhighlight>
===SAP HANA HDB Userstore neu setzen===
<syntaxhighlight lang="bash">
hdbuserstore SET DEFAULT <FQDN:30013>@TENANT SAPABAP1 'PASSWORD'
</syntaxhighlight>Hinweis: Auch wenn Kennwörter geändert werden, wie z.B. vom Backupuser, dann muss der hdbuserstore neu gesetzt werden.<syntaxhighlight lang="bash">
hdbuserstore SET BACKUP <FQDN>:30013@SYSTEMDB BACKUP_OPERATOR 'PASSWORD'
</syntaxhighlight>

===Primary/Secondary Schwenk===
ACHTUNG!! NUR CODESCHNIPSEL!!! Anleitung unvollständig
<syntaxhighlight lang="bash">
Primary
hdbnsutil -sr_enable --name=S4PDB1

Secondary
hdbnsutil -sr_unregister
hdbnsutil -sr_register --remoteHost=simas4pdb --remoteInstance=00 --replicationMode=sync --operationMode=logreplay --name=S4PDB2 --force_full_replica --online
HDB start
</syntaxhighlight>

===Test===
<syntaxhighlight lang="bash">
HDB info
HDB version
</syntaxhighlight>

=== SAP User anzeigen lassen ===
<syntaxhighlight lang="sql">
hdbsql -n localhost:30013 -d IMP -u SAPABAP1 -p '<PW>' "SELECT BNAME, USTYP FROM SAPABAP1.USR02 WHERE MANDT = '000'"
</syntaxhighlight>

===Störungen===
====Tabelle BALDAT====
in der SLG2 je Mandant alles löschen und den Job SBAL_DELETE einplanen
<syntaxhighlight lang="bash">
</syntaxhighlight>
====Transparent Huge Pages (THP) are activated or not readable====
HANA Cockpit -> SystemDB -> Database Overview -> Alerting and Diagnostics -> Alert Definitions
ID 116 auswählen Transparent Huge Pages status ( ID 116)
Edit und den Schalter "Schedule Active" auf "no" setzen und "save"
===Codeschnipsel===
SELECT * FROM SYS.M_TABLES where SCHEMA_NAME ='SAPHANADB' AND TABLE_NAME = 'LMSEMAPHORE';

===Nützliche Links===

Llama.cpp

2026-01-19T23:00:23Z

Hendrik: /* ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 */

=== Beschreibung ===
llama.cpp ist eine C/C++ Implementierung für Inference von Large Language Models (LLMs). Es unterstützt verschiedene Backends (CPU, Vulkan, ROCm, CUDA) und ermöglicht das Ausführen von quantisierten Modellen im GGUF-Format.

=== Download ===
<syntaxhighlight lang="bash" line="1">
git clone https://github.com/ggml-org/llama.cpp
cd llama.cpp
</syntaxhighlight>

=== Installation ===
==== Vulkan Build ====
<syntaxhighlight lang="bash" line="1">
sudo dnf install libcurl-devel -y
</syntaxhighlight>
[https://vulkan.lunarg.com/sdk/home Vulkan SDK] herunterladen
entpacken, ins Verzeichnis wechseln und source setup-env.sh ausführen. Dann ins Verzeichnis wechseln wo llama.cpp installiert werden soll.
Dort dann
<syntaxhighlight lang="bash" line="1">
source /home/hendrik/Programme/Vulkan_SDK/1.4.335.0/setup-env.sh
cd /home/hendrik/Programme/llama.cpp/
rm -R build-vulkan
cmake -B build-vulkan -DGGML_VULKAN=1
cmake --build build-vulkan --config Release -- -j $(nproc)
</syntaxhighlight>

==== ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 ====
gfx906 = Mi 50 Support 
gfx1100 = 7900 XTX Support

<syntaxhighlight lang="bash" line="1">
Copy & paste all the commands from the quick install https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/quick-start.html
Before rebooting to complete the install, download the 6.4 rocblas from the AUR: https://archlinux.org/packages/extra/x86_64/rocblas/
Extract it
Copy all tensor files that contain gfx906 in rocblas-6.4.3-3-x86_64.pkg/opt/rocm/lib/rocblas/library to /opt/rocm/lib/rocblas/library
Reboot
Check if it worked by running sudo update-alternatives --display rocm
</syntaxhighlight>
# ROCm Umgebung einrichten (optional falls Fehler auftreten)
<syntaxhighlight lang="bash" line="1">
echo 'export LD_LIBRARY_PATH=/opt/rocm/lib:$LD_LIBRARY_PATH' >> ~/.bashrc
echo 'export HSA_OVERRIDE_GFX_VERSION=9.0.6' >> ~/.bashrc # Für MI50/MI60
source ~/.bashrc
</syntaxhighlight>
<syntaxhighlight lang="bash" line="1">
sudo dnf install rocwmma-devel -y
</syntaxhighlight>
# llama.cpp kompilieren
<syntaxhighlight lang="bash" line="1">
cd /home/hendrik/Programme/llama.cpp/ && rm -R build-rocm-old && cp -R build-rocm build-rocm-old
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" cmake -B build-rocm -DGGML_HIP=ON -DAMDGPU_TARGETS=gfx1100 -DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 16
</syntaxhighlight>

Für mehrere GPU-Architekturen gleichzeitig:
<syntaxhighlight lang="bash" line="1">
# Für MI50 (gfx906) und RX 7900 XTX (gfx1100)
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
cmake -S . -B build -DGGML_HIP=ON -DAMDGPU_TARGETS="gfx906;gfx1100" \
-DCMAKE_BUILD_TYPE=Release && \
cmake --build build --config Release -- -j 8
</syntaxhighlight>

=== Konfiguration ===

==== llama-server als systemd Service einrichten ====

'''Service-Datei erstellen:'''
<syntaxhighlight lang="bash" line="1">
sudo nano /etc/systemd/system/llama-server.service
</syntaxhighlight>

'''Service-Datei Inhalt (Multi-GPU mit ROCm):'''
<syntaxhighlight lang="ini" line="1">
[Unit]
Description=Llama.cpp ROCm Multi-GPU Server
After=network.target

[Service]
Type=simple
User=username
Group=username
WorkingDirectory=/home/username/llama.cpp/build/bin

# Multi-GPU Konfiguration
Environment="HIP_VISIBLE_DEVICES=0,1"
Environment="HSA_OVERRIDE_GFX_VERSION=9.0.6"
Environment="PATH=/opt/rocm/bin:/usr/local/bin:/usr/bin:/bin"
Environment="LD_LIBRARY_PATH=/opt/rocm/lib"

# Server mit optimalen Multi-GPU Einstellungen
ExecStart=/home/username/llama.cpp/build/bin/llama-server \
-m /home/username/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja

Restart=always
RestartSec=10
LimitNOFILE=65535
LimitMEMLOCK=infinity
StandardOutput=journal
StandardError=journal
SyslogIdentifier=llama-server

[Install]
WantedBy=multi-user.target
</syntaxhighlight>

'''Service aktivieren und starten:'''
<syntaxhighlight lang="bash" line="1">
# Service neu laden
sudo systemctl daemon-reload

# Service aktivieren (Auto-Start beim Boot)
sudo systemctl enable llama-server

# Service starten
sudo systemctl start llama-server

# Status prüfen
sudo systemctl status llama-server

# Logs anzeigen
sudo journalctl -u llama-server -f
</syntaxhighlight>

'''Service Management:'''
<syntaxhighlight lang="bash" line="1">
# Service stoppen
sudo systemctl stop llama-server

# Service neustarten
sudo systemctl restart llama-server

# Service deaktivieren
sudo systemctl disable llama-server

==== Manuelle Server-Starts ====
'''Multi-GPU (ROCm) - Optimal:'''
<syntaxhighlight lang="bash" line="1">
HIP_VISIBLE_DEVICES=0,1 ./llama-server \
-m ~/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja
</syntaxhighlight>

=== Update ===
<syntaxhighlight lang="bash" line="1">
cd ~/llama.cpp
git pull
cmake --build build --config Release -- -j $(nproc)

# Service neu starten falls aktiv
sudo systemctl restart llama-server
</syntaxhighlight>
=== Befehlszeilenargumente ===
llama-server
<syntaxhighlight lang="bash" line="1">
-h, --help, --usage print usage and exit
--version show version and build info
-cl, --cache-list show list of models in cache
--completion-bash print source-able bash completion script for llama.cpp
--verbose-prompt print a verbose prompt before generation (default: false)
-t, --threads N number of CPU threads to use during generation (default: -1)
(env: LLAMA_ARG_THREADS)
-tb, --threads-batch N number of threads to use during batch and prompt processing (default:
same as --threads)
-C, --cpu-mask M CPU affinity mask: arbitrarily long hex. Complements cpu-range
(default: "")
-Cr, --cpu-range lo-hi range of CPUs for affinity. Complements --cpu-mask
--cpu-strict <0|1> use strict CPU placement (default: 0)
--prio N set process/thread priority : low(-1), normal(0), medium(1), high(2),
realtime(3) (default: 0)
--poll <0...100> use polling level to wait for work (0 - no polling, default: 50)
-Cb, --cpu-mask-batch M CPU affinity mask: arbitrarily long hex. Complements cpu-range-batch
(default: same as --cpu-mask)
-Crb, --cpu-range-batch lo-hi ranges of CPUs for affinity. Complements --cpu-mask-batch
--cpu-strict-batch <0|1> use strict CPU placement (default: same as --cpu-strict)
--prio-batch N set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime
(default: 0)
--poll-batch <0|1> use polling to wait for work (default: same as --poll)
-c, --ctx-size N size of the prompt context (default: 4096, 0 = loaded from model)
(env: LLAMA_ARG_CTX_SIZE)
-n, --predict, --n-predict N number of tokens to predict (default: -1, -1 = infinity)
(env: LLAMA_ARG_N_PREDICT)
-b, --batch-size N logical maximum batch size (default: 2048)
(env: LLAMA_ARG_BATCH)
-ub, --ubatch-size N physical maximum batch size (default: 512)
(env: LLAMA_ARG_UBATCH)
--keep N number of tokens to keep from the initial prompt (default: 0, -1 =
all)
--swa-full use full-size SWA cache (default: false)
[(more
info)](https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
(env: LLAMA_ARG_SWA_FULL)
--kv-unified, -kvu use single unified KV buffer for the KV cache of all sequences
(default: false)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/14363)
(env: LLAMA_ARG_KV_SPLIT)
-fa, --flash-attn [on|off|auto] set Flash Attention use ('on', 'off', or 'auto', default: 'auto')
(env: LLAMA_ARG_FLASH_ATTN)
--no-perf disable internal libllama performance timings (default: false)
(env: LLAMA_ARG_NO_PERF)
-e, --escape process escapes sequences (\n, \r, \t, \', \", \\) (default: true)
--no-escape do not process escape sequences
--rope-scaling {none,linear,yarn} RoPE frequency scaling method, defaults to linear unless specified by
the model
(env: LLAMA_ARG_ROPE_SCALING_TYPE)
--rope-scale N RoPE context scaling factor, expands context by a factor of N
(env: LLAMA_ARG_ROPE_SCALE)
--rope-freq-base N RoPE base frequency, used by NTK-aware scaling (default: loaded from
model)
(env: LLAMA_ARG_ROPE_FREQ_BASE)
--rope-freq-scale N RoPE frequency scaling factor, expands context by a factor of 1/N
(env: LLAMA_ARG_ROPE_FREQ_SCALE)
--yarn-orig-ctx N YaRN: original context size of model (default: 0 = model training
context size)
(env: LLAMA_ARG_YARN_ORIG_CTX)
--yarn-ext-factor N YaRN: extrapolation mix factor (default: -1.0, 0.0 = full
interpolation)
(env: LLAMA_ARG_YARN_EXT_FACTOR)
--yarn-attn-factor N YaRN: scale sqrt(t) or attention magnitude (default: -1.0)
(env: LLAMA_ARG_YARN_ATTN_FACTOR)
--yarn-beta-slow N YaRN: high correction dim or alpha (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_SLOW)
--yarn-beta-fast N YaRN: low correction dim or beta (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_FAST)
-nkvo, --no-kv-offload disable KV offload
(env: LLAMA_ARG_NO_KV_OFFLOAD)
-nr, --no-repack disable weight repacking
(env: LLAMA_ARG_NO_REPACK)
--no-host bypass host buffer allowing extra buffers to be used
(env: LLAMA_ARG_NO_HOST)
-ctk, --cache-type-k TYPE KV cache data type for K
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K)
-ctv, --cache-type-v TYPE KV cache data type for V
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V)
-dt, --defrag-thold N KV cache defragmentation threshold (DEPRECATED)
(env: LLAMA_ARG_DEFRAG_THOLD)
-np, --parallel N number of parallel sequences to decode (default: 1)
(env: LLAMA_ARG_N_PARALLEL)
--mlock force system to keep model in RAM rather than swapping or compressing
(env: LLAMA_ARG_MLOCK)
--no-mmap do not memory-map model (slower load but may reduce pageouts if not
using mlock)
(env: LLAMA_ARG_NO_MMAP)
--numa TYPE attempt optimizations that help on some NUMA systems
- distribute: spread execution evenly over all nodes
- isolate: only spawn threads on CPUs on the node that execution
started on
- numactl: use the CPU map provided by numactl
if run without this previously, it is recommended to drop the system
page cache before using this
see https://github.com/ggml-org/llama.cpp/issues/1437
(env: LLAMA_ARG_NUMA)
-dev, --device <dev1,dev2,..> comma-separated list of devices to use for offloading (none = don't
offload)
use --list-devices to see a list of available devices
(env: LLAMA_ARG_DEVICE)
--list-devices print list of available devices and exit
--override-tensor, -ot <tensor name pattern>=<buffer type>,...
override tensor buffer type
--cpu-moe, -cmoe keep all Mixture of Experts (MoE) weights in the CPU
(env: LLAMA_ARG_CPU_MOE)
--n-cpu-moe, -ncmoe N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU
(env: LLAMA_ARG_N_CPU_MOE)
-ngl, --gpu-layers, --n-gpu-layers N max. number of layers to store in VRAM (default: -1)
(env: LLAMA_ARG_N_GPU_LAYERS)
-sm, --split-mode {none,layer,row} how to split the model across multiple GPUs, one of:
- none: use one GPU only
- layer (default): split layers and KV across GPUs
- row: split rows across GPUs
(env: LLAMA_ARG_SPLIT_MODE)
-ts, --tensor-split N0,N1,N2,... fraction of the model to offload to each GPU, comma-separated list of
proportions, e.g. 3,1
(env: LLAMA_ARG_TENSOR_SPLIT)
-mg, --main-gpu INDEX the GPU to use for the model (with split-mode = none), or for
intermediate results and KV (with split-mode = row) (default: 0)
(env: LLAMA_ARG_MAIN_GPU)
--check-tensors check model tensor data for invalid values (default: false)
--override-kv KEY=TYPE:VALUE advanced option to override model metadata by key. may be specified
multiple times.
types: int, float, bool, str. example: --override-kv
tokenizer.ggml.add_bos_token=bool:false
--no-op-offload disable offloading host tensor operations to device (default: false)
--lora FNAME path to LoRA adapter (can be repeated to use multiple adapters)
--lora-scaled FNAME SCALE path to LoRA adapter with user defined scaling (can be repeated to use
multiple adapters)
--control-vector FNAME add a control vector
note: this argument can be repeated to add multiple control vectors
--control-vector-scaled FNAME SCALE add a control vector with user defined scaling SCALE
note: this argument can be repeated to add multiple scaled control
vectors
--control-vector-layer-range START END
layer range to apply the control vector(s) to, start and end inclusive
-m, --model FNAME model path (default: `models/$filename` with filename from `--hf-file`
or `--model-url` if set, otherwise models/7B/ggml-model-f16.gguf)
(env: LLAMA_ARG_MODEL)
-mu, --model-url MODEL_URL model download url (default: unused)
(env: LLAMA_ARG_MODEL_URL)
-dr, --docker-repo [<repo>/]<model>[:quant]
Docker Hub model repository. repo is optional, default to ai/. quant
is optional, default to :latest.
example: gemma3
(default: unused)
(env: LLAMA_ARG_DOCKER_REPO)
-hf, -hfr, --hf-repo <user>/<model>[:quant]
Hugging Face model repository; quant is optional, case-insensitive,
default to Q4_K_M, or falls back to the first file in the repo if
Q4_K_M doesn't exist.
mmproj is also downloaded automatically if available. to disable, add
--no-mmproj
example: unsloth/phi-4-GGUF:q4_k_m
(default: unused)
(env: LLAMA_ARG_HF_REPO)
-hfd, -hfrd, --hf-repo-draft <user>/<model>[:quant]
Same as --hf-repo, but for the draft model (default: unused)
(env: LLAMA_ARG_HFD_REPO)
-hff, --hf-file FILE Hugging Face model file. If specified, it will override the quant in
--hf-repo (default: unused)
(env: LLAMA_ARG_HF_FILE)
-hfv, -hfrv, --hf-repo-v <user>/<model>[:quant]
Hugging Face model repository for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_REPO_V)
-hffv, --hf-file-v FILE Hugging Face model file for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_FILE_V)
-hft, --hf-token TOKEN Hugging Face access token (default: value from HF_TOKEN environment
variable)
(env: HF_TOKEN)
--log-disable Log disable
--log-file FNAME Log to file
--log-colors [on|off|auto] Set colored logging ('on', 'off', or 'auto', default: 'auto')
'auto' enables colors when output is to a terminal
(env: LLAMA_LOG_COLORS)
-v, --verbose, --log-verbose Set verbosity level to infinity (i.e. log all messages, useful for
debugging)
--offline Offline mode: forces use of cache, prevents network access
(env: LLAMA_OFFLINE)
-lv, --verbosity, --log-verbosity N Set the verbosity threshold. Messages with a higher verbosity will be
ignored.
(env: LLAMA_LOG_VERBOSITY)
--log-prefix Enable prefix in log messages
(env: LLAMA_LOG_PREFIX)
--log-timestamps Enable timestamps in log messages
(env: LLAMA_LOG_TIMESTAMPS)
-ctkd, --cache-type-k-draft TYPE KV cache data type for K for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K_DRAFT)
-ctvd, --cache-type-v-draft TYPE KV cache data type for V for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V_DRAFT)

----- sampling params -----

--samplers SAMPLERS samplers that will be used for generation in the order, separated by
';'
(default:
penalties;dry;top_n_sigma;top_k;typ_p;top_p;min_p;xtc;temperature)
-s, --seed SEED RNG seed (default: -1, use random seed for -1)
--sampling-seq, --sampler-seq SEQUENCE
simplified sequence for samplers that will be used (default:
edskypmxt)
--ignore-eos ignore end of stream token and continue generating (implies
--logit-bias EOS-inf)
--temp N temperature (default: 0.8)
--top-k N top-k sampling (default: 40, 0 = disabled)
--top-p N top-p sampling (default: 0.9, 1.0 = disabled)
--min-p N min-p sampling (default: 0.1, 0.0 = disabled)
--top-nsigma N top-n-sigma sampling (default: -1.0, -1.0 = disabled)
--xtc-probability N xtc probability (default: 0.0, 0.0 = disabled)
--xtc-threshold N xtc threshold (default: 0.1, 1.0 = disabled)
--typical N locally typical sampling, parameter p (default: 1.0, 1.0 = disabled)
--repeat-last-n N last n tokens to consider for penalize (default: 64, 0 = disabled, -1
= ctx_size)
--repeat-penalty N penalize repeat sequence of tokens (default: 1.0, 1.0 = disabled)
--presence-penalty N repeat alpha presence penalty (default: 0.0, 0.0 = disabled)
--frequency-penalty N repeat alpha frequency penalty (default: 0.0, 0.0 = disabled)
--dry-multiplier N set DRY sampling multiplier (default: 0.0, 0.0 = disabled)
--dry-base N set DRY sampling base value (default: 1.75)
--dry-allowed-length N set allowed length for DRY sampling (default: 2)
--dry-penalty-last-n N set DRY penalty for the last n tokens (default: -1, 0 = disable, -1 =
context size)
--dry-sequence-breaker STRING add sequence breaker for DRY sampling, clearing out default breakers
('\n', ':', '"', '*') in the process; use "none" to not use any
sequence breakers
--dynatemp-range N dynamic temperature range (default: 0.0, 0.0 = disabled)
--dynatemp-exp N dynamic temperature exponent (default: 1.0)
--mirostat N use Mirostat sampling.
Top K, Nucleus and Locally Typical samplers are ignored if used.
(default: 0, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)
--mirostat-lr N Mirostat learning rate, parameter eta (default: 0.1)
--mirostat-ent N Mirostat target entropy, parameter tau (default: 5.0)
-l, --logit-bias TOKEN_ID(+/-)BIAS modifies the likelihood of token appearing in the completion,
i.e. `--logit-bias 15043+1` to increase likelihood of token ' Hello',
or `--logit-bias 15043-1` to decrease likelihood of token ' Hello'
--grammar GRAMMAR BNF-like grammar to constrain generations (see samples in grammars/
dir) (default: '')
--grammar-file FNAME file to read grammar from
-j, --json-schema SCHEMA JSON schema to constrain generations (https://json-schema.org/), e.g.
`{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead
-jf, --json-schema-file FILE File containing a JSON schema to constrain generations
(https://json-schema.org/), e.g. `{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead

----- example-specific params -----

--ctx-checkpoints, --swa-checkpoints N
max number of context checkpoints to create per slot (default: 8)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/15293)
(env: LLAMA_ARG_CTX_CHECKPOINTS)
--cache-ram, -cram N set the maximum cache size in MiB (default: 8192, -1 - no limit, 0 -
disable)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/16391)
(env: LLAMA_ARG_CACHE_RAM)
--no-context-shift disables context shift on infinite text generation (default: enabled)
(env: LLAMA_ARG_NO_CONTEXT_SHIFT)
--context-shift enables context shift on infinite text generation (default: disabled)
(env: LLAMA_ARG_CONTEXT_SHIFT)
-r, --reverse-prompt PROMPT halt generation at PROMPT, return control in interactive mode
-sp, --special special tokens output enabled (default: false)
--no-warmup skip warming up the model with an empty run
--spm-infill use Suffix/Prefix/Middle pattern for infill (instead of
Prefix/Suffix/Middle) as some models prefer this. (default: disabled)
--pooling {none,mean,cls,last,rank} pooling type for embeddings, use model default if unspecified
(env: LLAMA_ARG_POOLING)
-cb, --cont-batching enable continuous batching (a.k.a dynamic batching) (default: enabled)
(env: LLAMA_ARG_CONT_BATCHING)
-nocb, --no-cont-batching disable continuous batching
(env: LLAMA_ARG_NO_CONT_BATCHING)
--mmproj FILE path to a multimodal projector file. see tools/mtmd/README.md
note: if -hf is used, this argument can be omitted
(env: LLAMA_ARG_MMPROJ)
--mmproj-url URL URL to a multimodal projector file. see tools/mtmd/README.md
(env: LLAMA_ARG_MMPROJ_URL)
--no-mmproj explicitly disable multimodal projector, useful when using -hf
(env: LLAMA_ARG_NO_MMPROJ)
--no-mmproj-offload do not offload multimodal projector to GPU
(env: LLAMA_ARG_NO_MMPROJ_OFFLOAD)
--image-min-tokens N minimum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MIN_TOKENS)
--image-max-tokens N maximum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MAX_TOKENS)
--override-tensor-draft, -otd <tensor name pattern>=<buffer type>,...
override tensor buffer type for draft model
--cpu-moe-draft, -cmoed keep all Mixture of Experts (MoE) weights in the CPU for the draft
model
(env: LLAMA_ARG_CPU_MOE_DRAFT)
--n-cpu-moe-draft, -ncmoed N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU for the draft model
(env: LLAMA_ARG_N_CPU_MOE_DRAFT)
-a, --alias STRING set alias for model name (to be used by REST API)
(env: LLAMA_ARG_ALIAS)
--host HOST ip address to listen, or bind to an UNIX socket if the address ends
with .sock (default: 127.0.0.1)
(env: LLAMA_ARG_HOST)
--port PORT port to listen (default: 8080)
(env: LLAMA_ARG_PORT)
--path PATH path to serve static files from (default: )
(env: LLAMA_ARG_STATIC_PATH)
--api-prefix PREFIX prefix path the server serves from, without the trailing slash
(default: )
(env: LLAMA_ARG_API_PREFIX)
--no-webui Disable the Web UI (default: enabled)
(env: LLAMA_ARG_NO_WEBUI)
--embedding, --embeddings restrict to only support embedding use case; use only with dedicated
embedding models (default: disabled)
(env: LLAMA_ARG_EMBEDDINGS)
--reranking, --rerank enable reranking endpoint on server (default: disabled)
(env: LLAMA_ARG_RERANKING)
--api-key KEY API key to use for authentication (default: none)
(env: LLAMA_API_KEY)
--api-key-file FNAME path to file containing API keys (default: none)
--ssl-key-file FNAME path to file a PEM-encoded SSL private key
(env: LLAMA_ARG_SSL_KEY_FILE)
--ssl-cert-file FNAME path to file a PEM-encoded SSL certificate
(env: LLAMA_ARG_SSL_CERT_FILE)
--chat-template-kwargs STRING sets additional params for the json template parser
(env: LLAMA_CHAT_TEMPLATE_KWARGS)
-to, --timeout N server read/write timeout in seconds (default: 600)
(env: LLAMA_ARG_TIMEOUT)
--threads-http N number of threads used to process HTTP requests (default: -1)
(env: LLAMA_ARG_THREADS_HTTP)
--cache-reuse N min chunk size to attempt reusing from the cache via KV shifting
(default: 0)
[(card)](https://ggml.ai/f0.png)
(env: LLAMA_ARG_CACHE_REUSE)
--metrics enable prometheus compatible metrics endpoint (default: disabled)
(env: LLAMA_ARG_ENDPOINT_METRICS)
--props enable changing global properties via POST /props (default: disabled)
(env: LLAMA_ARG_ENDPOINT_PROPS)
--slots enable slots monitoring endpoint (default: enabled)
(env: LLAMA_ARG_ENDPOINT_SLOTS)
--no-slots disables slots monitoring endpoint
(env: LLAMA_ARG_NO_ENDPOINT_SLOTS)
--slot-save-path PATH path to save slot kv cache (default: disabled)
--jinja use jinja template for chat (default: disabled)
(env: LLAMA_ARG_JINJA)
--reasoning-format FORMAT controls whether thought tags are allowed and/or extracted from the
response, and in which format they're returned; one of:
- none: leaves thoughts unparsed in `message.content`
- deepseek: puts thoughts in `message.reasoning_content`
- deepseek-legacy: keeps `<think>` tags in `message.content` while
also populating `message.reasoning_content`
(default: auto)
(env: LLAMA_ARG_THINK)
--reasoning-budget N controls the amount of thinking allowed; currently only one of: -1 for
unrestricted thinking budget, or 0 to disable thinking (default: -1)
(env: LLAMA_ARG_THINK_BUDGET)
--chat-template JINJA_TEMPLATE set custom jinja chat template (default: template taken from model's
metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE)
--chat-template-file JINJA_TEMPLATE_FILE
set custom jinja chat template file (default: template taken from
model's metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE_FILE)
--no-prefill-assistant whether to prefill the assistant's response if the last message is an
assistant message (default: prefill enabled)
when this flag is set, if the last message is an assistant message
then it will be treated as a full message and not prefilled

(env: LLAMA_ARG_NO_PREFILL_ASSISTANT)
-sps, --slot-prompt-similarity SIMILARITY
how much the prompt of a request must match the prompt of a slot in
order to use that slot (default: 0.10, 0.0 = disabled)
--lora-init-without-apply load LoRA adapters without applying them (apply later via POST
/lora-adapters) (default: disabled)
-td, --threads-draft N number of threads to use during generation (default: same as
--threads)
-tbd, --threads-batch-draft N number of threads to use during batch and prompt processing (default:
same as --threads-draft)
--draft-max, --draft, --draft-n N number of tokens to draft for speculative decoding (default: 16)
(env: LLAMA_ARG_DRAFT_MAX)
--draft-min, --draft-n-min N minimum number of draft tokens to use for speculative decoding
(default: 0)
(env: LLAMA_ARG_DRAFT_MIN)
--draft-p-min P minimum speculative decoding probability (greedy) (default: 0.8)
(env: LLAMA_ARG_DRAFT_P_MIN)
-cd, --ctx-size-draft N size of the prompt context for the draft model (default: 0, 0 = loaded
from model)
(env: LLAMA_ARG_CTX_SIZE_DRAFT)
-devd, --device-draft <dev1,dev2,..> comma-separated list of devices to use for offloading the draft model
(none = don't offload)
use --list-devices to see a list of available devices
-ngld, --gpu-layers-draft, --n-gpu-layers-draft N
number of layers to store in VRAM for the draft model
(env: LLAMA_ARG_N_GPU_LAYERS_DRAFT)
-md, --model-draft FNAME draft model for speculative decoding (default: unused)
(env: LLAMA_ARG_MODEL_DRAFT)
--spec-replace TARGET DRAFT translate the string in TARGET into DRAFT if the draft model and main
model are not compatible
-mv, --model-vocoder FNAME vocoder model for audio generation (default: unused)
--tts-use-guide-tokens Use guide tokens to improve TTS word recall
--embd-gemma-default use default EmbeddingGemma model (note: can download weights from the
internet)
--fim-qwen-1.5b-default use default Qwen 2.5 Coder 1.5B (note: can download weights from the
internet)
--fim-qwen-3b-default use default Qwen 2.5 Coder 3B (note: can download weights from the
internet)
--fim-qwen-7b-default use default Qwen 2.5 Coder 7B (note: can download weights from the
internet)
--fim-qwen-7b-spec use Qwen 2.5 Coder 7B + 0.5B draft for speculative decoding (note: can
download weights from the internet)
--fim-qwen-14b-spec use Qwen 2.5 Coder 14B + 0.5B draft for speculative decoding (note:
can download weights from the internet)
--fim-qwen-30b-default use default Qwen 3 Coder 30B A3B Instruct (note: can download weights
from the internet)
--gpt-oss-20b-default use gpt-oss-20b (note: can download weights from the internet)
--gpt-oss-120b-default use gpt-oss-120b (note: can download weights from the internet)
--vision-gemma-4b-default use Gemma 3 4B QAT (note: can download weights from the internet)
--vision-gemma-12b-default use Gemma 3 12B QAT (note: can download weights from the internet)
</syntaxhighlight>
=== Test ===
=== Bekannte Probleme ===
<syntaxhighlight lang="bash" line="1">
Error 1 [ 34%] Linking HIP shared library ../../../bin/libggml-hip.so [ 34%] Built target ggml-hip gmake[1]: *** [CMakeFiles/Makefile2:1775: ggml/src/CMakeFiles/ggml-cpu.dir/all] Error 2 gmake: *** [Makefile:146: all] Error 2
</syntaxhighlight>
Lösung:
<syntaxhighlight lang="bash" line="1">
GCC 13.3.0 konnte damit nicht umgehen, aber GCC 14.2.0 hat bessere C++-Standard-Compliance und kann diesen Header-Konflikt korrekt auflösen – deshalb funktioniert die Kompilierung jetzt mit der neueren Compiler-Version.
sudo update-alternatives --remove gcc /usr/bin/gcc-13
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-14 100
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-13 50
sudo update-alternatives --config gcc
# Wähle gcc-14

sudo update-alternatives --remove g++ /usr/bin/g++-13
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-14 100
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-13 50
sudo update-alternatives --config g++
# Wähle g++-14
</syntaxhighlight>
=== Nützliche Links ===
* [https://github.com/ggml-org/llama.cpp Offizielles llama.cpp Repository]
* [https://github.com/ggml-org/llama.cpp/blob/master/examples/server/README.md llama-server Dokumentation]
* [https://github.com/ggml-org/llama.cpp/discussions llama.cpp Discussions]
* [https://huggingface.co/models?library=gguf GGUF Modelle auf Hugging Face]
* [https://docs.rocm.com/ ROCm Dokumentation]
* [https://vulkan.lunarg.com/doc/sdk Vulkan SDK Dokumentation]

Llama.cpp

2026-01-07T14:50:58Z

Hendrik: /* ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 */

=== Beschreibung ===
llama.cpp ist eine C/C++ Implementierung für Inference von Large Language Models (LLMs). Es unterstützt verschiedene Backends (CPU, Vulkan, ROCm, CUDA) und ermöglicht das Ausführen von quantisierten Modellen im GGUF-Format.

=== Download ===
<syntaxhighlight lang="bash" line="1">
git clone https://github.com/ggml-org/llama.cpp
cd llama.cpp
</syntaxhighlight>

=== Installation ===
==== Vulkan Build ====
<syntaxhighlight lang="bash" line="1">
sudo dnf install libcurl-devel -y
</syntaxhighlight>
[https://vulkan.lunarg.com/sdk/home Vulkan SDK] herunterladen
entpacken, ins Verzeichnis wechseln und source setup-env.sh ausführen. Dann ins Verzeichnis wechseln wo llama.cpp installiert werden soll.
Dort dann
<syntaxhighlight lang="bash" line="1">
source /home/hendrik/Programme/Vulkan_SDK/1.4.335.0/setup-env.sh
cd /home/hendrik/Programme/llama.cpp/
rm -R build-vulkan
cmake -B build-vulkan -DGGML_VULKAN=1
cmake --build build-vulkan --config Release -- -j $(nproc)
</syntaxhighlight>

==== ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 ====
gfx906 = Mi 50 Support 
gfx1100 = 7900 XTX Support

<syntaxhighlight lang="bash" line="1">
Copy & paste all the commands from the quick install https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/quick-start.html
Before rebooting to complete the install, download the 6.4 rocblas from the AUR: https://archlinux.org/packages/extra/x86_64/rocblas/
Extract it
Copy all tensor files that contain gfx906 in rocblas-6.4.3-3-x86_64.pkg/opt/rocm/lib/rocblas/library to /opt/rocm/lib/rocblas/library
Reboot
Check if it worked by running sudo update-alternatives --display rocm
</syntaxhighlight>
# ROCm Umgebung einrichten (optional falls Fehler auftreten)
<syntaxhighlight lang="bash" line="1">
echo 'export LD_LIBRARY_PATH=/opt/rocm/lib:$LD_LIBRARY_PATH' >> ~/.bashrc
echo 'export HSA_OVERRIDE_GFX_VERSION=9.0.6' >> ~/.bashrc # Für MI50/MI60
source ~/.bashrc
</syntaxhighlight>
<syntaxhighlight lang="bash" line="1">
sudo dnf install rocwmma-devel -y
</syntaxhighlight>
# llama.cpp kompilieren
<syntaxhighlight lang="bash" line="1">
cd /home/hendrik/Programme/llama.cpp/ && rm -R build-rocm
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" cmake -B build-rocm -DGGML_HIP=ON -DAMDGPU_TARGETS=gfx1100 -DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 16
</syntaxhighlight>

Für mehrere GPU-Architekturen gleichzeitig:
<syntaxhighlight lang="bash" line="1">
# Für MI50 (gfx906) und RX 7900 XTX (gfx1100)
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
cmake -S . -B build -DGGML_HIP=ON -DAMDGPU_TARGETS="gfx906;gfx1100" \
-DCMAKE_BUILD_TYPE=Release && \
cmake --build build --config Release -- -j 8
</syntaxhighlight>

=== Konfiguration ===

==== llama-server als systemd Service einrichten ====

'''Service-Datei erstellen:'''
<syntaxhighlight lang="bash" line="1">
sudo nano /etc/systemd/system/llama-server.service
</syntaxhighlight>

'''Service-Datei Inhalt (Multi-GPU mit ROCm):'''
<syntaxhighlight lang="ini" line="1">
[Unit]
Description=Llama.cpp ROCm Multi-GPU Server
After=network.target

[Service]
Type=simple
User=username
Group=username
WorkingDirectory=/home/username/llama.cpp/build/bin

# Multi-GPU Konfiguration
Environment="HIP_VISIBLE_DEVICES=0,1"
Environment="HSA_OVERRIDE_GFX_VERSION=9.0.6"
Environment="PATH=/opt/rocm/bin:/usr/local/bin:/usr/bin:/bin"
Environment="LD_LIBRARY_PATH=/opt/rocm/lib"

# Server mit optimalen Multi-GPU Einstellungen
ExecStart=/home/username/llama.cpp/build/bin/llama-server \
-m /home/username/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja

Restart=always
RestartSec=10
LimitNOFILE=65535
LimitMEMLOCK=infinity
StandardOutput=journal
StandardError=journal
SyslogIdentifier=llama-server

[Install]
WantedBy=multi-user.target
</syntaxhighlight>

'''Service aktivieren und starten:'''
<syntaxhighlight lang="bash" line="1">
# Service neu laden
sudo systemctl daemon-reload

# Service aktivieren (Auto-Start beim Boot)
sudo systemctl enable llama-server

# Service starten
sudo systemctl start llama-server

# Status prüfen
sudo systemctl status llama-server

# Logs anzeigen
sudo journalctl -u llama-server -f
</syntaxhighlight>

'''Service Management:'''
<syntaxhighlight lang="bash" line="1">
# Service stoppen
sudo systemctl stop llama-server

# Service neustarten
sudo systemctl restart llama-server

# Service deaktivieren
sudo systemctl disable llama-server

==== Manuelle Server-Starts ====
'''Multi-GPU (ROCm) - Optimal:'''
<syntaxhighlight lang="bash" line="1">
HIP_VISIBLE_DEVICES=0,1 ./llama-server \
-m ~/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja
</syntaxhighlight>

=== Update ===
<syntaxhighlight lang="bash" line="1">
cd ~/llama.cpp
git pull
cmake --build build --config Release -- -j $(nproc)

# Service neu starten falls aktiv
sudo systemctl restart llama-server
</syntaxhighlight>
=== Befehlszeilenargumente ===
llama-server
<syntaxhighlight lang="bash" line="1">
-h, --help, --usage print usage and exit
--version show version and build info
-cl, --cache-list show list of models in cache
--completion-bash print source-able bash completion script for llama.cpp
--verbose-prompt print a verbose prompt before generation (default: false)
-t, --threads N number of CPU threads to use during generation (default: -1)
(env: LLAMA_ARG_THREADS)
-tb, --threads-batch N number of threads to use during batch and prompt processing (default:
same as --threads)
-C, --cpu-mask M CPU affinity mask: arbitrarily long hex. Complements cpu-range
(default: "")
-Cr, --cpu-range lo-hi range of CPUs for affinity. Complements --cpu-mask
--cpu-strict <0|1> use strict CPU placement (default: 0)
--prio N set process/thread priority : low(-1), normal(0), medium(1), high(2),
realtime(3) (default: 0)
--poll <0...100> use polling level to wait for work (0 - no polling, default: 50)
-Cb, --cpu-mask-batch M CPU affinity mask: arbitrarily long hex. Complements cpu-range-batch
(default: same as --cpu-mask)
-Crb, --cpu-range-batch lo-hi ranges of CPUs for affinity. Complements --cpu-mask-batch
--cpu-strict-batch <0|1> use strict CPU placement (default: same as --cpu-strict)
--prio-batch N set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime
(default: 0)
--poll-batch <0|1> use polling to wait for work (default: same as --poll)
-c, --ctx-size N size of the prompt context (default: 4096, 0 = loaded from model)
(env: LLAMA_ARG_CTX_SIZE)
-n, --predict, --n-predict N number of tokens to predict (default: -1, -1 = infinity)
(env: LLAMA_ARG_N_PREDICT)
-b, --batch-size N logical maximum batch size (default: 2048)
(env: LLAMA_ARG_BATCH)
-ub, --ubatch-size N physical maximum batch size (default: 512)
(env: LLAMA_ARG_UBATCH)
--keep N number of tokens to keep from the initial prompt (default: 0, -1 =
all)
--swa-full use full-size SWA cache (default: false)
[(more
info)](https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
(env: LLAMA_ARG_SWA_FULL)
--kv-unified, -kvu use single unified KV buffer for the KV cache of all sequences
(default: false)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/14363)
(env: LLAMA_ARG_KV_SPLIT)
-fa, --flash-attn [on|off|auto] set Flash Attention use ('on', 'off', or 'auto', default: 'auto')
(env: LLAMA_ARG_FLASH_ATTN)
--no-perf disable internal libllama performance timings (default: false)
(env: LLAMA_ARG_NO_PERF)
-e, --escape process escapes sequences (\n, \r, \t, \', \", \\) (default: true)
--no-escape do not process escape sequences
--rope-scaling {none,linear,yarn} RoPE frequency scaling method, defaults to linear unless specified by
the model
(env: LLAMA_ARG_ROPE_SCALING_TYPE)
--rope-scale N RoPE context scaling factor, expands context by a factor of N
(env: LLAMA_ARG_ROPE_SCALE)
--rope-freq-base N RoPE base frequency, used by NTK-aware scaling (default: loaded from
model)
(env: LLAMA_ARG_ROPE_FREQ_BASE)
--rope-freq-scale N RoPE frequency scaling factor, expands context by a factor of 1/N
(env: LLAMA_ARG_ROPE_FREQ_SCALE)
--yarn-orig-ctx N YaRN: original context size of model (default: 0 = model training
context size)
(env: LLAMA_ARG_YARN_ORIG_CTX)
--yarn-ext-factor N YaRN: extrapolation mix factor (default: -1.0, 0.0 = full
interpolation)
(env: LLAMA_ARG_YARN_EXT_FACTOR)
--yarn-attn-factor N YaRN: scale sqrt(t) or attention magnitude (default: -1.0)
(env: LLAMA_ARG_YARN_ATTN_FACTOR)
--yarn-beta-slow N YaRN: high correction dim or alpha (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_SLOW)
--yarn-beta-fast N YaRN: low correction dim or beta (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_FAST)
-nkvo, --no-kv-offload disable KV offload
(env: LLAMA_ARG_NO_KV_OFFLOAD)
-nr, --no-repack disable weight repacking
(env: LLAMA_ARG_NO_REPACK)
--no-host bypass host buffer allowing extra buffers to be used
(env: LLAMA_ARG_NO_HOST)
-ctk, --cache-type-k TYPE KV cache data type for K
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K)
-ctv, --cache-type-v TYPE KV cache data type for V
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V)
-dt, --defrag-thold N KV cache defragmentation threshold (DEPRECATED)
(env: LLAMA_ARG_DEFRAG_THOLD)
-np, --parallel N number of parallel sequences to decode (default: 1)
(env: LLAMA_ARG_N_PARALLEL)
--mlock force system to keep model in RAM rather than swapping or compressing
(env: LLAMA_ARG_MLOCK)
--no-mmap do not memory-map model (slower load but may reduce pageouts if not
using mlock)
(env: LLAMA_ARG_NO_MMAP)
--numa TYPE attempt optimizations that help on some NUMA systems
- distribute: spread execution evenly over all nodes
- isolate: only spawn threads on CPUs on the node that execution
started on
- numactl: use the CPU map provided by numactl
if run without this previously, it is recommended to drop the system
page cache before using this
see https://github.com/ggml-org/llama.cpp/issues/1437
(env: LLAMA_ARG_NUMA)
-dev, --device <dev1,dev2,..> comma-separated list of devices to use for offloading (none = don't
offload)
use --list-devices to see a list of available devices
(env: LLAMA_ARG_DEVICE)
--list-devices print list of available devices and exit
--override-tensor, -ot <tensor name pattern>=<buffer type>,...
override tensor buffer type
--cpu-moe, -cmoe keep all Mixture of Experts (MoE) weights in the CPU
(env: LLAMA_ARG_CPU_MOE)
--n-cpu-moe, -ncmoe N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU
(env: LLAMA_ARG_N_CPU_MOE)
-ngl, --gpu-layers, --n-gpu-layers N max. number of layers to store in VRAM (default: -1)
(env: LLAMA_ARG_N_GPU_LAYERS)
-sm, --split-mode {none,layer,row} how to split the model across multiple GPUs, one of:
- none: use one GPU only
- layer (default): split layers and KV across GPUs
- row: split rows across GPUs
(env: LLAMA_ARG_SPLIT_MODE)
-ts, --tensor-split N0,N1,N2,... fraction of the model to offload to each GPU, comma-separated list of
proportions, e.g. 3,1
(env: LLAMA_ARG_TENSOR_SPLIT)
-mg, --main-gpu INDEX the GPU to use for the model (with split-mode = none), or for
intermediate results and KV (with split-mode = row) (default: 0)
(env: LLAMA_ARG_MAIN_GPU)
--check-tensors check model tensor data for invalid values (default: false)
--override-kv KEY=TYPE:VALUE advanced option to override model metadata by key. may be specified
multiple times.
types: int, float, bool, str. example: --override-kv
tokenizer.ggml.add_bos_token=bool:false
--no-op-offload disable offloading host tensor operations to device (default: false)
--lora FNAME path to LoRA adapter (can be repeated to use multiple adapters)
--lora-scaled FNAME SCALE path to LoRA adapter with user defined scaling (can be repeated to use
multiple adapters)
--control-vector FNAME add a control vector
note: this argument can be repeated to add multiple control vectors
--control-vector-scaled FNAME SCALE add a control vector with user defined scaling SCALE
note: this argument can be repeated to add multiple scaled control
vectors
--control-vector-layer-range START END
layer range to apply the control vector(s) to, start and end inclusive
-m, --model FNAME model path (default: `models/$filename` with filename from `--hf-file`
or `--model-url` if set, otherwise models/7B/ggml-model-f16.gguf)
(env: LLAMA_ARG_MODEL)
-mu, --model-url MODEL_URL model download url (default: unused)
(env: LLAMA_ARG_MODEL_URL)
-dr, --docker-repo [<repo>/]<model>[:quant]
Docker Hub model repository. repo is optional, default to ai/. quant
is optional, default to :latest.
example: gemma3
(default: unused)
(env: LLAMA_ARG_DOCKER_REPO)
-hf, -hfr, --hf-repo <user>/<model>[:quant]
Hugging Face model repository; quant is optional, case-insensitive,
default to Q4_K_M, or falls back to the first file in the repo if
Q4_K_M doesn't exist.
mmproj is also downloaded automatically if available. to disable, add
--no-mmproj
example: unsloth/phi-4-GGUF:q4_k_m
(default: unused)
(env: LLAMA_ARG_HF_REPO)
-hfd, -hfrd, --hf-repo-draft <user>/<model>[:quant]
Same as --hf-repo, but for the draft model (default: unused)
(env: LLAMA_ARG_HFD_REPO)
-hff, --hf-file FILE Hugging Face model file. If specified, it will override the quant in
--hf-repo (default: unused)
(env: LLAMA_ARG_HF_FILE)
-hfv, -hfrv, --hf-repo-v <user>/<model>[:quant]
Hugging Face model repository for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_REPO_V)
-hffv, --hf-file-v FILE Hugging Face model file for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_FILE_V)
-hft, --hf-token TOKEN Hugging Face access token (default: value from HF_TOKEN environment
variable)
(env: HF_TOKEN)
--log-disable Log disable
--log-file FNAME Log to file
--log-colors [on|off|auto] Set colored logging ('on', 'off', or 'auto', default: 'auto')
'auto' enables colors when output is to a terminal
(env: LLAMA_LOG_COLORS)
-v, --verbose, --log-verbose Set verbosity level to infinity (i.e. log all messages, useful for
debugging)
--offline Offline mode: forces use of cache, prevents network access
(env: LLAMA_OFFLINE)
-lv, --verbosity, --log-verbosity N Set the verbosity threshold. Messages with a higher verbosity will be
ignored.
(env: LLAMA_LOG_VERBOSITY)
--log-prefix Enable prefix in log messages
(env: LLAMA_LOG_PREFIX)
--log-timestamps Enable timestamps in log messages
(env: LLAMA_LOG_TIMESTAMPS)
-ctkd, --cache-type-k-draft TYPE KV cache data type for K for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K_DRAFT)
-ctvd, --cache-type-v-draft TYPE KV cache data type for V for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V_DRAFT)

----- sampling params -----

--samplers SAMPLERS samplers that will be used for generation in the order, separated by
';'
(default:
penalties;dry;top_n_sigma;top_k;typ_p;top_p;min_p;xtc;temperature)
-s, --seed SEED RNG seed (default: -1, use random seed for -1)
--sampling-seq, --sampler-seq SEQUENCE
simplified sequence for samplers that will be used (default:
edskypmxt)
--ignore-eos ignore end of stream token and continue generating (implies
--logit-bias EOS-inf)
--temp N temperature (default: 0.8)
--top-k N top-k sampling (default: 40, 0 = disabled)
--top-p N top-p sampling (default: 0.9, 1.0 = disabled)
--min-p N min-p sampling (default: 0.1, 0.0 = disabled)
--top-nsigma N top-n-sigma sampling (default: -1.0, -1.0 = disabled)
--xtc-probability N xtc probability (default: 0.0, 0.0 = disabled)
--xtc-threshold N xtc threshold (default: 0.1, 1.0 = disabled)
--typical N locally typical sampling, parameter p (default: 1.0, 1.0 = disabled)
--repeat-last-n N last n tokens to consider for penalize (default: 64, 0 = disabled, -1
= ctx_size)
--repeat-penalty N penalize repeat sequence of tokens (default: 1.0, 1.0 = disabled)
--presence-penalty N repeat alpha presence penalty (default: 0.0, 0.0 = disabled)
--frequency-penalty N repeat alpha frequency penalty (default: 0.0, 0.0 = disabled)
--dry-multiplier N set DRY sampling multiplier (default: 0.0, 0.0 = disabled)
--dry-base N set DRY sampling base value (default: 1.75)
--dry-allowed-length N set allowed length for DRY sampling (default: 2)
--dry-penalty-last-n N set DRY penalty for the last n tokens (default: -1, 0 = disable, -1 =
context size)
--dry-sequence-breaker STRING add sequence breaker for DRY sampling, clearing out default breakers
('\n', ':', '"', '*') in the process; use "none" to not use any
sequence breakers
--dynatemp-range N dynamic temperature range (default: 0.0, 0.0 = disabled)
--dynatemp-exp N dynamic temperature exponent (default: 1.0)
--mirostat N use Mirostat sampling.
Top K, Nucleus and Locally Typical samplers are ignored if used.
(default: 0, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)
--mirostat-lr N Mirostat learning rate, parameter eta (default: 0.1)
--mirostat-ent N Mirostat target entropy, parameter tau (default: 5.0)
-l, --logit-bias TOKEN_ID(+/-)BIAS modifies the likelihood of token appearing in the completion,
i.e. `--logit-bias 15043+1` to increase likelihood of token ' Hello',
or `--logit-bias 15043-1` to decrease likelihood of token ' Hello'
--grammar GRAMMAR BNF-like grammar to constrain generations (see samples in grammars/
dir) (default: '')
--grammar-file FNAME file to read grammar from
-j, --json-schema SCHEMA JSON schema to constrain generations (https://json-schema.org/), e.g.
`{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead
-jf, --json-schema-file FILE File containing a JSON schema to constrain generations
(https://json-schema.org/), e.g. `{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead

----- example-specific params -----

--ctx-checkpoints, --swa-checkpoints N
max number of context checkpoints to create per slot (default: 8)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/15293)
(env: LLAMA_ARG_CTX_CHECKPOINTS)
--cache-ram, -cram N set the maximum cache size in MiB (default: 8192, -1 - no limit, 0 -
disable)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/16391)
(env: LLAMA_ARG_CACHE_RAM)
--no-context-shift disables context shift on infinite text generation (default: enabled)
(env: LLAMA_ARG_NO_CONTEXT_SHIFT)
--context-shift enables context shift on infinite text generation (default: disabled)
(env: LLAMA_ARG_CONTEXT_SHIFT)
-r, --reverse-prompt PROMPT halt generation at PROMPT, return control in interactive mode
-sp, --special special tokens output enabled (default: false)
--no-warmup skip warming up the model with an empty run
--spm-infill use Suffix/Prefix/Middle pattern for infill (instead of
Prefix/Suffix/Middle) as some models prefer this. (default: disabled)
--pooling {none,mean,cls,last,rank} pooling type for embeddings, use model default if unspecified
(env: LLAMA_ARG_POOLING)
-cb, --cont-batching enable continuous batching (a.k.a dynamic batching) (default: enabled)
(env: LLAMA_ARG_CONT_BATCHING)
-nocb, --no-cont-batching disable continuous batching
(env: LLAMA_ARG_NO_CONT_BATCHING)
--mmproj FILE path to a multimodal projector file. see tools/mtmd/README.md
note: if -hf is used, this argument can be omitted
(env: LLAMA_ARG_MMPROJ)
--mmproj-url URL URL to a multimodal projector file. see tools/mtmd/README.md
(env: LLAMA_ARG_MMPROJ_URL)
--no-mmproj explicitly disable multimodal projector, useful when using -hf
(env: LLAMA_ARG_NO_MMPROJ)
--no-mmproj-offload do not offload multimodal projector to GPU
(env: LLAMA_ARG_NO_MMPROJ_OFFLOAD)
--image-min-tokens N minimum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MIN_TOKENS)
--image-max-tokens N maximum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MAX_TOKENS)
--override-tensor-draft, -otd <tensor name pattern>=<buffer type>,...
override tensor buffer type for draft model
--cpu-moe-draft, -cmoed keep all Mixture of Experts (MoE) weights in the CPU for the draft
model
(env: LLAMA_ARG_CPU_MOE_DRAFT)
--n-cpu-moe-draft, -ncmoed N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU for the draft model
(env: LLAMA_ARG_N_CPU_MOE_DRAFT)
-a, --alias STRING set alias for model name (to be used by REST API)
(env: LLAMA_ARG_ALIAS)
--host HOST ip address to listen, or bind to an UNIX socket if the address ends
with .sock (default: 127.0.0.1)
(env: LLAMA_ARG_HOST)
--port PORT port to listen (default: 8080)
(env: LLAMA_ARG_PORT)
--path PATH path to serve static files from (default: )
(env: LLAMA_ARG_STATIC_PATH)
--api-prefix PREFIX prefix path the server serves from, without the trailing slash
(default: )
(env: LLAMA_ARG_API_PREFIX)
--no-webui Disable the Web UI (default: enabled)
(env: LLAMA_ARG_NO_WEBUI)
--embedding, --embeddings restrict to only support embedding use case; use only with dedicated
embedding models (default: disabled)
(env: LLAMA_ARG_EMBEDDINGS)
--reranking, --rerank enable reranking endpoint on server (default: disabled)
(env: LLAMA_ARG_RERANKING)
--api-key KEY API key to use for authentication (default: none)
(env: LLAMA_API_KEY)
--api-key-file FNAME path to file containing API keys (default: none)
--ssl-key-file FNAME path to file a PEM-encoded SSL private key
(env: LLAMA_ARG_SSL_KEY_FILE)
--ssl-cert-file FNAME path to file a PEM-encoded SSL certificate
(env: LLAMA_ARG_SSL_CERT_FILE)
--chat-template-kwargs STRING sets additional params for the json template parser
(env: LLAMA_CHAT_TEMPLATE_KWARGS)
-to, --timeout N server read/write timeout in seconds (default: 600)
(env: LLAMA_ARG_TIMEOUT)
--threads-http N number of threads used to process HTTP requests (default: -1)
(env: LLAMA_ARG_THREADS_HTTP)
--cache-reuse N min chunk size to attempt reusing from the cache via KV shifting
(default: 0)
[(card)](https://ggml.ai/f0.png)
(env: LLAMA_ARG_CACHE_REUSE)
--metrics enable prometheus compatible metrics endpoint (default: disabled)
(env: LLAMA_ARG_ENDPOINT_METRICS)
--props enable changing global properties via POST /props (default: disabled)
(env: LLAMA_ARG_ENDPOINT_PROPS)
--slots enable slots monitoring endpoint (default: enabled)
(env: LLAMA_ARG_ENDPOINT_SLOTS)
--no-slots disables slots monitoring endpoint
(env: LLAMA_ARG_NO_ENDPOINT_SLOTS)
--slot-save-path PATH path to save slot kv cache (default: disabled)
--jinja use jinja template for chat (default: disabled)
(env: LLAMA_ARG_JINJA)
--reasoning-format FORMAT controls whether thought tags are allowed and/or extracted from the
response, and in which format they're returned; one of:
- none: leaves thoughts unparsed in `message.content`
- deepseek: puts thoughts in `message.reasoning_content`
- deepseek-legacy: keeps `<think>` tags in `message.content` while
also populating `message.reasoning_content`
(default: auto)
(env: LLAMA_ARG_THINK)
--reasoning-budget N controls the amount of thinking allowed; currently only one of: -1 for
unrestricted thinking budget, or 0 to disable thinking (default: -1)
(env: LLAMA_ARG_THINK_BUDGET)
--chat-template JINJA_TEMPLATE set custom jinja chat template (default: template taken from model's
metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE)
--chat-template-file JINJA_TEMPLATE_FILE
set custom jinja chat template file (default: template taken from
model's metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE_FILE)
--no-prefill-assistant whether to prefill the assistant's response if the last message is an
assistant message (default: prefill enabled)
when this flag is set, if the last message is an assistant message
then it will be treated as a full message and not prefilled

(env: LLAMA_ARG_NO_PREFILL_ASSISTANT)
-sps, --slot-prompt-similarity SIMILARITY
how much the prompt of a request must match the prompt of a slot in
order to use that slot (default: 0.10, 0.0 = disabled)
--lora-init-without-apply load LoRA adapters without applying them (apply later via POST
/lora-adapters) (default: disabled)
-td, --threads-draft N number of threads to use during generation (default: same as
--threads)
-tbd, --threads-batch-draft N number of threads to use during batch and prompt processing (default:
same as --threads-draft)
--draft-max, --draft, --draft-n N number of tokens to draft for speculative decoding (default: 16)
(env: LLAMA_ARG_DRAFT_MAX)
--draft-min, --draft-n-min N minimum number of draft tokens to use for speculative decoding
(default: 0)
(env: LLAMA_ARG_DRAFT_MIN)
--draft-p-min P minimum speculative decoding probability (greedy) (default: 0.8)
(env: LLAMA_ARG_DRAFT_P_MIN)
-cd, --ctx-size-draft N size of the prompt context for the draft model (default: 0, 0 = loaded
from model)
(env: LLAMA_ARG_CTX_SIZE_DRAFT)
-devd, --device-draft <dev1,dev2,..> comma-separated list of devices to use for offloading the draft model
(none = don't offload)
use --list-devices to see a list of available devices
-ngld, --gpu-layers-draft, --n-gpu-layers-draft N
number of layers to store in VRAM for the draft model
(env: LLAMA_ARG_N_GPU_LAYERS_DRAFT)
-md, --model-draft FNAME draft model for speculative decoding (default: unused)
(env: LLAMA_ARG_MODEL_DRAFT)
--spec-replace TARGET DRAFT translate the string in TARGET into DRAFT if the draft model and main
model are not compatible
-mv, --model-vocoder FNAME vocoder model for audio generation (default: unused)
--tts-use-guide-tokens Use guide tokens to improve TTS word recall
--embd-gemma-default use default EmbeddingGemma model (note: can download weights from the
internet)
--fim-qwen-1.5b-default use default Qwen 2.5 Coder 1.5B (note: can download weights from the
internet)
--fim-qwen-3b-default use default Qwen 2.5 Coder 3B (note: can download weights from the
internet)
--fim-qwen-7b-default use default Qwen 2.5 Coder 7B (note: can download weights from the
internet)
--fim-qwen-7b-spec use Qwen 2.5 Coder 7B + 0.5B draft for speculative decoding (note: can
download weights from the internet)
--fim-qwen-14b-spec use Qwen 2.5 Coder 14B + 0.5B draft for speculative decoding (note:
can download weights from the internet)
--fim-qwen-30b-default use default Qwen 3 Coder 30B A3B Instruct (note: can download weights
from the internet)
--gpt-oss-20b-default use gpt-oss-20b (note: can download weights from the internet)
--gpt-oss-120b-default use gpt-oss-120b (note: can download weights from the internet)
--vision-gemma-4b-default use Gemma 3 4B QAT (note: can download weights from the internet)
--vision-gemma-12b-default use Gemma 3 12B QAT (note: can download weights from the internet)
</syntaxhighlight>
=== Test ===
=== Bekannte Probleme ===
<syntaxhighlight lang="bash" line="1">
Error 1 [ 34%] Linking HIP shared library ../../../bin/libggml-hip.so [ 34%] Built target ggml-hip gmake[1]: *** [CMakeFiles/Makefile2:1775: ggml/src/CMakeFiles/ggml-cpu.dir/all] Error 2 gmake: *** [Makefile:146: all] Error 2
</syntaxhighlight>
Lösung:
<syntaxhighlight lang="bash" line="1">
GCC 13.3.0 konnte damit nicht umgehen, aber GCC 14.2.0 hat bessere C++-Standard-Compliance und kann diesen Header-Konflikt korrekt auflösen – deshalb funktioniert die Kompilierung jetzt mit der neueren Compiler-Version.
sudo update-alternatives --remove gcc /usr/bin/gcc-13
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-14 100
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-13 50
sudo update-alternatives --config gcc
# Wähle gcc-14

sudo update-alternatives --remove g++ /usr/bin/g++-13
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-14 100
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-13 50
sudo update-alternatives --config g++
# Wähle g++-14
</syntaxhighlight>
=== Nützliche Links ===
* [https://github.com/ggml-org/llama.cpp Offizielles llama.cpp Repository]
* [https://github.com/ggml-org/llama.cpp/blob/master/examples/server/README.md llama-server Dokumentation]
* [https://github.com/ggml-org/llama.cpp/discussions llama.cpp Discussions]
* [https://huggingface.co/models?library=gguf GGUF Modelle auf Hugging Face]
* [https://docs.rocm.com/ ROCm Dokumentation]
* [https://vulkan.lunarg.com/doc/sdk Vulkan SDK Dokumentation]

Llama.cpp

2026-01-07T14:49:43Z

Hendrik: /* ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 */

=== Beschreibung ===
llama.cpp ist eine C/C++ Implementierung für Inference von Large Language Models (LLMs). Es unterstützt verschiedene Backends (CPU, Vulkan, ROCm, CUDA) und ermöglicht das Ausführen von quantisierten Modellen im GGUF-Format.

=== Download ===
<syntaxhighlight lang="bash" line="1">
git clone https://github.com/ggml-org/llama.cpp
cd llama.cpp
</syntaxhighlight>

=== Installation ===
==== Vulkan Build ====
<syntaxhighlight lang="bash" line="1">
sudo dnf install libcurl-devel -y
</syntaxhighlight>
[https://vulkan.lunarg.com/sdk/home Vulkan SDK] herunterladen
entpacken, ins Verzeichnis wechseln und source setup-env.sh ausführen. Dann ins Verzeichnis wechseln wo llama.cpp installiert werden soll.
Dort dann
<syntaxhighlight lang="bash" line="1">
source /home/hendrik/Programme/Vulkan_SDK/1.4.335.0/setup-env.sh
cd /home/hendrik/Programme/llama.cpp/
rm -R build-vulkan
cmake -B build-vulkan -DGGML_VULKAN=1
cmake --build build-vulkan --config Release -- -j $(nproc)
</syntaxhighlight>

==== ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 ====
gfx906 = Mi 50 Support 
gfx1100 = 7900 XTX Support

<syntaxhighlight lang="bash" line="1">
Copy & paste all the commands from the quick install https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/quick-start.html
Before rebooting to complete the install, download the 6.4 rocblas from the AUR: https://archlinux.org/packages/extra/x86_64/rocblas/
Extract it
Copy all tensor files that contain gfx906 in rocblas-6.4.3-3-x86_64.pkg/opt/rocm/lib/rocblas/library to /opt/rocm/lib/rocblas/library
Reboot
Check if it worked by running sudo update-alternatives --display rocm
</syntaxhighlight>
# ROCm Umgebung einrichten (optional falls Fehler auftreten)
<syntaxhighlight lang="bash" line="1">
echo 'export LD_LIBRARY_PATH=/opt/rocm/lib:$LD_LIBRARY_PATH' >> ~/.bashrc
echo 'export HSA_OVERRIDE_GFX_VERSION=9.0.6' >> ~/.bashrc # Für MI50/MI60
source ~/.bashrc
</syntaxhighlight>
<syntaxhighlight lang="bash" line="1">
sudo dnf install rocwmma-devel -y
</syntaxhighlight>
# llama.cpp kompilieren
<syntaxhighlight lang="bash" line="1">
cd /home/hendrik/Programme/llama.cpp/ && rm -R build-rocm
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" cmake -B build-rocm -DGGML_HIP=ON -DAMDGPU_TARGETS=gfx1100 -DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 8
</syntaxhighlight>

Für mehrere GPU-Architekturen gleichzeitig:
<syntaxhighlight lang="bash" line="1">
# Für MI50 (gfx906) und RX 7900 XTX (gfx1100)
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
cmake -S . -B build -DGGML_HIP=ON -DAMDGPU_TARGETS="gfx906;gfx1100" \
-DCMAKE_BUILD_TYPE=Release && \
cmake --build build --config Release -- -j 8
</syntaxhighlight>

=== Konfiguration ===

==== llama-server als systemd Service einrichten ====

'''Service-Datei erstellen:'''
<syntaxhighlight lang="bash" line="1">
sudo nano /etc/systemd/system/llama-server.service
</syntaxhighlight>

'''Service-Datei Inhalt (Multi-GPU mit ROCm):'''
<syntaxhighlight lang="ini" line="1">
[Unit]
Description=Llama.cpp ROCm Multi-GPU Server
After=network.target

[Service]
Type=simple
User=username
Group=username
WorkingDirectory=/home/username/llama.cpp/build/bin

# Multi-GPU Konfiguration
Environment="HIP_VISIBLE_DEVICES=0,1"
Environment="HSA_OVERRIDE_GFX_VERSION=9.0.6"
Environment="PATH=/opt/rocm/bin:/usr/local/bin:/usr/bin:/bin"
Environment="LD_LIBRARY_PATH=/opt/rocm/lib"

# Server mit optimalen Multi-GPU Einstellungen
ExecStart=/home/username/llama.cpp/build/bin/llama-server \
-m /home/username/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja

Restart=always
RestartSec=10
LimitNOFILE=65535
LimitMEMLOCK=infinity
StandardOutput=journal
StandardError=journal
SyslogIdentifier=llama-server

[Install]
WantedBy=multi-user.target
</syntaxhighlight>

'''Service aktivieren und starten:'''
<syntaxhighlight lang="bash" line="1">
# Service neu laden
sudo systemctl daemon-reload

# Service aktivieren (Auto-Start beim Boot)
sudo systemctl enable llama-server

# Service starten
sudo systemctl start llama-server

# Status prüfen
sudo systemctl status llama-server

# Logs anzeigen
sudo journalctl -u llama-server -f
</syntaxhighlight>

'''Service Management:'''
<syntaxhighlight lang="bash" line="1">
# Service stoppen
sudo systemctl stop llama-server

# Service neustarten
sudo systemctl restart llama-server

# Service deaktivieren
sudo systemctl disable llama-server

==== Manuelle Server-Starts ====
'''Multi-GPU (ROCm) - Optimal:'''
<syntaxhighlight lang="bash" line="1">
HIP_VISIBLE_DEVICES=0,1 ./llama-server \
-m ~/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja
</syntaxhighlight>

=== Update ===
<syntaxhighlight lang="bash" line="1">
cd ~/llama.cpp
git pull
cmake --build build --config Release -- -j $(nproc)

# Service neu starten falls aktiv
sudo systemctl restart llama-server
</syntaxhighlight>
=== Befehlszeilenargumente ===
llama-server
<syntaxhighlight lang="bash" line="1">
-h, --help, --usage print usage and exit
--version show version and build info
-cl, --cache-list show list of models in cache
--completion-bash print source-able bash completion script for llama.cpp
--verbose-prompt print a verbose prompt before generation (default: false)
-t, --threads N number of CPU threads to use during generation (default: -1)
(env: LLAMA_ARG_THREADS)
-tb, --threads-batch N number of threads to use during batch and prompt processing (default:
same as --threads)
-C, --cpu-mask M CPU affinity mask: arbitrarily long hex. Complements cpu-range
(default: "")
-Cr, --cpu-range lo-hi range of CPUs for affinity. Complements --cpu-mask
--cpu-strict <0|1> use strict CPU placement (default: 0)
--prio N set process/thread priority : low(-1), normal(0), medium(1), high(2),
realtime(3) (default: 0)
--poll <0...100> use polling level to wait for work (0 - no polling, default: 50)
-Cb, --cpu-mask-batch M CPU affinity mask: arbitrarily long hex. Complements cpu-range-batch
(default: same as --cpu-mask)
-Crb, --cpu-range-batch lo-hi ranges of CPUs for affinity. Complements --cpu-mask-batch
--cpu-strict-batch <0|1> use strict CPU placement (default: same as --cpu-strict)
--prio-batch N set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime
(default: 0)
--poll-batch <0|1> use polling to wait for work (default: same as --poll)
-c, --ctx-size N size of the prompt context (default: 4096, 0 = loaded from model)
(env: LLAMA_ARG_CTX_SIZE)
-n, --predict, --n-predict N number of tokens to predict (default: -1, -1 = infinity)
(env: LLAMA_ARG_N_PREDICT)
-b, --batch-size N logical maximum batch size (default: 2048)
(env: LLAMA_ARG_BATCH)
-ub, --ubatch-size N physical maximum batch size (default: 512)
(env: LLAMA_ARG_UBATCH)
--keep N number of tokens to keep from the initial prompt (default: 0, -1 =
all)
--swa-full use full-size SWA cache (default: false)
[(more
info)](https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
(env: LLAMA_ARG_SWA_FULL)
--kv-unified, -kvu use single unified KV buffer for the KV cache of all sequences
(default: false)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/14363)
(env: LLAMA_ARG_KV_SPLIT)
-fa, --flash-attn [on|off|auto] set Flash Attention use ('on', 'off', or 'auto', default: 'auto')
(env: LLAMA_ARG_FLASH_ATTN)
--no-perf disable internal libllama performance timings (default: false)
(env: LLAMA_ARG_NO_PERF)
-e, --escape process escapes sequences (\n, \r, \t, \', \", \\) (default: true)
--no-escape do not process escape sequences
--rope-scaling {none,linear,yarn} RoPE frequency scaling method, defaults to linear unless specified by
the model
(env: LLAMA_ARG_ROPE_SCALING_TYPE)
--rope-scale N RoPE context scaling factor, expands context by a factor of N
(env: LLAMA_ARG_ROPE_SCALE)
--rope-freq-base N RoPE base frequency, used by NTK-aware scaling (default: loaded from
model)
(env: LLAMA_ARG_ROPE_FREQ_BASE)
--rope-freq-scale N RoPE frequency scaling factor, expands context by a factor of 1/N
(env: LLAMA_ARG_ROPE_FREQ_SCALE)
--yarn-orig-ctx N YaRN: original context size of model (default: 0 = model training
context size)
(env: LLAMA_ARG_YARN_ORIG_CTX)
--yarn-ext-factor N YaRN: extrapolation mix factor (default: -1.0, 0.0 = full
interpolation)
(env: LLAMA_ARG_YARN_EXT_FACTOR)
--yarn-attn-factor N YaRN: scale sqrt(t) or attention magnitude (default: -1.0)
(env: LLAMA_ARG_YARN_ATTN_FACTOR)
--yarn-beta-slow N YaRN: high correction dim or alpha (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_SLOW)
--yarn-beta-fast N YaRN: low correction dim or beta (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_FAST)
-nkvo, --no-kv-offload disable KV offload
(env: LLAMA_ARG_NO_KV_OFFLOAD)
-nr, --no-repack disable weight repacking
(env: LLAMA_ARG_NO_REPACK)
--no-host bypass host buffer allowing extra buffers to be used
(env: LLAMA_ARG_NO_HOST)
-ctk, --cache-type-k TYPE KV cache data type for K
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K)
-ctv, --cache-type-v TYPE KV cache data type for V
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V)
-dt, --defrag-thold N KV cache defragmentation threshold (DEPRECATED)
(env: LLAMA_ARG_DEFRAG_THOLD)
-np, --parallel N number of parallel sequences to decode (default: 1)
(env: LLAMA_ARG_N_PARALLEL)
--mlock force system to keep model in RAM rather than swapping or compressing
(env: LLAMA_ARG_MLOCK)
--no-mmap do not memory-map model (slower load but may reduce pageouts if not
using mlock)
(env: LLAMA_ARG_NO_MMAP)
--numa TYPE attempt optimizations that help on some NUMA systems
- distribute: spread execution evenly over all nodes
- isolate: only spawn threads on CPUs on the node that execution
started on
- numactl: use the CPU map provided by numactl
if run without this previously, it is recommended to drop the system
page cache before using this
see https://github.com/ggml-org/llama.cpp/issues/1437
(env: LLAMA_ARG_NUMA)
-dev, --device <dev1,dev2,..> comma-separated list of devices to use for offloading (none = don't
offload)
use --list-devices to see a list of available devices
(env: LLAMA_ARG_DEVICE)
--list-devices print list of available devices and exit
--override-tensor, -ot <tensor name pattern>=<buffer type>,...
override tensor buffer type
--cpu-moe, -cmoe keep all Mixture of Experts (MoE) weights in the CPU
(env: LLAMA_ARG_CPU_MOE)
--n-cpu-moe, -ncmoe N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU
(env: LLAMA_ARG_N_CPU_MOE)
-ngl, --gpu-layers, --n-gpu-layers N max. number of layers to store in VRAM (default: -1)
(env: LLAMA_ARG_N_GPU_LAYERS)
-sm, --split-mode {none,layer,row} how to split the model across multiple GPUs, one of:
- none: use one GPU only
- layer (default): split layers and KV across GPUs
- row: split rows across GPUs
(env: LLAMA_ARG_SPLIT_MODE)
-ts, --tensor-split N0,N1,N2,... fraction of the model to offload to each GPU, comma-separated list of
proportions, e.g. 3,1
(env: LLAMA_ARG_TENSOR_SPLIT)
-mg, --main-gpu INDEX the GPU to use for the model (with split-mode = none), or for
intermediate results and KV (with split-mode = row) (default: 0)
(env: LLAMA_ARG_MAIN_GPU)
--check-tensors check model tensor data for invalid values (default: false)
--override-kv KEY=TYPE:VALUE advanced option to override model metadata by key. may be specified
multiple times.
types: int, float, bool, str. example: --override-kv
tokenizer.ggml.add_bos_token=bool:false
--no-op-offload disable offloading host tensor operations to device (default: false)
--lora FNAME path to LoRA adapter (can be repeated to use multiple adapters)
--lora-scaled FNAME SCALE path to LoRA adapter with user defined scaling (can be repeated to use
multiple adapters)
--control-vector FNAME add a control vector
note: this argument can be repeated to add multiple control vectors
--control-vector-scaled FNAME SCALE add a control vector with user defined scaling SCALE
note: this argument can be repeated to add multiple scaled control
vectors
--control-vector-layer-range START END
layer range to apply the control vector(s) to, start and end inclusive
-m, --model FNAME model path (default: `models/$filename` with filename from `--hf-file`
or `--model-url` if set, otherwise models/7B/ggml-model-f16.gguf)
(env: LLAMA_ARG_MODEL)
-mu, --model-url MODEL_URL model download url (default: unused)
(env: LLAMA_ARG_MODEL_URL)
-dr, --docker-repo [<repo>/]<model>[:quant]
Docker Hub model repository. repo is optional, default to ai/. quant
is optional, default to :latest.
example: gemma3
(default: unused)
(env: LLAMA_ARG_DOCKER_REPO)
-hf, -hfr, --hf-repo <user>/<model>[:quant]
Hugging Face model repository; quant is optional, case-insensitive,
default to Q4_K_M, or falls back to the first file in the repo if
Q4_K_M doesn't exist.
mmproj is also downloaded automatically if available. to disable, add
--no-mmproj
example: unsloth/phi-4-GGUF:q4_k_m
(default: unused)
(env: LLAMA_ARG_HF_REPO)
-hfd, -hfrd, --hf-repo-draft <user>/<model>[:quant]
Same as --hf-repo, but for the draft model (default: unused)
(env: LLAMA_ARG_HFD_REPO)
-hff, --hf-file FILE Hugging Face model file. If specified, it will override the quant in
--hf-repo (default: unused)
(env: LLAMA_ARG_HF_FILE)
-hfv, -hfrv, --hf-repo-v <user>/<model>[:quant]
Hugging Face model repository for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_REPO_V)
-hffv, --hf-file-v FILE Hugging Face model file for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_FILE_V)
-hft, --hf-token TOKEN Hugging Face access token (default: value from HF_TOKEN environment
variable)
(env: HF_TOKEN)
--log-disable Log disable
--log-file FNAME Log to file
--log-colors [on|off|auto] Set colored logging ('on', 'off', or 'auto', default: 'auto')
'auto' enables colors when output is to a terminal
(env: LLAMA_LOG_COLORS)
-v, --verbose, --log-verbose Set verbosity level to infinity (i.e. log all messages, useful for
debugging)
--offline Offline mode: forces use of cache, prevents network access
(env: LLAMA_OFFLINE)
-lv, --verbosity, --log-verbosity N Set the verbosity threshold. Messages with a higher verbosity will be
ignored.
(env: LLAMA_LOG_VERBOSITY)
--log-prefix Enable prefix in log messages
(env: LLAMA_LOG_PREFIX)
--log-timestamps Enable timestamps in log messages
(env: LLAMA_LOG_TIMESTAMPS)
-ctkd, --cache-type-k-draft TYPE KV cache data type for K for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K_DRAFT)
-ctvd, --cache-type-v-draft TYPE KV cache data type for V for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V_DRAFT)

----- sampling params -----

--samplers SAMPLERS samplers that will be used for generation in the order, separated by
';'
(default:
penalties;dry;top_n_sigma;top_k;typ_p;top_p;min_p;xtc;temperature)
-s, --seed SEED RNG seed (default: -1, use random seed for -1)
--sampling-seq, --sampler-seq SEQUENCE
simplified sequence for samplers that will be used (default:
edskypmxt)
--ignore-eos ignore end of stream token and continue generating (implies
--logit-bias EOS-inf)
--temp N temperature (default: 0.8)
--top-k N top-k sampling (default: 40, 0 = disabled)
--top-p N top-p sampling (default: 0.9, 1.0 = disabled)
--min-p N min-p sampling (default: 0.1, 0.0 = disabled)
--top-nsigma N top-n-sigma sampling (default: -1.0, -1.0 = disabled)
--xtc-probability N xtc probability (default: 0.0, 0.0 = disabled)
--xtc-threshold N xtc threshold (default: 0.1, 1.0 = disabled)
--typical N locally typical sampling, parameter p (default: 1.0, 1.0 = disabled)
--repeat-last-n N last n tokens to consider for penalize (default: 64, 0 = disabled, -1
= ctx_size)
--repeat-penalty N penalize repeat sequence of tokens (default: 1.0, 1.0 = disabled)
--presence-penalty N repeat alpha presence penalty (default: 0.0, 0.0 = disabled)
--frequency-penalty N repeat alpha frequency penalty (default: 0.0, 0.0 = disabled)
--dry-multiplier N set DRY sampling multiplier (default: 0.0, 0.0 = disabled)
--dry-base N set DRY sampling base value (default: 1.75)
--dry-allowed-length N set allowed length for DRY sampling (default: 2)
--dry-penalty-last-n N set DRY penalty for the last n tokens (default: -1, 0 = disable, -1 =
context size)
--dry-sequence-breaker STRING add sequence breaker for DRY sampling, clearing out default breakers
('\n', ':', '"', '*') in the process; use "none" to not use any
sequence breakers
--dynatemp-range N dynamic temperature range (default: 0.0, 0.0 = disabled)
--dynatemp-exp N dynamic temperature exponent (default: 1.0)
--mirostat N use Mirostat sampling.
Top K, Nucleus and Locally Typical samplers are ignored if used.
(default: 0, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)
--mirostat-lr N Mirostat learning rate, parameter eta (default: 0.1)
--mirostat-ent N Mirostat target entropy, parameter tau (default: 5.0)
-l, --logit-bias TOKEN_ID(+/-)BIAS modifies the likelihood of token appearing in the completion,
i.e. `--logit-bias 15043+1` to increase likelihood of token ' Hello',
or `--logit-bias 15043-1` to decrease likelihood of token ' Hello'
--grammar GRAMMAR BNF-like grammar to constrain generations (see samples in grammars/
dir) (default: '')
--grammar-file FNAME file to read grammar from
-j, --json-schema SCHEMA JSON schema to constrain generations (https://json-schema.org/), e.g.
`{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead
-jf, --json-schema-file FILE File containing a JSON schema to constrain generations
(https://json-schema.org/), e.g. `{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead

----- example-specific params -----

--ctx-checkpoints, --swa-checkpoints N
max number of context checkpoints to create per slot (default: 8)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/15293)
(env: LLAMA_ARG_CTX_CHECKPOINTS)
--cache-ram, -cram N set the maximum cache size in MiB (default: 8192, -1 - no limit, 0 -
disable)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/16391)
(env: LLAMA_ARG_CACHE_RAM)
--no-context-shift disables context shift on infinite text generation (default: enabled)
(env: LLAMA_ARG_NO_CONTEXT_SHIFT)
--context-shift enables context shift on infinite text generation (default: disabled)
(env: LLAMA_ARG_CONTEXT_SHIFT)
-r, --reverse-prompt PROMPT halt generation at PROMPT, return control in interactive mode
-sp, --special special tokens output enabled (default: false)
--no-warmup skip warming up the model with an empty run
--spm-infill use Suffix/Prefix/Middle pattern for infill (instead of
Prefix/Suffix/Middle) as some models prefer this. (default: disabled)
--pooling {none,mean,cls,last,rank} pooling type for embeddings, use model default if unspecified
(env: LLAMA_ARG_POOLING)
-cb, --cont-batching enable continuous batching (a.k.a dynamic batching) (default: enabled)
(env: LLAMA_ARG_CONT_BATCHING)
-nocb, --no-cont-batching disable continuous batching
(env: LLAMA_ARG_NO_CONT_BATCHING)
--mmproj FILE path to a multimodal projector file. see tools/mtmd/README.md
note: if -hf is used, this argument can be omitted
(env: LLAMA_ARG_MMPROJ)
--mmproj-url URL URL to a multimodal projector file. see tools/mtmd/README.md
(env: LLAMA_ARG_MMPROJ_URL)
--no-mmproj explicitly disable multimodal projector, useful when using -hf
(env: LLAMA_ARG_NO_MMPROJ)
--no-mmproj-offload do not offload multimodal projector to GPU
(env: LLAMA_ARG_NO_MMPROJ_OFFLOAD)
--image-min-tokens N minimum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MIN_TOKENS)
--image-max-tokens N maximum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MAX_TOKENS)
--override-tensor-draft, -otd <tensor name pattern>=<buffer type>,...
override tensor buffer type for draft model
--cpu-moe-draft, -cmoed keep all Mixture of Experts (MoE) weights in the CPU for the draft
model
(env: LLAMA_ARG_CPU_MOE_DRAFT)
--n-cpu-moe-draft, -ncmoed N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU for the draft model
(env: LLAMA_ARG_N_CPU_MOE_DRAFT)
-a, --alias STRING set alias for model name (to be used by REST API)
(env: LLAMA_ARG_ALIAS)
--host HOST ip address to listen, or bind to an UNIX socket if the address ends
with .sock (default: 127.0.0.1)
(env: LLAMA_ARG_HOST)
--port PORT port to listen (default: 8080)
(env: LLAMA_ARG_PORT)
--path PATH path to serve static files from (default: )
(env: LLAMA_ARG_STATIC_PATH)
--api-prefix PREFIX prefix path the server serves from, without the trailing slash
(default: )
(env: LLAMA_ARG_API_PREFIX)
--no-webui Disable the Web UI (default: enabled)
(env: LLAMA_ARG_NO_WEBUI)
--embedding, --embeddings restrict to only support embedding use case; use only with dedicated
embedding models (default: disabled)
(env: LLAMA_ARG_EMBEDDINGS)
--reranking, --rerank enable reranking endpoint on server (default: disabled)
(env: LLAMA_ARG_RERANKING)
--api-key KEY API key to use for authentication (default: none)
(env: LLAMA_API_KEY)
--api-key-file FNAME path to file containing API keys (default: none)
--ssl-key-file FNAME path to file a PEM-encoded SSL private key
(env: LLAMA_ARG_SSL_KEY_FILE)
--ssl-cert-file FNAME path to file a PEM-encoded SSL certificate
(env: LLAMA_ARG_SSL_CERT_FILE)
--chat-template-kwargs STRING sets additional params for the json template parser
(env: LLAMA_CHAT_TEMPLATE_KWARGS)
-to, --timeout N server read/write timeout in seconds (default: 600)
(env: LLAMA_ARG_TIMEOUT)
--threads-http N number of threads used to process HTTP requests (default: -1)
(env: LLAMA_ARG_THREADS_HTTP)
--cache-reuse N min chunk size to attempt reusing from the cache via KV shifting
(default: 0)
[(card)](https://ggml.ai/f0.png)
(env: LLAMA_ARG_CACHE_REUSE)
--metrics enable prometheus compatible metrics endpoint (default: disabled)
(env: LLAMA_ARG_ENDPOINT_METRICS)
--props enable changing global properties via POST /props (default: disabled)
(env: LLAMA_ARG_ENDPOINT_PROPS)
--slots enable slots monitoring endpoint (default: enabled)
(env: LLAMA_ARG_ENDPOINT_SLOTS)
--no-slots disables slots monitoring endpoint
(env: LLAMA_ARG_NO_ENDPOINT_SLOTS)
--slot-save-path PATH path to save slot kv cache (default: disabled)
--jinja use jinja template for chat (default: disabled)
(env: LLAMA_ARG_JINJA)
--reasoning-format FORMAT controls whether thought tags are allowed and/or extracted from the
response, and in which format they're returned; one of:
- none: leaves thoughts unparsed in `message.content`
- deepseek: puts thoughts in `message.reasoning_content`
- deepseek-legacy: keeps `<think>` tags in `message.content` while
also populating `message.reasoning_content`
(default: auto)
(env: LLAMA_ARG_THINK)
--reasoning-budget N controls the amount of thinking allowed; currently only one of: -1 for
unrestricted thinking budget, or 0 to disable thinking (default: -1)
(env: LLAMA_ARG_THINK_BUDGET)
--chat-template JINJA_TEMPLATE set custom jinja chat template (default: template taken from model's
metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE)
--chat-template-file JINJA_TEMPLATE_FILE
set custom jinja chat template file (default: template taken from
model's metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE_FILE)
--no-prefill-assistant whether to prefill the assistant's response if the last message is an
assistant message (default: prefill enabled)
when this flag is set, if the last message is an assistant message
then it will be treated as a full message and not prefilled

(env: LLAMA_ARG_NO_PREFILL_ASSISTANT)
-sps, --slot-prompt-similarity SIMILARITY
how much the prompt of a request must match the prompt of a slot in
order to use that slot (default: 0.10, 0.0 = disabled)
--lora-init-without-apply load LoRA adapters without applying them (apply later via POST
/lora-adapters) (default: disabled)
-td, --threads-draft N number of threads to use during generation (default: same as
--threads)
-tbd, --threads-batch-draft N number of threads to use during batch and prompt processing (default:
same as --threads-draft)
--draft-max, --draft, --draft-n N number of tokens to draft for speculative decoding (default: 16)
(env: LLAMA_ARG_DRAFT_MAX)
--draft-min, --draft-n-min N minimum number of draft tokens to use for speculative decoding
(default: 0)
(env: LLAMA_ARG_DRAFT_MIN)
--draft-p-min P minimum speculative decoding probability (greedy) (default: 0.8)
(env: LLAMA_ARG_DRAFT_P_MIN)
-cd, --ctx-size-draft N size of the prompt context for the draft model (default: 0, 0 = loaded
from model)
(env: LLAMA_ARG_CTX_SIZE_DRAFT)
-devd, --device-draft <dev1,dev2,..> comma-separated list of devices to use for offloading the draft model
(none = don't offload)
use --list-devices to see a list of available devices
-ngld, --gpu-layers-draft, --n-gpu-layers-draft N
number of layers to store in VRAM for the draft model
(env: LLAMA_ARG_N_GPU_LAYERS_DRAFT)
-md, --model-draft FNAME draft model for speculative decoding (default: unused)
(env: LLAMA_ARG_MODEL_DRAFT)
--spec-replace TARGET DRAFT translate the string in TARGET into DRAFT if the draft model and main
model are not compatible
-mv, --model-vocoder FNAME vocoder model for audio generation (default: unused)
--tts-use-guide-tokens Use guide tokens to improve TTS word recall
--embd-gemma-default use default EmbeddingGemma model (note: can download weights from the
internet)
--fim-qwen-1.5b-default use default Qwen 2.5 Coder 1.5B (note: can download weights from the
internet)
--fim-qwen-3b-default use default Qwen 2.5 Coder 3B (note: can download weights from the
internet)
--fim-qwen-7b-default use default Qwen 2.5 Coder 7B (note: can download weights from the
internet)
--fim-qwen-7b-spec use Qwen 2.5 Coder 7B + 0.5B draft for speculative decoding (note: can
download weights from the internet)
--fim-qwen-14b-spec use Qwen 2.5 Coder 14B + 0.5B draft for speculative decoding (note:
can download weights from the internet)
--fim-qwen-30b-default use default Qwen 3 Coder 30B A3B Instruct (note: can download weights
from the internet)
--gpt-oss-20b-default use gpt-oss-20b (note: can download weights from the internet)
--gpt-oss-120b-default use gpt-oss-120b (note: can download weights from the internet)
--vision-gemma-4b-default use Gemma 3 4B QAT (note: can download weights from the internet)
--vision-gemma-12b-default use Gemma 3 12B QAT (note: can download weights from the internet)
</syntaxhighlight>
=== Test ===
=== Bekannte Probleme ===
<syntaxhighlight lang="bash" line="1">
Error 1 [ 34%] Linking HIP shared library ../../../bin/libggml-hip.so [ 34%] Built target ggml-hip gmake[1]: *** [CMakeFiles/Makefile2:1775: ggml/src/CMakeFiles/ggml-cpu.dir/all] Error 2 gmake: *** [Makefile:146: all] Error 2
</syntaxhighlight>
Lösung:
<syntaxhighlight lang="bash" line="1">
GCC 13.3.0 konnte damit nicht umgehen, aber GCC 14.2.0 hat bessere C++-Standard-Compliance und kann diesen Header-Konflikt korrekt auflösen – deshalb funktioniert die Kompilierung jetzt mit der neueren Compiler-Version.
sudo update-alternatives --remove gcc /usr/bin/gcc-13
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-14 100
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-13 50
sudo update-alternatives --config gcc
# Wähle gcc-14

sudo update-alternatives --remove g++ /usr/bin/g++-13
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-14 100
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-13 50
sudo update-alternatives --config g++
# Wähle g++-14
</syntaxhighlight>
=== Nützliche Links ===
* [https://github.com/ggml-org/llama.cpp Offizielles llama.cpp Repository]
* [https://github.com/ggml-org/llama.cpp/blob/master/examples/server/README.md llama-server Dokumentation]
* [https://github.com/ggml-org/llama.cpp/discussions llama.cpp Discussions]
* [https://huggingface.co/models?library=gguf GGUF Modelle auf Hugging Face]
* [https://docs.rocm.com/ ROCm Dokumentation]
* [https://vulkan.lunarg.com/doc/sdk Vulkan SDK Dokumentation]

Llama.cpp

2026-01-07T14:36:00Z

Hendrik: /* ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 */

=== Beschreibung ===
llama.cpp ist eine C/C++ Implementierung für Inference von Large Language Models (LLMs). Es unterstützt verschiedene Backends (CPU, Vulkan, ROCm, CUDA) und ermöglicht das Ausführen von quantisierten Modellen im GGUF-Format.

=== Download ===
<syntaxhighlight lang="bash" line="1">
git clone https://github.com/ggml-org/llama.cpp
cd llama.cpp
</syntaxhighlight>

=== Installation ===
==== Vulkan Build ====
<syntaxhighlight lang="bash" line="1">
sudo dnf install libcurl-devel -y
</syntaxhighlight>
[https://vulkan.lunarg.com/sdk/home Vulkan SDK] herunterladen
entpacken, ins Verzeichnis wechseln und source setup-env.sh ausführen. Dann ins Verzeichnis wechseln wo llama.cpp installiert werden soll.
Dort dann
<syntaxhighlight lang="bash" line="1">
source /home/hendrik/Programme/Vulkan_SDK/1.4.335.0/setup-env.sh
cd /home/hendrik/Programme/llama.cpp/
rm -R build-vulkan
cmake -B build-vulkan -DGGML_VULKAN=1
cmake --build build-vulkan --config Release -- -j $(nproc)
</syntaxhighlight>

==== ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 ====
gfx906 = Mi 50 Support 
gfx1100 = 7900 XTX Support

<syntaxhighlight lang="bash" line="1">
Copy & paste all the commands from the quick install https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/quick-start.html
Before rebooting to complete the install, download the 6.4 rocblas from the AUR: https://archlinux.org/packages/extra/x86_64/rocblas/
Extract it
Copy all tensor files that contain gfx906 in rocblas-6.4.3-3-x86_64.pkg/opt/rocm/lib/rocblas/library to /opt/rocm/lib/rocblas/library
Reboot
Check if it worked by running sudo update-alternatives --display rocm
</syntaxhighlight>
# ROCm Umgebung einrichten (optional falls Fehler auftreten)
<syntaxhighlight lang="bash" line="1">
echo 'export LD_LIBRARY_PATH=/opt/rocm/lib:$LD_LIBRARY_PATH' >> ~/.bashrc
echo 'export HSA_OVERRIDE_GFX_VERSION=9.0.6' >> ~/.bashrc # Für MI50/MI60
source ~/.bashrc
</syntaxhighlight>
<syntaxhighlight lang="bash" line="1">
sudo dnf install rocwmma-devel -y
</syntaxhighlight>
# llama.cpp kompilieren
<syntaxhighlight lang="bash" line="1">
cd /home/hendrik/Programme/llama.cpp/ && rm -R build-rocm
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
cmake -B build-rocm -DGGML_HIP=ON -DAMDGPU_TARGETS=gfx906 -DGGML_HIP_ROCWMMA_FATTN=ON \
-DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 8
</syntaxhighlight>

Für mehrere GPU-Architekturen gleichzeitig:
<syntaxhighlight lang="bash" line="1">
# Für MI50 (gfx906) und RX 7900 XTX (gfx1100)
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
cmake -S . -B build -DGGML_HIP=ON -DAMDGPU_TARGETS="gfx906;gfx1100" \
-DCMAKE_BUILD_TYPE=Release && \
cmake --build build --config Release -- -j 8
</syntaxhighlight>

=== Konfiguration ===

==== llama-server als systemd Service einrichten ====

'''Service-Datei erstellen:'''
<syntaxhighlight lang="bash" line="1">
sudo nano /etc/systemd/system/llama-server.service
</syntaxhighlight>

'''Service-Datei Inhalt (Multi-GPU mit ROCm):'''
<syntaxhighlight lang="ini" line="1">
[Unit]
Description=Llama.cpp ROCm Multi-GPU Server
After=network.target

[Service]
Type=simple
User=username
Group=username
WorkingDirectory=/home/username/llama.cpp/build/bin

# Multi-GPU Konfiguration
Environment="HIP_VISIBLE_DEVICES=0,1"
Environment="HSA_OVERRIDE_GFX_VERSION=9.0.6"
Environment="PATH=/opt/rocm/bin:/usr/local/bin:/usr/bin:/bin"
Environment="LD_LIBRARY_PATH=/opt/rocm/lib"

# Server mit optimalen Multi-GPU Einstellungen
ExecStart=/home/username/llama.cpp/build/bin/llama-server \
-m /home/username/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja

Restart=always
RestartSec=10
LimitNOFILE=65535
LimitMEMLOCK=infinity
StandardOutput=journal
StandardError=journal
SyslogIdentifier=llama-server

[Install]
WantedBy=multi-user.target
</syntaxhighlight>

'''Service aktivieren und starten:'''
<syntaxhighlight lang="bash" line="1">
# Service neu laden
sudo systemctl daemon-reload

# Service aktivieren (Auto-Start beim Boot)
sudo systemctl enable llama-server

# Service starten
sudo systemctl start llama-server

# Status prüfen
sudo systemctl status llama-server

# Logs anzeigen
sudo journalctl -u llama-server -f
</syntaxhighlight>

'''Service Management:'''
<syntaxhighlight lang="bash" line="1">
# Service stoppen
sudo systemctl stop llama-server

# Service neustarten
sudo systemctl restart llama-server

# Service deaktivieren
sudo systemctl disable llama-server

==== Manuelle Server-Starts ====
'''Multi-GPU (ROCm) - Optimal:'''
<syntaxhighlight lang="bash" line="1">
HIP_VISIBLE_DEVICES=0,1 ./llama-server \
-m ~/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja
</syntaxhighlight>

=== Update ===
<syntaxhighlight lang="bash" line="1">
cd ~/llama.cpp
git pull
cmake --build build --config Release -- -j $(nproc)

# Service neu starten falls aktiv
sudo systemctl restart llama-server
</syntaxhighlight>
=== Befehlszeilenargumente ===
llama-server
<syntaxhighlight lang="bash" line="1">
-h, --help, --usage print usage and exit
--version show version and build info
-cl, --cache-list show list of models in cache
--completion-bash print source-able bash completion script for llama.cpp
--verbose-prompt print a verbose prompt before generation (default: false)
-t, --threads N number of CPU threads to use during generation (default: -1)
(env: LLAMA_ARG_THREADS)
-tb, --threads-batch N number of threads to use during batch and prompt processing (default:
same as --threads)
-C, --cpu-mask M CPU affinity mask: arbitrarily long hex. Complements cpu-range
(default: "")
-Cr, --cpu-range lo-hi range of CPUs for affinity. Complements --cpu-mask
--cpu-strict <0|1> use strict CPU placement (default: 0)
--prio N set process/thread priority : low(-1), normal(0), medium(1), high(2),
realtime(3) (default: 0)
--poll <0...100> use polling level to wait for work (0 - no polling, default: 50)
-Cb, --cpu-mask-batch M CPU affinity mask: arbitrarily long hex. Complements cpu-range-batch
(default: same as --cpu-mask)
-Crb, --cpu-range-batch lo-hi ranges of CPUs for affinity. Complements --cpu-mask-batch
--cpu-strict-batch <0|1> use strict CPU placement (default: same as --cpu-strict)
--prio-batch N set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime
(default: 0)
--poll-batch <0|1> use polling to wait for work (default: same as --poll)
-c, --ctx-size N size of the prompt context (default: 4096, 0 = loaded from model)
(env: LLAMA_ARG_CTX_SIZE)
-n, --predict, --n-predict N number of tokens to predict (default: -1, -1 = infinity)
(env: LLAMA_ARG_N_PREDICT)
-b, --batch-size N logical maximum batch size (default: 2048)
(env: LLAMA_ARG_BATCH)
-ub, --ubatch-size N physical maximum batch size (default: 512)
(env: LLAMA_ARG_UBATCH)
--keep N number of tokens to keep from the initial prompt (default: 0, -1 =
all)
--swa-full use full-size SWA cache (default: false)
[(more
info)](https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
(env: LLAMA_ARG_SWA_FULL)
--kv-unified, -kvu use single unified KV buffer for the KV cache of all sequences
(default: false)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/14363)
(env: LLAMA_ARG_KV_SPLIT)
-fa, --flash-attn [on|off|auto] set Flash Attention use ('on', 'off', or 'auto', default: 'auto')
(env: LLAMA_ARG_FLASH_ATTN)
--no-perf disable internal libllama performance timings (default: false)
(env: LLAMA_ARG_NO_PERF)
-e, --escape process escapes sequences (\n, \r, \t, \', \", \\) (default: true)
--no-escape do not process escape sequences
--rope-scaling {none,linear,yarn} RoPE frequency scaling method, defaults to linear unless specified by
the model
(env: LLAMA_ARG_ROPE_SCALING_TYPE)
--rope-scale N RoPE context scaling factor, expands context by a factor of N
(env: LLAMA_ARG_ROPE_SCALE)
--rope-freq-base N RoPE base frequency, used by NTK-aware scaling (default: loaded from
model)
(env: LLAMA_ARG_ROPE_FREQ_BASE)
--rope-freq-scale N RoPE frequency scaling factor, expands context by a factor of 1/N
(env: LLAMA_ARG_ROPE_FREQ_SCALE)
--yarn-orig-ctx N YaRN: original context size of model (default: 0 = model training
context size)
(env: LLAMA_ARG_YARN_ORIG_CTX)
--yarn-ext-factor N YaRN: extrapolation mix factor (default: -1.0, 0.0 = full
interpolation)
(env: LLAMA_ARG_YARN_EXT_FACTOR)
--yarn-attn-factor N YaRN: scale sqrt(t) or attention magnitude (default: -1.0)
(env: LLAMA_ARG_YARN_ATTN_FACTOR)
--yarn-beta-slow N YaRN: high correction dim or alpha (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_SLOW)
--yarn-beta-fast N YaRN: low correction dim or beta (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_FAST)
-nkvo, --no-kv-offload disable KV offload
(env: LLAMA_ARG_NO_KV_OFFLOAD)
-nr, --no-repack disable weight repacking
(env: LLAMA_ARG_NO_REPACK)
--no-host bypass host buffer allowing extra buffers to be used
(env: LLAMA_ARG_NO_HOST)
-ctk, --cache-type-k TYPE KV cache data type for K
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K)
-ctv, --cache-type-v TYPE KV cache data type for V
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V)
-dt, --defrag-thold N KV cache defragmentation threshold (DEPRECATED)
(env: LLAMA_ARG_DEFRAG_THOLD)
-np, --parallel N number of parallel sequences to decode (default: 1)
(env: LLAMA_ARG_N_PARALLEL)
--mlock force system to keep model in RAM rather than swapping or compressing
(env: LLAMA_ARG_MLOCK)
--no-mmap do not memory-map model (slower load but may reduce pageouts if not
using mlock)
(env: LLAMA_ARG_NO_MMAP)
--numa TYPE attempt optimizations that help on some NUMA systems
- distribute: spread execution evenly over all nodes
- isolate: only spawn threads on CPUs on the node that execution
started on
- numactl: use the CPU map provided by numactl
if run without this previously, it is recommended to drop the system
page cache before using this
see https://github.com/ggml-org/llama.cpp/issues/1437
(env: LLAMA_ARG_NUMA)
-dev, --device <dev1,dev2,..> comma-separated list of devices to use for offloading (none = don't
offload)
use --list-devices to see a list of available devices
(env: LLAMA_ARG_DEVICE)
--list-devices print list of available devices and exit
--override-tensor, -ot <tensor name pattern>=<buffer type>,...
override tensor buffer type
--cpu-moe, -cmoe keep all Mixture of Experts (MoE) weights in the CPU
(env: LLAMA_ARG_CPU_MOE)
--n-cpu-moe, -ncmoe N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU
(env: LLAMA_ARG_N_CPU_MOE)
-ngl, --gpu-layers, --n-gpu-layers N max. number of layers to store in VRAM (default: -1)
(env: LLAMA_ARG_N_GPU_LAYERS)
-sm, --split-mode {none,layer,row} how to split the model across multiple GPUs, one of:
- none: use one GPU only
- layer (default): split layers and KV across GPUs
- row: split rows across GPUs
(env: LLAMA_ARG_SPLIT_MODE)
-ts, --tensor-split N0,N1,N2,... fraction of the model to offload to each GPU, comma-separated list of
proportions, e.g. 3,1
(env: LLAMA_ARG_TENSOR_SPLIT)
-mg, --main-gpu INDEX the GPU to use for the model (with split-mode = none), or for
intermediate results and KV (with split-mode = row) (default: 0)
(env: LLAMA_ARG_MAIN_GPU)
--check-tensors check model tensor data for invalid values (default: false)
--override-kv KEY=TYPE:VALUE advanced option to override model metadata by key. may be specified
multiple times.
types: int, float, bool, str. example: --override-kv
tokenizer.ggml.add_bos_token=bool:false
--no-op-offload disable offloading host tensor operations to device (default: false)
--lora FNAME path to LoRA adapter (can be repeated to use multiple adapters)
--lora-scaled FNAME SCALE path to LoRA adapter with user defined scaling (can be repeated to use
multiple adapters)
--control-vector FNAME add a control vector
note: this argument can be repeated to add multiple control vectors
--control-vector-scaled FNAME SCALE add a control vector with user defined scaling SCALE
note: this argument can be repeated to add multiple scaled control
vectors
--control-vector-layer-range START END
layer range to apply the control vector(s) to, start and end inclusive
-m, --model FNAME model path (default: `models/$filename` with filename from `--hf-file`
or `--model-url` if set, otherwise models/7B/ggml-model-f16.gguf)
(env: LLAMA_ARG_MODEL)
-mu, --model-url MODEL_URL model download url (default: unused)
(env: LLAMA_ARG_MODEL_URL)
-dr, --docker-repo [<repo>/]<model>[:quant]
Docker Hub model repository. repo is optional, default to ai/. quant
is optional, default to :latest.
example: gemma3
(default: unused)
(env: LLAMA_ARG_DOCKER_REPO)
-hf, -hfr, --hf-repo <user>/<model>[:quant]
Hugging Face model repository; quant is optional, case-insensitive,
default to Q4_K_M, or falls back to the first file in the repo if
Q4_K_M doesn't exist.
mmproj is also downloaded automatically if available. to disable, add
--no-mmproj
example: unsloth/phi-4-GGUF:q4_k_m
(default: unused)
(env: LLAMA_ARG_HF_REPO)
-hfd, -hfrd, --hf-repo-draft <user>/<model>[:quant]
Same as --hf-repo, but for the draft model (default: unused)
(env: LLAMA_ARG_HFD_REPO)
-hff, --hf-file FILE Hugging Face model file. If specified, it will override the quant in
--hf-repo (default: unused)
(env: LLAMA_ARG_HF_FILE)
-hfv, -hfrv, --hf-repo-v <user>/<model>[:quant]
Hugging Face model repository for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_REPO_V)
-hffv, --hf-file-v FILE Hugging Face model file for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_FILE_V)
-hft, --hf-token TOKEN Hugging Face access token (default: value from HF_TOKEN environment
variable)
(env: HF_TOKEN)
--log-disable Log disable
--log-file FNAME Log to file
--log-colors [on|off|auto] Set colored logging ('on', 'off', or 'auto', default: 'auto')
'auto' enables colors when output is to a terminal
(env: LLAMA_LOG_COLORS)
-v, --verbose, --log-verbose Set verbosity level to infinity (i.e. log all messages, useful for
debugging)
--offline Offline mode: forces use of cache, prevents network access
(env: LLAMA_OFFLINE)
-lv, --verbosity, --log-verbosity N Set the verbosity threshold. Messages with a higher verbosity will be
ignored.
(env: LLAMA_LOG_VERBOSITY)
--log-prefix Enable prefix in log messages
(env: LLAMA_LOG_PREFIX)
--log-timestamps Enable timestamps in log messages
(env: LLAMA_LOG_TIMESTAMPS)
-ctkd, --cache-type-k-draft TYPE KV cache data type for K for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K_DRAFT)
-ctvd, --cache-type-v-draft TYPE KV cache data type for V for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V_DRAFT)

----- sampling params -----

--samplers SAMPLERS samplers that will be used for generation in the order, separated by
';'
(default:
penalties;dry;top_n_sigma;top_k;typ_p;top_p;min_p;xtc;temperature)
-s, --seed SEED RNG seed (default: -1, use random seed for -1)
--sampling-seq, --sampler-seq SEQUENCE
simplified sequence for samplers that will be used (default:
edskypmxt)
--ignore-eos ignore end of stream token and continue generating (implies
--logit-bias EOS-inf)
--temp N temperature (default: 0.8)
--top-k N top-k sampling (default: 40, 0 = disabled)
--top-p N top-p sampling (default: 0.9, 1.0 = disabled)
--min-p N min-p sampling (default: 0.1, 0.0 = disabled)
--top-nsigma N top-n-sigma sampling (default: -1.0, -1.0 = disabled)
--xtc-probability N xtc probability (default: 0.0, 0.0 = disabled)
--xtc-threshold N xtc threshold (default: 0.1, 1.0 = disabled)
--typical N locally typical sampling, parameter p (default: 1.0, 1.0 = disabled)
--repeat-last-n N last n tokens to consider for penalize (default: 64, 0 = disabled, -1
= ctx_size)
--repeat-penalty N penalize repeat sequence of tokens (default: 1.0, 1.0 = disabled)
--presence-penalty N repeat alpha presence penalty (default: 0.0, 0.0 = disabled)
--frequency-penalty N repeat alpha frequency penalty (default: 0.0, 0.0 = disabled)
--dry-multiplier N set DRY sampling multiplier (default: 0.0, 0.0 = disabled)
--dry-base N set DRY sampling base value (default: 1.75)
--dry-allowed-length N set allowed length for DRY sampling (default: 2)
--dry-penalty-last-n N set DRY penalty for the last n tokens (default: -1, 0 = disable, -1 =
context size)
--dry-sequence-breaker STRING add sequence breaker for DRY sampling, clearing out default breakers
('\n', ':', '"', '*') in the process; use "none" to not use any
sequence breakers
--dynatemp-range N dynamic temperature range (default: 0.0, 0.0 = disabled)
--dynatemp-exp N dynamic temperature exponent (default: 1.0)
--mirostat N use Mirostat sampling.
Top K, Nucleus and Locally Typical samplers are ignored if used.
(default: 0, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)
--mirostat-lr N Mirostat learning rate, parameter eta (default: 0.1)
--mirostat-ent N Mirostat target entropy, parameter tau (default: 5.0)
-l, --logit-bias TOKEN_ID(+/-)BIAS modifies the likelihood of token appearing in the completion,
i.e. `--logit-bias 15043+1` to increase likelihood of token ' Hello',
or `--logit-bias 15043-1` to decrease likelihood of token ' Hello'
--grammar GRAMMAR BNF-like grammar to constrain generations (see samples in grammars/
dir) (default: '')
--grammar-file FNAME file to read grammar from
-j, --json-schema SCHEMA JSON schema to constrain generations (https://json-schema.org/), e.g.
`{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead
-jf, --json-schema-file FILE File containing a JSON schema to constrain generations
(https://json-schema.org/), e.g. `{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead

----- example-specific params -----

--ctx-checkpoints, --swa-checkpoints N
max number of context checkpoints to create per slot (default: 8)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/15293)
(env: LLAMA_ARG_CTX_CHECKPOINTS)
--cache-ram, -cram N set the maximum cache size in MiB (default: 8192, -1 - no limit, 0 -
disable)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/16391)
(env: LLAMA_ARG_CACHE_RAM)
--no-context-shift disables context shift on infinite text generation (default: enabled)
(env: LLAMA_ARG_NO_CONTEXT_SHIFT)
--context-shift enables context shift on infinite text generation (default: disabled)
(env: LLAMA_ARG_CONTEXT_SHIFT)
-r, --reverse-prompt PROMPT halt generation at PROMPT, return control in interactive mode
-sp, --special special tokens output enabled (default: false)
--no-warmup skip warming up the model with an empty run
--spm-infill use Suffix/Prefix/Middle pattern for infill (instead of
Prefix/Suffix/Middle) as some models prefer this. (default: disabled)
--pooling {none,mean,cls,last,rank} pooling type for embeddings, use model default if unspecified
(env: LLAMA_ARG_POOLING)
-cb, --cont-batching enable continuous batching (a.k.a dynamic batching) (default: enabled)
(env: LLAMA_ARG_CONT_BATCHING)
-nocb, --no-cont-batching disable continuous batching
(env: LLAMA_ARG_NO_CONT_BATCHING)
--mmproj FILE path to a multimodal projector file. see tools/mtmd/README.md
note: if -hf is used, this argument can be omitted
(env: LLAMA_ARG_MMPROJ)
--mmproj-url URL URL to a multimodal projector file. see tools/mtmd/README.md
(env: LLAMA_ARG_MMPROJ_URL)
--no-mmproj explicitly disable multimodal projector, useful when using -hf
(env: LLAMA_ARG_NO_MMPROJ)
--no-mmproj-offload do not offload multimodal projector to GPU
(env: LLAMA_ARG_NO_MMPROJ_OFFLOAD)
--image-min-tokens N minimum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MIN_TOKENS)
--image-max-tokens N maximum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MAX_TOKENS)
--override-tensor-draft, -otd <tensor name pattern>=<buffer type>,...
override tensor buffer type for draft model
--cpu-moe-draft, -cmoed keep all Mixture of Experts (MoE) weights in the CPU for the draft
model
(env: LLAMA_ARG_CPU_MOE_DRAFT)
--n-cpu-moe-draft, -ncmoed N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU for the draft model
(env: LLAMA_ARG_N_CPU_MOE_DRAFT)
-a, --alias STRING set alias for model name (to be used by REST API)
(env: LLAMA_ARG_ALIAS)
--host HOST ip address to listen, or bind to an UNIX socket if the address ends
with .sock (default: 127.0.0.1)
(env: LLAMA_ARG_HOST)
--port PORT port to listen (default: 8080)
(env: LLAMA_ARG_PORT)
--path PATH path to serve static files from (default: )
(env: LLAMA_ARG_STATIC_PATH)
--api-prefix PREFIX prefix path the server serves from, without the trailing slash
(default: )
(env: LLAMA_ARG_API_PREFIX)
--no-webui Disable the Web UI (default: enabled)
(env: LLAMA_ARG_NO_WEBUI)
--embedding, --embeddings restrict to only support embedding use case; use only with dedicated
embedding models (default: disabled)
(env: LLAMA_ARG_EMBEDDINGS)
--reranking, --rerank enable reranking endpoint on server (default: disabled)
(env: LLAMA_ARG_RERANKING)
--api-key KEY API key to use for authentication (default: none)
(env: LLAMA_API_KEY)
--api-key-file FNAME path to file containing API keys (default: none)
--ssl-key-file FNAME path to file a PEM-encoded SSL private key
(env: LLAMA_ARG_SSL_KEY_FILE)
--ssl-cert-file FNAME path to file a PEM-encoded SSL certificate
(env: LLAMA_ARG_SSL_CERT_FILE)
--chat-template-kwargs STRING sets additional params for the json template parser
(env: LLAMA_CHAT_TEMPLATE_KWARGS)
-to, --timeout N server read/write timeout in seconds (default: 600)
(env: LLAMA_ARG_TIMEOUT)
--threads-http N number of threads used to process HTTP requests (default: -1)
(env: LLAMA_ARG_THREADS_HTTP)
--cache-reuse N min chunk size to attempt reusing from the cache via KV shifting
(default: 0)
[(card)](https://ggml.ai/f0.png)
(env: LLAMA_ARG_CACHE_REUSE)
--metrics enable prometheus compatible metrics endpoint (default: disabled)
(env: LLAMA_ARG_ENDPOINT_METRICS)
--props enable changing global properties via POST /props (default: disabled)
(env: LLAMA_ARG_ENDPOINT_PROPS)
--slots enable slots monitoring endpoint (default: enabled)
(env: LLAMA_ARG_ENDPOINT_SLOTS)
--no-slots disables slots monitoring endpoint
(env: LLAMA_ARG_NO_ENDPOINT_SLOTS)
--slot-save-path PATH path to save slot kv cache (default: disabled)
--jinja use jinja template for chat (default: disabled)
(env: LLAMA_ARG_JINJA)
--reasoning-format FORMAT controls whether thought tags are allowed and/or extracted from the
response, and in which format they're returned; one of:
- none: leaves thoughts unparsed in `message.content`
- deepseek: puts thoughts in `message.reasoning_content`
- deepseek-legacy: keeps `<think>` tags in `message.content` while
also populating `message.reasoning_content`
(default: auto)
(env: LLAMA_ARG_THINK)
--reasoning-budget N controls the amount of thinking allowed; currently only one of: -1 for
unrestricted thinking budget, or 0 to disable thinking (default: -1)
(env: LLAMA_ARG_THINK_BUDGET)
--chat-template JINJA_TEMPLATE set custom jinja chat template (default: template taken from model's
metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE)
--chat-template-file JINJA_TEMPLATE_FILE
set custom jinja chat template file (default: template taken from
model's metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE_FILE)
--no-prefill-assistant whether to prefill the assistant's response if the last message is an
assistant message (default: prefill enabled)
when this flag is set, if the last message is an assistant message
then it will be treated as a full message and not prefilled

(env: LLAMA_ARG_NO_PREFILL_ASSISTANT)
-sps, --slot-prompt-similarity SIMILARITY
how much the prompt of a request must match the prompt of a slot in
order to use that slot (default: 0.10, 0.0 = disabled)
--lora-init-without-apply load LoRA adapters without applying them (apply later via POST
/lora-adapters) (default: disabled)
-td, --threads-draft N number of threads to use during generation (default: same as
--threads)
-tbd, --threads-batch-draft N number of threads to use during batch and prompt processing (default:
same as --threads-draft)
--draft-max, --draft, --draft-n N number of tokens to draft for speculative decoding (default: 16)
(env: LLAMA_ARG_DRAFT_MAX)
--draft-min, --draft-n-min N minimum number of draft tokens to use for speculative decoding
(default: 0)
(env: LLAMA_ARG_DRAFT_MIN)
--draft-p-min P minimum speculative decoding probability (greedy) (default: 0.8)
(env: LLAMA_ARG_DRAFT_P_MIN)
-cd, --ctx-size-draft N size of the prompt context for the draft model (default: 0, 0 = loaded
from model)
(env: LLAMA_ARG_CTX_SIZE_DRAFT)
-devd, --device-draft <dev1,dev2,..> comma-separated list of devices to use for offloading the draft model
(none = don't offload)
use --list-devices to see a list of available devices
-ngld, --gpu-layers-draft, --n-gpu-layers-draft N
number of layers to store in VRAM for the draft model
(env: LLAMA_ARG_N_GPU_LAYERS_DRAFT)
-md, --model-draft FNAME draft model for speculative decoding (default: unused)
(env: LLAMA_ARG_MODEL_DRAFT)
--spec-replace TARGET DRAFT translate the string in TARGET into DRAFT if the draft model and main
model are not compatible
-mv, --model-vocoder FNAME vocoder model for audio generation (default: unused)
--tts-use-guide-tokens Use guide tokens to improve TTS word recall
--embd-gemma-default use default EmbeddingGemma model (note: can download weights from the
internet)
--fim-qwen-1.5b-default use default Qwen 2.5 Coder 1.5B (note: can download weights from the
internet)
--fim-qwen-3b-default use default Qwen 2.5 Coder 3B (note: can download weights from the
internet)
--fim-qwen-7b-default use default Qwen 2.5 Coder 7B (note: can download weights from the
internet)
--fim-qwen-7b-spec use Qwen 2.5 Coder 7B + 0.5B draft for speculative decoding (note: can
download weights from the internet)
--fim-qwen-14b-spec use Qwen 2.5 Coder 14B + 0.5B draft for speculative decoding (note:
can download weights from the internet)
--fim-qwen-30b-default use default Qwen 3 Coder 30B A3B Instruct (note: can download weights
from the internet)
--gpt-oss-20b-default use gpt-oss-20b (note: can download weights from the internet)
--gpt-oss-120b-default use gpt-oss-120b (note: can download weights from the internet)
--vision-gemma-4b-default use Gemma 3 4B QAT (note: can download weights from the internet)
--vision-gemma-12b-default use Gemma 3 12B QAT (note: can download weights from the internet)
</syntaxhighlight>
=== Test ===
=== Bekannte Probleme ===
<syntaxhighlight lang="bash" line="1">
Error 1 [ 34%] Linking HIP shared library ../../../bin/libggml-hip.so [ 34%] Built target ggml-hip gmake[1]: *** [CMakeFiles/Makefile2:1775: ggml/src/CMakeFiles/ggml-cpu.dir/all] Error 2 gmake: *** [Makefile:146: all] Error 2
</syntaxhighlight>
Lösung:
<syntaxhighlight lang="bash" line="1">
GCC 13.3.0 konnte damit nicht umgehen, aber GCC 14.2.0 hat bessere C++-Standard-Compliance und kann diesen Header-Konflikt korrekt auflösen – deshalb funktioniert die Kompilierung jetzt mit der neueren Compiler-Version.
sudo update-alternatives --remove gcc /usr/bin/gcc-13
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-14 100
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-13 50
sudo update-alternatives --config gcc
# Wähle gcc-14

sudo update-alternatives --remove g++ /usr/bin/g++-13
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-14 100
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-13 50
sudo update-alternatives --config g++
# Wähle g++-14
</syntaxhighlight>
=== Nützliche Links ===
* [https://github.com/ggml-org/llama.cpp Offizielles llama.cpp Repository]
* [https://github.com/ggml-org/llama.cpp/blob/master/examples/server/README.md llama-server Dokumentation]
* [https://github.com/ggml-org/llama.cpp/discussions llama.cpp Discussions]
* [https://huggingface.co/models?library=gguf GGUF Modelle auf Hugging Face]
* [https://docs.rocm.com/ ROCm Dokumentation]
* [https://vulkan.lunarg.com/doc/sdk Vulkan SDK Dokumentation]

Llama.cpp

2026-01-07T14:33:58Z

Hendrik: /* ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 */

=== Beschreibung ===
llama.cpp ist eine C/C++ Implementierung für Inference von Large Language Models (LLMs). Es unterstützt verschiedene Backends (CPU, Vulkan, ROCm, CUDA) und ermöglicht das Ausführen von quantisierten Modellen im GGUF-Format.

=== Download ===
<syntaxhighlight lang="bash" line="1">
git clone https://github.com/ggml-org/llama.cpp
cd llama.cpp
</syntaxhighlight>

=== Installation ===
==== Vulkan Build ====
<syntaxhighlight lang="bash" line="1">
sudo dnf install libcurl-devel -y
</syntaxhighlight>
[https://vulkan.lunarg.com/sdk/home Vulkan SDK] herunterladen
entpacken, ins Verzeichnis wechseln und source setup-env.sh ausführen. Dann ins Verzeichnis wechseln wo llama.cpp installiert werden soll.
Dort dann
<syntaxhighlight lang="bash" line="1">
source /home/hendrik/Programme/Vulkan_SDK/1.4.335.0/setup-env.sh
cd /home/hendrik/Programme/llama.cpp/
rm -R build-vulkan
cmake -B build-vulkan -DGGML_VULKAN=1
cmake --build build-vulkan --config Release -- -j $(nproc)
</syntaxhighlight>

==== ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 ====
gfx906 = Mi 50 Support 
gfx1100 = 7900 XTX Support

<syntaxhighlight lang="bash" line="1">
Copy & paste all the commands from the quick install https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/quick-start.html
Before rebooting to complete the install, download the 6.4 rocblas from the AUR: https://archlinux.org/packages/extra/x86_64/rocblas/
Extract it
Copy all tensor files that contain gfx906 in rocblas-6.4.3-3-x86_64.pkg/opt/rocm/lib/rocblas/library to /opt/rocm/lib/rocblas/library
Reboot
Check if it worked by running sudo update-alternatives --display rocm
</syntaxhighlight>
# ROCm Umgebung einrichten (optional falls Fehler auftreten)
<syntaxhighlight lang="bash" line="1">
echo 'export LD_LIBRARY_PATH=/opt/rocm/lib:$LD_LIBRARY_PATH' >> ~/.bashrc
echo 'export HSA_OVERRIDE_GFX_VERSION=9.0.6' >> ~/.bashrc # Für MI50/MI60
source ~/.bashrc
</syntaxhighlight>
# llama.cpp kompilieren
<syntaxhighlight lang="bash" line="1">
cd /home/hendrik/Programme/llama.cpp/ && rm -R build-rocm
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
cmake -B build-rocm -DGGML_HIP=ON -DAMDGPU_TARGETS=gfx906 -DGGML_HIP_ROCWMMA_FATTN=ON \
-DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 8
</syntaxhighlight>

Für mehrere GPU-Architekturen gleichzeitig:
<syntaxhighlight lang="bash" line="1">
# Für MI50 (gfx906) und RX 7900 XTX (gfx1100)
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
cmake -S . -B build -DGGML_HIP=ON -DAMDGPU_TARGETS="gfx906;gfx1100" \
-DCMAKE_BUILD_TYPE=Release && \
cmake --build build --config Release -- -j 8
</syntaxhighlight>

=== Konfiguration ===

==== llama-server als systemd Service einrichten ====

'''Service-Datei erstellen:'''
<syntaxhighlight lang="bash" line="1">
sudo nano /etc/systemd/system/llama-server.service
</syntaxhighlight>

'''Service-Datei Inhalt (Multi-GPU mit ROCm):'''
<syntaxhighlight lang="ini" line="1">
[Unit]
Description=Llama.cpp ROCm Multi-GPU Server
After=network.target

[Service]
Type=simple
User=username
Group=username
WorkingDirectory=/home/username/llama.cpp/build/bin

# Multi-GPU Konfiguration
Environment="HIP_VISIBLE_DEVICES=0,1"
Environment="HSA_OVERRIDE_GFX_VERSION=9.0.6"
Environment="PATH=/opt/rocm/bin:/usr/local/bin:/usr/bin:/bin"
Environment="LD_LIBRARY_PATH=/opt/rocm/lib"

# Server mit optimalen Multi-GPU Einstellungen
ExecStart=/home/username/llama.cpp/build/bin/llama-server \
-m /home/username/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja

Restart=always
RestartSec=10
LimitNOFILE=65535
LimitMEMLOCK=infinity
StandardOutput=journal
StandardError=journal
SyslogIdentifier=llama-server

[Install]
WantedBy=multi-user.target
</syntaxhighlight>

'''Service aktivieren und starten:'''
<syntaxhighlight lang="bash" line="1">
# Service neu laden
sudo systemctl daemon-reload

# Service aktivieren (Auto-Start beim Boot)
sudo systemctl enable llama-server

# Service starten
sudo systemctl start llama-server

# Status prüfen
sudo systemctl status llama-server

# Logs anzeigen
sudo journalctl -u llama-server -f
</syntaxhighlight>

'''Service Management:'''
<syntaxhighlight lang="bash" line="1">
# Service stoppen
sudo systemctl stop llama-server

# Service neustarten
sudo systemctl restart llama-server

# Service deaktivieren
sudo systemctl disable llama-server

==== Manuelle Server-Starts ====
'''Multi-GPU (ROCm) - Optimal:'''
<syntaxhighlight lang="bash" line="1">
HIP_VISIBLE_DEVICES=0,1 ./llama-server \
-m ~/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja
</syntaxhighlight>

=== Update ===
<syntaxhighlight lang="bash" line="1">
cd ~/llama.cpp
git pull
cmake --build build --config Release -- -j $(nproc)

# Service neu starten falls aktiv
sudo systemctl restart llama-server
</syntaxhighlight>
=== Befehlszeilenargumente ===
llama-server
<syntaxhighlight lang="bash" line="1">
-h, --help, --usage print usage and exit
--version show version and build info
-cl, --cache-list show list of models in cache
--completion-bash print source-able bash completion script for llama.cpp
--verbose-prompt print a verbose prompt before generation (default: false)
-t, --threads N number of CPU threads to use during generation (default: -1)
(env: LLAMA_ARG_THREADS)
-tb, --threads-batch N number of threads to use during batch and prompt processing (default:
same as --threads)
-C, --cpu-mask M CPU affinity mask: arbitrarily long hex. Complements cpu-range
(default: "")
-Cr, --cpu-range lo-hi range of CPUs for affinity. Complements --cpu-mask
--cpu-strict <0|1> use strict CPU placement (default: 0)
--prio N set process/thread priority : low(-1), normal(0), medium(1), high(2),
realtime(3) (default: 0)
--poll <0...100> use polling level to wait for work (0 - no polling, default: 50)
-Cb, --cpu-mask-batch M CPU affinity mask: arbitrarily long hex. Complements cpu-range-batch
(default: same as --cpu-mask)
-Crb, --cpu-range-batch lo-hi ranges of CPUs for affinity. Complements --cpu-mask-batch
--cpu-strict-batch <0|1> use strict CPU placement (default: same as --cpu-strict)
--prio-batch N set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime
(default: 0)
--poll-batch <0|1> use polling to wait for work (default: same as --poll)
-c, --ctx-size N size of the prompt context (default: 4096, 0 = loaded from model)
(env: LLAMA_ARG_CTX_SIZE)
-n, --predict, --n-predict N number of tokens to predict (default: -1, -1 = infinity)
(env: LLAMA_ARG_N_PREDICT)
-b, --batch-size N logical maximum batch size (default: 2048)
(env: LLAMA_ARG_BATCH)
-ub, --ubatch-size N physical maximum batch size (default: 512)
(env: LLAMA_ARG_UBATCH)
--keep N number of tokens to keep from the initial prompt (default: 0, -1 =
all)
--swa-full use full-size SWA cache (default: false)
[(more
info)](https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
(env: LLAMA_ARG_SWA_FULL)
--kv-unified, -kvu use single unified KV buffer for the KV cache of all sequences
(default: false)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/14363)
(env: LLAMA_ARG_KV_SPLIT)
-fa, --flash-attn [on|off|auto] set Flash Attention use ('on', 'off', or 'auto', default: 'auto')
(env: LLAMA_ARG_FLASH_ATTN)
--no-perf disable internal libllama performance timings (default: false)
(env: LLAMA_ARG_NO_PERF)
-e, --escape process escapes sequences (\n, \r, \t, \', \", \\) (default: true)
--no-escape do not process escape sequences
--rope-scaling {none,linear,yarn} RoPE frequency scaling method, defaults to linear unless specified by
the model
(env: LLAMA_ARG_ROPE_SCALING_TYPE)
--rope-scale N RoPE context scaling factor, expands context by a factor of N
(env: LLAMA_ARG_ROPE_SCALE)
--rope-freq-base N RoPE base frequency, used by NTK-aware scaling (default: loaded from
model)
(env: LLAMA_ARG_ROPE_FREQ_BASE)
--rope-freq-scale N RoPE frequency scaling factor, expands context by a factor of 1/N
(env: LLAMA_ARG_ROPE_FREQ_SCALE)
--yarn-orig-ctx N YaRN: original context size of model (default: 0 = model training
context size)
(env: LLAMA_ARG_YARN_ORIG_CTX)
--yarn-ext-factor N YaRN: extrapolation mix factor (default: -1.0, 0.0 = full
interpolation)
(env: LLAMA_ARG_YARN_EXT_FACTOR)
--yarn-attn-factor N YaRN: scale sqrt(t) or attention magnitude (default: -1.0)
(env: LLAMA_ARG_YARN_ATTN_FACTOR)
--yarn-beta-slow N YaRN: high correction dim or alpha (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_SLOW)
--yarn-beta-fast N YaRN: low correction dim or beta (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_FAST)
-nkvo, --no-kv-offload disable KV offload
(env: LLAMA_ARG_NO_KV_OFFLOAD)
-nr, --no-repack disable weight repacking
(env: LLAMA_ARG_NO_REPACK)
--no-host bypass host buffer allowing extra buffers to be used
(env: LLAMA_ARG_NO_HOST)
-ctk, --cache-type-k TYPE KV cache data type for K
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K)
-ctv, --cache-type-v TYPE KV cache data type for V
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V)
-dt, --defrag-thold N KV cache defragmentation threshold (DEPRECATED)
(env: LLAMA_ARG_DEFRAG_THOLD)
-np, --parallel N number of parallel sequences to decode (default: 1)
(env: LLAMA_ARG_N_PARALLEL)
--mlock force system to keep model in RAM rather than swapping or compressing
(env: LLAMA_ARG_MLOCK)
--no-mmap do not memory-map model (slower load but may reduce pageouts if not
using mlock)
(env: LLAMA_ARG_NO_MMAP)
--numa TYPE attempt optimizations that help on some NUMA systems
- distribute: spread execution evenly over all nodes
- isolate: only spawn threads on CPUs on the node that execution
started on
- numactl: use the CPU map provided by numactl
if run without this previously, it is recommended to drop the system
page cache before using this
see https://github.com/ggml-org/llama.cpp/issues/1437
(env: LLAMA_ARG_NUMA)
-dev, --device <dev1,dev2,..> comma-separated list of devices to use for offloading (none = don't
offload)
use --list-devices to see a list of available devices
(env: LLAMA_ARG_DEVICE)
--list-devices print list of available devices and exit
--override-tensor, -ot <tensor name pattern>=<buffer type>,...
override tensor buffer type
--cpu-moe, -cmoe keep all Mixture of Experts (MoE) weights in the CPU
(env: LLAMA_ARG_CPU_MOE)
--n-cpu-moe, -ncmoe N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU
(env: LLAMA_ARG_N_CPU_MOE)
-ngl, --gpu-layers, --n-gpu-layers N max. number of layers to store in VRAM (default: -1)
(env: LLAMA_ARG_N_GPU_LAYERS)
-sm, --split-mode {none,layer,row} how to split the model across multiple GPUs, one of:
- none: use one GPU only
- layer (default): split layers and KV across GPUs
- row: split rows across GPUs
(env: LLAMA_ARG_SPLIT_MODE)
-ts, --tensor-split N0,N1,N2,... fraction of the model to offload to each GPU, comma-separated list of
proportions, e.g. 3,1
(env: LLAMA_ARG_TENSOR_SPLIT)
-mg, --main-gpu INDEX the GPU to use for the model (with split-mode = none), or for
intermediate results and KV (with split-mode = row) (default: 0)
(env: LLAMA_ARG_MAIN_GPU)
--check-tensors check model tensor data for invalid values (default: false)
--override-kv KEY=TYPE:VALUE advanced option to override model metadata by key. may be specified
multiple times.
types: int, float, bool, str. example: --override-kv
tokenizer.ggml.add_bos_token=bool:false
--no-op-offload disable offloading host tensor operations to device (default: false)
--lora FNAME path to LoRA adapter (can be repeated to use multiple adapters)
--lora-scaled FNAME SCALE path to LoRA adapter with user defined scaling (can be repeated to use
multiple adapters)
--control-vector FNAME add a control vector
note: this argument can be repeated to add multiple control vectors
--control-vector-scaled FNAME SCALE add a control vector with user defined scaling SCALE
note: this argument can be repeated to add multiple scaled control
vectors
--control-vector-layer-range START END
layer range to apply the control vector(s) to, start and end inclusive
-m, --model FNAME model path (default: `models/$filename` with filename from `--hf-file`
or `--model-url` if set, otherwise models/7B/ggml-model-f16.gguf)
(env: LLAMA_ARG_MODEL)
-mu, --model-url MODEL_URL model download url (default: unused)
(env: LLAMA_ARG_MODEL_URL)
-dr, --docker-repo [<repo>/]<model>[:quant]
Docker Hub model repository. repo is optional, default to ai/. quant
is optional, default to :latest.
example: gemma3
(default: unused)
(env: LLAMA_ARG_DOCKER_REPO)
-hf, -hfr, --hf-repo <user>/<model>[:quant]
Hugging Face model repository; quant is optional, case-insensitive,
default to Q4_K_M, or falls back to the first file in the repo if
Q4_K_M doesn't exist.
mmproj is also downloaded automatically if available. to disable, add
--no-mmproj
example: unsloth/phi-4-GGUF:q4_k_m
(default: unused)
(env: LLAMA_ARG_HF_REPO)
-hfd, -hfrd, --hf-repo-draft <user>/<model>[:quant]
Same as --hf-repo, but for the draft model (default: unused)
(env: LLAMA_ARG_HFD_REPO)
-hff, --hf-file FILE Hugging Face model file. If specified, it will override the quant in
--hf-repo (default: unused)
(env: LLAMA_ARG_HF_FILE)
-hfv, -hfrv, --hf-repo-v <user>/<model>[:quant]
Hugging Face model repository for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_REPO_V)
-hffv, --hf-file-v FILE Hugging Face model file for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_FILE_V)
-hft, --hf-token TOKEN Hugging Face access token (default: value from HF_TOKEN environment
variable)
(env: HF_TOKEN)
--log-disable Log disable
--log-file FNAME Log to file
--log-colors [on|off|auto] Set colored logging ('on', 'off', or 'auto', default: 'auto')
'auto' enables colors when output is to a terminal
(env: LLAMA_LOG_COLORS)
-v, --verbose, --log-verbose Set verbosity level to infinity (i.e. log all messages, useful for
debugging)
--offline Offline mode: forces use of cache, prevents network access
(env: LLAMA_OFFLINE)
-lv, --verbosity, --log-verbosity N Set the verbosity threshold. Messages with a higher verbosity will be
ignored.
(env: LLAMA_LOG_VERBOSITY)
--log-prefix Enable prefix in log messages
(env: LLAMA_LOG_PREFIX)
--log-timestamps Enable timestamps in log messages
(env: LLAMA_LOG_TIMESTAMPS)
-ctkd, --cache-type-k-draft TYPE KV cache data type for K for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K_DRAFT)
-ctvd, --cache-type-v-draft TYPE KV cache data type for V for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V_DRAFT)

----- sampling params -----

--samplers SAMPLERS samplers that will be used for generation in the order, separated by
';'
(default:
penalties;dry;top_n_sigma;top_k;typ_p;top_p;min_p;xtc;temperature)
-s, --seed SEED RNG seed (default: -1, use random seed for -1)
--sampling-seq, --sampler-seq SEQUENCE
simplified sequence for samplers that will be used (default:
edskypmxt)
--ignore-eos ignore end of stream token and continue generating (implies
--logit-bias EOS-inf)
--temp N temperature (default: 0.8)
--top-k N top-k sampling (default: 40, 0 = disabled)
--top-p N top-p sampling (default: 0.9, 1.0 = disabled)
--min-p N min-p sampling (default: 0.1, 0.0 = disabled)
--top-nsigma N top-n-sigma sampling (default: -1.0, -1.0 = disabled)
--xtc-probability N xtc probability (default: 0.0, 0.0 = disabled)
--xtc-threshold N xtc threshold (default: 0.1, 1.0 = disabled)
--typical N locally typical sampling, parameter p (default: 1.0, 1.0 = disabled)
--repeat-last-n N last n tokens to consider for penalize (default: 64, 0 = disabled, -1
= ctx_size)
--repeat-penalty N penalize repeat sequence of tokens (default: 1.0, 1.0 = disabled)
--presence-penalty N repeat alpha presence penalty (default: 0.0, 0.0 = disabled)
--frequency-penalty N repeat alpha frequency penalty (default: 0.0, 0.0 = disabled)
--dry-multiplier N set DRY sampling multiplier (default: 0.0, 0.0 = disabled)
--dry-base N set DRY sampling base value (default: 1.75)
--dry-allowed-length N set allowed length for DRY sampling (default: 2)
--dry-penalty-last-n N set DRY penalty for the last n tokens (default: -1, 0 = disable, -1 =
context size)
--dry-sequence-breaker STRING add sequence breaker for DRY sampling, clearing out default breakers
('\n', ':', '"', '*') in the process; use "none" to not use any
sequence breakers
--dynatemp-range N dynamic temperature range (default: 0.0, 0.0 = disabled)
--dynatemp-exp N dynamic temperature exponent (default: 1.0)
--mirostat N use Mirostat sampling.
Top K, Nucleus and Locally Typical samplers are ignored if used.
(default: 0, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)
--mirostat-lr N Mirostat learning rate, parameter eta (default: 0.1)
--mirostat-ent N Mirostat target entropy, parameter tau (default: 5.0)
-l, --logit-bias TOKEN_ID(+/-)BIAS modifies the likelihood of token appearing in the completion,
i.e. `--logit-bias 15043+1` to increase likelihood of token ' Hello',
or `--logit-bias 15043-1` to decrease likelihood of token ' Hello'
--grammar GRAMMAR BNF-like grammar to constrain generations (see samples in grammars/
dir) (default: '')
--grammar-file FNAME file to read grammar from
-j, --json-schema SCHEMA JSON schema to constrain generations (https://json-schema.org/), e.g.
`{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead
-jf, --json-schema-file FILE File containing a JSON schema to constrain generations
(https://json-schema.org/), e.g. `{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead

----- example-specific params -----

--ctx-checkpoints, --swa-checkpoints N
max number of context checkpoints to create per slot (default: 8)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/15293)
(env: LLAMA_ARG_CTX_CHECKPOINTS)
--cache-ram, -cram N set the maximum cache size in MiB (default: 8192, -1 - no limit, 0 -
disable)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/16391)
(env: LLAMA_ARG_CACHE_RAM)
--no-context-shift disables context shift on infinite text generation (default: enabled)
(env: LLAMA_ARG_NO_CONTEXT_SHIFT)
--context-shift enables context shift on infinite text generation (default: disabled)
(env: LLAMA_ARG_CONTEXT_SHIFT)
-r, --reverse-prompt PROMPT halt generation at PROMPT, return control in interactive mode
-sp, --special special tokens output enabled (default: false)
--no-warmup skip warming up the model with an empty run
--spm-infill use Suffix/Prefix/Middle pattern for infill (instead of
Prefix/Suffix/Middle) as some models prefer this. (default: disabled)
--pooling {none,mean,cls,last,rank} pooling type for embeddings, use model default if unspecified
(env: LLAMA_ARG_POOLING)
-cb, --cont-batching enable continuous batching (a.k.a dynamic batching) (default: enabled)
(env: LLAMA_ARG_CONT_BATCHING)
-nocb, --no-cont-batching disable continuous batching
(env: LLAMA_ARG_NO_CONT_BATCHING)
--mmproj FILE path to a multimodal projector file. see tools/mtmd/README.md
note: if -hf is used, this argument can be omitted
(env: LLAMA_ARG_MMPROJ)
--mmproj-url URL URL to a multimodal projector file. see tools/mtmd/README.md
(env: LLAMA_ARG_MMPROJ_URL)
--no-mmproj explicitly disable multimodal projector, useful when using -hf
(env: LLAMA_ARG_NO_MMPROJ)
--no-mmproj-offload do not offload multimodal projector to GPU
(env: LLAMA_ARG_NO_MMPROJ_OFFLOAD)
--image-min-tokens N minimum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MIN_TOKENS)
--image-max-tokens N maximum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MAX_TOKENS)
--override-tensor-draft, -otd <tensor name pattern>=<buffer type>,...
override tensor buffer type for draft model
--cpu-moe-draft, -cmoed keep all Mixture of Experts (MoE) weights in the CPU for the draft
model
(env: LLAMA_ARG_CPU_MOE_DRAFT)
--n-cpu-moe-draft, -ncmoed N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU for the draft model
(env: LLAMA_ARG_N_CPU_MOE_DRAFT)
-a, --alias STRING set alias for model name (to be used by REST API)
(env: LLAMA_ARG_ALIAS)
--host HOST ip address to listen, or bind to an UNIX socket if the address ends
with .sock (default: 127.0.0.1)
(env: LLAMA_ARG_HOST)
--port PORT port to listen (default: 8080)
(env: LLAMA_ARG_PORT)
--path PATH path to serve static files from (default: )
(env: LLAMA_ARG_STATIC_PATH)
--api-prefix PREFIX prefix path the server serves from, without the trailing slash
(default: )
(env: LLAMA_ARG_API_PREFIX)
--no-webui Disable the Web UI (default: enabled)
(env: LLAMA_ARG_NO_WEBUI)
--embedding, --embeddings restrict to only support embedding use case; use only with dedicated
embedding models (default: disabled)
(env: LLAMA_ARG_EMBEDDINGS)
--reranking, --rerank enable reranking endpoint on server (default: disabled)
(env: LLAMA_ARG_RERANKING)
--api-key KEY API key to use for authentication (default: none)
(env: LLAMA_API_KEY)
--api-key-file FNAME path to file containing API keys (default: none)
--ssl-key-file FNAME path to file a PEM-encoded SSL private key
(env: LLAMA_ARG_SSL_KEY_FILE)
--ssl-cert-file FNAME path to file a PEM-encoded SSL certificate
(env: LLAMA_ARG_SSL_CERT_FILE)
--chat-template-kwargs STRING sets additional params for the json template parser
(env: LLAMA_CHAT_TEMPLATE_KWARGS)
-to, --timeout N server read/write timeout in seconds (default: 600)
(env: LLAMA_ARG_TIMEOUT)
--threads-http N number of threads used to process HTTP requests (default: -1)
(env: LLAMA_ARG_THREADS_HTTP)
--cache-reuse N min chunk size to attempt reusing from the cache via KV shifting
(default: 0)
[(card)](https://ggml.ai/f0.png)
(env: LLAMA_ARG_CACHE_REUSE)
--metrics enable prometheus compatible metrics endpoint (default: disabled)
(env: LLAMA_ARG_ENDPOINT_METRICS)
--props enable changing global properties via POST /props (default: disabled)
(env: LLAMA_ARG_ENDPOINT_PROPS)
--slots enable slots monitoring endpoint (default: enabled)
(env: LLAMA_ARG_ENDPOINT_SLOTS)
--no-slots disables slots monitoring endpoint
(env: LLAMA_ARG_NO_ENDPOINT_SLOTS)
--slot-save-path PATH path to save slot kv cache (default: disabled)
--jinja use jinja template for chat (default: disabled)
(env: LLAMA_ARG_JINJA)
--reasoning-format FORMAT controls whether thought tags are allowed and/or extracted from the
response, and in which format they're returned; one of:
- none: leaves thoughts unparsed in `message.content`
- deepseek: puts thoughts in `message.reasoning_content`
- deepseek-legacy: keeps `<think>` tags in `message.content` while
also populating `message.reasoning_content`
(default: auto)
(env: LLAMA_ARG_THINK)
--reasoning-budget N controls the amount of thinking allowed; currently only one of: -1 for
unrestricted thinking budget, or 0 to disable thinking (default: -1)
(env: LLAMA_ARG_THINK_BUDGET)
--chat-template JINJA_TEMPLATE set custom jinja chat template (default: template taken from model's
metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE)
--chat-template-file JINJA_TEMPLATE_FILE
set custom jinja chat template file (default: template taken from
model's metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE_FILE)
--no-prefill-assistant whether to prefill the assistant's response if the last message is an
assistant message (default: prefill enabled)
when this flag is set, if the last message is an assistant message
then it will be treated as a full message and not prefilled

(env: LLAMA_ARG_NO_PREFILL_ASSISTANT)
-sps, --slot-prompt-similarity SIMILARITY
how much the prompt of a request must match the prompt of a slot in
order to use that slot (default: 0.10, 0.0 = disabled)
--lora-init-without-apply load LoRA adapters without applying them (apply later via POST
/lora-adapters) (default: disabled)
-td, --threads-draft N number of threads to use during generation (default: same as
--threads)
-tbd, --threads-batch-draft N number of threads to use during batch and prompt processing (default:
same as --threads-draft)
--draft-max, --draft, --draft-n N number of tokens to draft for speculative decoding (default: 16)
(env: LLAMA_ARG_DRAFT_MAX)
--draft-min, --draft-n-min N minimum number of draft tokens to use for speculative decoding
(default: 0)
(env: LLAMA_ARG_DRAFT_MIN)
--draft-p-min P minimum speculative decoding probability (greedy) (default: 0.8)
(env: LLAMA_ARG_DRAFT_P_MIN)
-cd, --ctx-size-draft N size of the prompt context for the draft model (default: 0, 0 = loaded
from model)
(env: LLAMA_ARG_CTX_SIZE_DRAFT)
-devd, --device-draft <dev1,dev2,..> comma-separated list of devices to use for offloading the draft model
(none = don't offload)
use --list-devices to see a list of available devices
-ngld, --gpu-layers-draft, --n-gpu-layers-draft N
number of layers to store in VRAM for the draft model
(env: LLAMA_ARG_N_GPU_LAYERS_DRAFT)
-md, --model-draft FNAME draft model for speculative decoding (default: unused)
(env: LLAMA_ARG_MODEL_DRAFT)
--spec-replace TARGET DRAFT translate the string in TARGET into DRAFT if the draft model and main
model are not compatible
-mv, --model-vocoder FNAME vocoder model for audio generation (default: unused)
--tts-use-guide-tokens Use guide tokens to improve TTS word recall
--embd-gemma-default use default EmbeddingGemma model (note: can download weights from the
internet)
--fim-qwen-1.5b-default use default Qwen 2.5 Coder 1.5B (note: can download weights from the
internet)
--fim-qwen-3b-default use default Qwen 2.5 Coder 3B (note: can download weights from the
internet)
--fim-qwen-7b-default use default Qwen 2.5 Coder 7B (note: can download weights from the
internet)
--fim-qwen-7b-spec use Qwen 2.5 Coder 7B + 0.5B draft for speculative decoding (note: can
download weights from the internet)
--fim-qwen-14b-spec use Qwen 2.5 Coder 14B + 0.5B draft for speculative decoding (note:
can download weights from the internet)
--fim-qwen-30b-default use default Qwen 3 Coder 30B A3B Instruct (note: can download weights
from the internet)
--gpt-oss-20b-default use gpt-oss-20b (note: can download weights from the internet)
--gpt-oss-120b-default use gpt-oss-120b (note: can download weights from the internet)
--vision-gemma-4b-default use Gemma 3 4B QAT (note: can download weights from the internet)
--vision-gemma-12b-default use Gemma 3 12B QAT (note: can download weights from the internet)
</syntaxhighlight>
=== Test ===
=== Bekannte Probleme ===
<syntaxhighlight lang="bash" line="1">
Error 1 [ 34%] Linking HIP shared library ../../../bin/libggml-hip.so [ 34%] Built target ggml-hip gmake[1]: *** [CMakeFiles/Makefile2:1775: ggml/src/CMakeFiles/ggml-cpu.dir/all] Error 2 gmake: *** [Makefile:146: all] Error 2
</syntaxhighlight>
Lösung:
<syntaxhighlight lang="bash" line="1">
GCC 13.3.0 konnte damit nicht umgehen, aber GCC 14.2.0 hat bessere C++-Standard-Compliance und kann diesen Header-Konflikt korrekt auflösen – deshalb funktioniert die Kompilierung jetzt mit der neueren Compiler-Version.
sudo update-alternatives --remove gcc /usr/bin/gcc-13
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-14 100
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-13 50
sudo update-alternatives --config gcc
# Wähle gcc-14

sudo update-alternatives --remove g++ /usr/bin/g++-13
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-14 100
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-13 50
sudo update-alternatives --config g++
# Wähle g++-14
</syntaxhighlight>
=== Nützliche Links ===
* [https://github.com/ggml-org/llama.cpp Offizielles llama.cpp Repository]
* [https://github.com/ggml-org/llama.cpp/blob/master/examples/server/README.md llama-server Dokumentation]
* [https://github.com/ggml-org/llama.cpp/discussions llama.cpp Discussions]
* [https://huggingface.co/models?library=gguf GGUF Modelle auf Hugging Face]
* [https://docs.rocm.com/ ROCm Dokumentation]
* [https://vulkan.lunarg.com/doc/sdk Vulkan SDK Dokumentation]

Llama.cpp

2026-01-07T14:33:39Z

Hendrik: /* ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 */

=== Beschreibung ===
llama.cpp ist eine C/C++ Implementierung für Inference von Large Language Models (LLMs). Es unterstützt verschiedene Backends (CPU, Vulkan, ROCm, CUDA) und ermöglicht das Ausführen von quantisierten Modellen im GGUF-Format.

=== Download ===
<syntaxhighlight lang="bash" line="1">
git clone https://github.com/ggml-org/llama.cpp
cd llama.cpp
</syntaxhighlight>

=== Installation ===
==== Vulkan Build ====
<syntaxhighlight lang="bash" line="1">
sudo dnf install libcurl-devel -y
</syntaxhighlight>
[https://vulkan.lunarg.com/sdk/home Vulkan SDK] herunterladen
entpacken, ins Verzeichnis wechseln und source setup-env.sh ausführen. Dann ins Verzeichnis wechseln wo llama.cpp installiert werden soll.
Dort dann
<syntaxhighlight lang="bash" line="1">
source /home/hendrik/Programme/Vulkan_SDK/1.4.335.0/setup-env.sh
cd /home/hendrik/Programme/llama.cpp/
rm -R build-vulkan
cmake -B build-vulkan -DGGML_VULKAN=1
cmake --build build-vulkan --config Release -- -j $(nproc)
</syntaxhighlight>

==== ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 ====
gfx906 = Mi 50 Support 
gfx1100 = 7900 XTX Support

<syntaxhighlight lang="bash" line="1">
Copy & paste all the commands from the quick install https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/quick-start.html
Before rebooting to complete the install, download the 6.4 rocblas from the AUR: https://archlinux.org/packages/extra/x86_64/rocblas/
Extract it
Copy all tensor files that contain gfx906 in rocblas-6.4.3-3-x86_64.pkg/opt/rocm/lib/rocblas/library to /opt/rocm/lib/rocblas/library
Reboot
Check if it worked by running sudo update-alternatives --display rocm
</syntaxhighlight>
# ROCm Umgebung einrichten (optional falls Fehler auftreten)
<syntaxhighlight lang="bash" line="1">
echo 'export LD_LIBRARY_PATH=/opt/rocm/lib:$LD_LIBRARY_PATH' >> ~/.bashrc
echo 'export HSA_OVERRIDE_GFX_VERSION=9.0.6' >> ~/.bashrc # Für MI50/MI60
source ~/.bashrc
</syntaxhighlight>
# llama.cpp kompilieren
<syntaxhighlight lang="bash" line="1">
rm -R build-rocm
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
cmake -B build-rocm -DGGML_HIP=ON -DAMDGPU_TARGETS=gfx906 -DGGML_HIP_ROCWMMA_FATTN=ON \
-DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 8
</syntaxhighlight>

Für mehrere GPU-Architekturen gleichzeitig:
<syntaxhighlight lang="bash" line="1">
# Für MI50 (gfx906) und RX 7900 XTX (gfx1100)
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
cmake -S . -B build -DGGML_HIP=ON -DAMDGPU_TARGETS="gfx906;gfx1100" \
-DCMAKE_BUILD_TYPE=Release && \
cmake --build build --config Release -- -j 8
</syntaxhighlight>

=== Konfiguration ===

==== llama-server als systemd Service einrichten ====

'''Service-Datei erstellen:'''
<syntaxhighlight lang="bash" line="1">
sudo nano /etc/systemd/system/llama-server.service
</syntaxhighlight>

'''Service-Datei Inhalt (Multi-GPU mit ROCm):'''
<syntaxhighlight lang="ini" line="1">
[Unit]
Description=Llama.cpp ROCm Multi-GPU Server
After=network.target

[Service]
Type=simple
User=username
Group=username
WorkingDirectory=/home/username/llama.cpp/build/bin

# Multi-GPU Konfiguration
Environment="HIP_VISIBLE_DEVICES=0,1"
Environment="HSA_OVERRIDE_GFX_VERSION=9.0.6"
Environment="PATH=/opt/rocm/bin:/usr/local/bin:/usr/bin:/bin"
Environment="LD_LIBRARY_PATH=/opt/rocm/lib"

# Server mit optimalen Multi-GPU Einstellungen
ExecStart=/home/username/llama.cpp/build/bin/llama-server \
-m /home/username/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja

Restart=always
RestartSec=10
LimitNOFILE=65535
LimitMEMLOCK=infinity
StandardOutput=journal
StandardError=journal
SyslogIdentifier=llama-server

[Install]
WantedBy=multi-user.target
</syntaxhighlight>

'''Service aktivieren und starten:'''
<syntaxhighlight lang="bash" line="1">
# Service neu laden
sudo systemctl daemon-reload

# Service aktivieren (Auto-Start beim Boot)
sudo systemctl enable llama-server

# Service starten
sudo systemctl start llama-server

# Status prüfen
sudo systemctl status llama-server

# Logs anzeigen
sudo journalctl -u llama-server -f
</syntaxhighlight>

'''Service Management:'''
<syntaxhighlight lang="bash" line="1">
# Service stoppen
sudo systemctl stop llama-server

# Service neustarten
sudo systemctl restart llama-server

# Service deaktivieren
sudo systemctl disable llama-server

==== Manuelle Server-Starts ====
'''Multi-GPU (ROCm) - Optimal:'''
<syntaxhighlight lang="bash" line="1">
HIP_VISIBLE_DEVICES=0,1 ./llama-server \
-m ~/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja
</syntaxhighlight>

=== Update ===
<syntaxhighlight lang="bash" line="1">
cd ~/llama.cpp
git pull
cmake --build build --config Release -- -j $(nproc)

# Service neu starten falls aktiv
sudo systemctl restart llama-server
</syntaxhighlight>
=== Befehlszeilenargumente ===
llama-server
<syntaxhighlight lang="bash" line="1">
-h, --help, --usage print usage and exit
--version show version and build info
-cl, --cache-list show list of models in cache
--completion-bash print source-able bash completion script for llama.cpp
--verbose-prompt print a verbose prompt before generation (default: false)
-t, --threads N number of CPU threads to use during generation (default: -1)
(env: LLAMA_ARG_THREADS)
-tb, --threads-batch N number of threads to use during batch and prompt processing (default:
same as --threads)
-C, --cpu-mask M CPU affinity mask: arbitrarily long hex. Complements cpu-range
(default: "")
-Cr, --cpu-range lo-hi range of CPUs for affinity. Complements --cpu-mask
--cpu-strict <0|1> use strict CPU placement (default: 0)
--prio N set process/thread priority : low(-1), normal(0), medium(1), high(2),
realtime(3) (default: 0)
--poll <0...100> use polling level to wait for work (0 - no polling, default: 50)
-Cb, --cpu-mask-batch M CPU affinity mask: arbitrarily long hex. Complements cpu-range-batch
(default: same as --cpu-mask)
-Crb, --cpu-range-batch lo-hi ranges of CPUs for affinity. Complements --cpu-mask-batch
--cpu-strict-batch <0|1> use strict CPU placement (default: same as --cpu-strict)
--prio-batch N set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime
(default: 0)
--poll-batch <0|1> use polling to wait for work (default: same as --poll)
-c, --ctx-size N size of the prompt context (default: 4096, 0 = loaded from model)
(env: LLAMA_ARG_CTX_SIZE)
-n, --predict, --n-predict N number of tokens to predict (default: -1, -1 = infinity)
(env: LLAMA_ARG_N_PREDICT)
-b, --batch-size N logical maximum batch size (default: 2048)
(env: LLAMA_ARG_BATCH)
-ub, --ubatch-size N physical maximum batch size (default: 512)
(env: LLAMA_ARG_UBATCH)
--keep N number of tokens to keep from the initial prompt (default: 0, -1 =
all)
--swa-full use full-size SWA cache (default: false)
[(more
info)](https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
(env: LLAMA_ARG_SWA_FULL)
--kv-unified, -kvu use single unified KV buffer for the KV cache of all sequences
(default: false)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/14363)
(env: LLAMA_ARG_KV_SPLIT)
-fa, --flash-attn [on|off|auto] set Flash Attention use ('on', 'off', or 'auto', default: 'auto')
(env: LLAMA_ARG_FLASH_ATTN)
--no-perf disable internal libllama performance timings (default: false)
(env: LLAMA_ARG_NO_PERF)
-e, --escape process escapes sequences (\n, \r, \t, \', \", \\) (default: true)
--no-escape do not process escape sequences
--rope-scaling {none,linear,yarn} RoPE frequency scaling method, defaults to linear unless specified by
the model
(env: LLAMA_ARG_ROPE_SCALING_TYPE)
--rope-scale N RoPE context scaling factor, expands context by a factor of N
(env: LLAMA_ARG_ROPE_SCALE)
--rope-freq-base N RoPE base frequency, used by NTK-aware scaling (default: loaded from
model)
(env: LLAMA_ARG_ROPE_FREQ_BASE)
--rope-freq-scale N RoPE frequency scaling factor, expands context by a factor of 1/N
(env: LLAMA_ARG_ROPE_FREQ_SCALE)
--yarn-orig-ctx N YaRN: original context size of model (default: 0 = model training
context size)
(env: LLAMA_ARG_YARN_ORIG_CTX)
--yarn-ext-factor N YaRN: extrapolation mix factor (default: -1.0, 0.0 = full
interpolation)
(env: LLAMA_ARG_YARN_EXT_FACTOR)
--yarn-attn-factor N YaRN: scale sqrt(t) or attention magnitude (default: -1.0)
(env: LLAMA_ARG_YARN_ATTN_FACTOR)
--yarn-beta-slow N YaRN: high correction dim or alpha (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_SLOW)
--yarn-beta-fast N YaRN: low correction dim or beta (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_FAST)
-nkvo, --no-kv-offload disable KV offload
(env: LLAMA_ARG_NO_KV_OFFLOAD)
-nr, --no-repack disable weight repacking
(env: LLAMA_ARG_NO_REPACK)
--no-host bypass host buffer allowing extra buffers to be used
(env: LLAMA_ARG_NO_HOST)
-ctk, --cache-type-k TYPE KV cache data type for K
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K)
-ctv, --cache-type-v TYPE KV cache data type for V
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V)
-dt, --defrag-thold N KV cache defragmentation threshold (DEPRECATED)
(env: LLAMA_ARG_DEFRAG_THOLD)
-np, --parallel N number of parallel sequences to decode (default: 1)
(env: LLAMA_ARG_N_PARALLEL)
--mlock force system to keep model in RAM rather than swapping or compressing
(env: LLAMA_ARG_MLOCK)
--no-mmap do not memory-map model (slower load but may reduce pageouts if not
using mlock)
(env: LLAMA_ARG_NO_MMAP)
--numa TYPE attempt optimizations that help on some NUMA systems
- distribute: spread execution evenly over all nodes
- isolate: only spawn threads on CPUs on the node that execution
started on
- numactl: use the CPU map provided by numactl
if run without this previously, it is recommended to drop the system
page cache before using this
see https://github.com/ggml-org/llama.cpp/issues/1437
(env: LLAMA_ARG_NUMA)
-dev, --device <dev1,dev2,..> comma-separated list of devices to use for offloading (none = don't
offload)
use --list-devices to see a list of available devices
(env: LLAMA_ARG_DEVICE)
--list-devices print list of available devices and exit
--override-tensor, -ot <tensor name pattern>=<buffer type>,...
override tensor buffer type
--cpu-moe, -cmoe keep all Mixture of Experts (MoE) weights in the CPU
(env: LLAMA_ARG_CPU_MOE)
--n-cpu-moe, -ncmoe N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU
(env: LLAMA_ARG_N_CPU_MOE)
-ngl, --gpu-layers, --n-gpu-layers N max. number of layers to store in VRAM (default: -1)
(env: LLAMA_ARG_N_GPU_LAYERS)
-sm, --split-mode {none,layer,row} how to split the model across multiple GPUs, one of:
- none: use one GPU only
- layer (default): split layers and KV across GPUs
- row: split rows across GPUs
(env: LLAMA_ARG_SPLIT_MODE)
-ts, --tensor-split N0,N1,N2,... fraction of the model to offload to each GPU, comma-separated list of
proportions, e.g. 3,1
(env: LLAMA_ARG_TENSOR_SPLIT)
-mg, --main-gpu INDEX the GPU to use for the model (with split-mode = none), or for
intermediate results and KV (with split-mode = row) (default: 0)
(env: LLAMA_ARG_MAIN_GPU)
--check-tensors check model tensor data for invalid values (default: false)
--override-kv KEY=TYPE:VALUE advanced option to override model metadata by key. may be specified
multiple times.
types: int, float, bool, str. example: --override-kv
tokenizer.ggml.add_bos_token=bool:false
--no-op-offload disable offloading host tensor operations to device (default: false)
--lora FNAME path to LoRA adapter (can be repeated to use multiple adapters)
--lora-scaled FNAME SCALE path to LoRA adapter with user defined scaling (can be repeated to use
multiple adapters)
--control-vector FNAME add a control vector
note: this argument can be repeated to add multiple control vectors
--control-vector-scaled FNAME SCALE add a control vector with user defined scaling SCALE
note: this argument can be repeated to add multiple scaled control
vectors
--control-vector-layer-range START END
layer range to apply the control vector(s) to, start and end inclusive
-m, --model FNAME model path (default: `models/$filename` with filename from `--hf-file`
or `--model-url` if set, otherwise models/7B/ggml-model-f16.gguf)
(env: LLAMA_ARG_MODEL)
-mu, --model-url MODEL_URL model download url (default: unused)
(env: LLAMA_ARG_MODEL_URL)
-dr, --docker-repo [<repo>/]<model>[:quant]
Docker Hub model repository. repo is optional, default to ai/. quant
is optional, default to :latest.
example: gemma3
(default: unused)
(env: LLAMA_ARG_DOCKER_REPO)
-hf, -hfr, --hf-repo <user>/<model>[:quant]
Hugging Face model repository; quant is optional, case-insensitive,
default to Q4_K_M, or falls back to the first file in the repo if
Q4_K_M doesn't exist.
mmproj is also downloaded automatically if available. to disable, add
--no-mmproj
example: unsloth/phi-4-GGUF:q4_k_m
(default: unused)
(env: LLAMA_ARG_HF_REPO)
-hfd, -hfrd, --hf-repo-draft <user>/<model>[:quant]
Same as --hf-repo, but for the draft model (default: unused)
(env: LLAMA_ARG_HFD_REPO)
-hff, --hf-file FILE Hugging Face model file. If specified, it will override the quant in
--hf-repo (default: unused)
(env: LLAMA_ARG_HF_FILE)
-hfv, -hfrv, --hf-repo-v <user>/<model>[:quant]
Hugging Face model repository for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_REPO_V)
-hffv, --hf-file-v FILE Hugging Face model file for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_FILE_V)
-hft, --hf-token TOKEN Hugging Face access token (default: value from HF_TOKEN environment
variable)
(env: HF_TOKEN)
--log-disable Log disable
--log-file FNAME Log to file
--log-colors [on|off|auto] Set colored logging ('on', 'off', or 'auto', default: 'auto')
'auto' enables colors when output is to a terminal
(env: LLAMA_LOG_COLORS)
-v, --verbose, --log-verbose Set verbosity level to infinity (i.e. log all messages, useful for
debugging)
--offline Offline mode: forces use of cache, prevents network access
(env: LLAMA_OFFLINE)
-lv, --verbosity, --log-verbosity N Set the verbosity threshold. Messages with a higher verbosity will be
ignored.
(env: LLAMA_LOG_VERBOSITY)
--log-prefix Enable prefix in log messages
(env: LLAMA_LOG_PREFIX)
--log-timestamps Enable timestamps in log messages
(env: LLAMA_LOG_TIMESTAMPS)
-ctkd, --cache-type-k-draft TYPE KV cache data type for K for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K_DRAFT)
-ctvd, --cache-type-v-draft TYPE KV cache data type for V for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V_DRAFT)

----- sampling params -----

--samplers SAMPLERS samplers that will be used for generation in the order, separated by
';'
(default:
penalties;dry;top_n_sigma;top_k;typ_p;top_p;min_p;xtc;temperature)
-s, --seed SEED RNG seed (default: -1, use random seed for -1)
--sampling-seq, --sampler-seq SEQUENCE
simplified sequence for samplers that will be used (default:
edskypmxt)
--ignore-eos ignore end of stream token and continue generating (implies
--logit-bias EOS-inf)
--temp N temperature (default: 0.8)
--top-k N top-k sampling (default: 40, 0 = disabled)
--top-p N top-p sampling (default: 0.9, 1.0 = disabled)
--min-p N min-p sampling (default: 0.1, 0.0 = disabled)
--top-nsigma N top-n-sigma sampling (default: -1.0, -1.0 = disabled)
--xtc-probability N xtc probability (default: 0.0, 0.0 = disabled)
--xtc-threshold N xtc threshold (default: 0.1, 1.0 = disabled)
--typical N locally typical sampling, parameter p (default: 1.0, 1.0 = disabled)
--repeat-last-n N last n tokens to consider for penalize (default: 64, 0 = disabled, -1
= ctx_size)
--repeat-penalty N penalize repeat sequence of tokens (default: 1.0, 1.0 = disabled)
--presence-penalty N repeat alpha presence penalty (default: 0.0, 0.0 = disabled)
--frequency-penalty N repeat alpha frequency penalty (default: 0.0, 0.0 = disabled)
--dry-multiplier N set DRY sampling multiplier (default: 0.0, 0.0 = disabled)
--dry-base N set DRY sampling base value (default: 1.75)
--dry-allowed-length N set allowed length for DRY sampling (default: 2)
--dry-penalty-last-n N set DRY penalty for the last n tokens (default: -1, 0 = disable, -1 =
context size)
--dry-sequence-breaker STRING add sequence breaker for DRY sampling, clearing out default breakers
('\n', ':', '"', '*') in the process; use "none" to not use any
sequence breakers
--dynatemp-range N dynamic temperature range (default: 0.0, 0.0 = disabled)
--dynatemp-exp N dynamic temperature exponent (default: 1.0)
--mirostat N use Mirostat sampling.
Top K, Nucleus and Locally Typical samplers are ignored if used.
(default: 0, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)
--mirostat-lr N Mirostat learning rate, parameter eta (default: 0.1)
--mirostat-ent N Mirostat target entropy, parameter tau (default: 5.0)
-l, --logit-bias TOKEN_ID(+/-)BIAS modifies the likelihood of token appearing in the completion,
i.e. `--logit-bias 15043+1` to increase likelihood of token ' Hello',
or `--logit-bias 15043-1` to decrease likelihood of token ' Hello'
--grammar GRAMMAR BNF-like grammar to constrain generations (see samples in grammars/
dir) (default: '')
--grammar-file FNAME file to read grammar from
-j, --json-schema SCHEMA JSON schema to constrain generations (https://json-schema.org/), e.g.
`{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead
-jf, --json-schema-file FILE File containing a JSON schema to constrain generations
(https://json-schema.org/), e.g. `{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead

----- example-specific params -----

--ctx-checkpoints, --swa-checkpoints N
max number of context checkpoints to create per slot (default: 8)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/15293)
(env: LLAMA_ARG_CTX_CHECKPOINTS)
--cache-ram, -cram N set the maximum cache size in MiB (default: 8192, -1 - no limit, 0 -
disable)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/16391)
(env: LLAMA_ARG_CACHE_RAM)
--no-context-shift disables context shift on infinite text generation (default: enabled)
(env: LLAMA_ARG_NO_CONTEXT_SHIFT)
--context-shift enables context shift on infinite text generation (default: disabled)
(env: LLAMA_ARG_CONTEXT_SHIFT)
-r, --reverse-prompt PROMPT halt generation at PROMPT, return control in interactive mode
-sp, --special special tokens output enabled (default: false)
--no-warmup skip warming up the model with an empty run
--spm-infill use Suffix/Prefix/Middle pattern for infill (instead of
Prefix/Suffix/Middle) as some models prefer this. (default: disabled)
--pooling {none,mean,cls,last,rank} pooling type for embeddings, use model default if unspecified
(env: LLAMA_ARG_POOLING)
-cb, --cont-batching enable continuous batching (a.k.a dynamic batching) (default: enabled)
(env: LLAMA_ARG_CONT_BATCHING)
-nocb, --no-cont-batching disable continuous batching
(env: LLAMA_ARG_NO_CONT_BATCHING)
--mmproj FILE path to a multimodal projector file. see tools/mtmd/README.md
note: if -hf is used, this argument can be omitted
(env: LLAMA_ARG_MMPROJ)
--mmproj-url URL URL to a multimodal projector file. see tools/mtmd/README.md
(env: LLAMA_ARG_MMPROJ_URL)
--no-mmproj explicitly disable multimodal projector, useful when using -hf
(env: LLAMA_ARG_NO_MMPROJ)
--no-mmproj-offload do not offload multimodal projector to GPU
(env: LLAMA_ARG_NO_MMPROJ_OFFLOAD)
--image-min-tokens N minimum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MIN_TOKENS)
--image-max-tokens N maximum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MAX_TOKENS)
--override-tensor-draft, -otd <tensor name pattern>=<buffer type>,...
override tensor buffer type for draft model
--cpu-moe-draft, -cmoed keep all Mixture of Experts (MoE) weights in the CPU for the draft
model
(env: LLAMA_ARG_CPU_MOE_DRAFT)
--n-cpu-moe-draft, -ncmoed N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU for the draft model
(env: LLAMA_ARG_N_CPU_MOE_DRAFT)
-a, --alias STRING set alias for model name (to be used by REST API)
(env: LLAMA_ARG_ALIAS)
--host HOST ip address to listen, or bind to an UNIX socket if the address ends
with .sock (default: 127.0.0.1)
(env: LLAMA_ARG_HOST)
--port PORT port to listen (default: 8080)
(env: LLAMA_ARG_PORT)
--path PATH path to serve static files from (default: )
(env: LLAMA_ARG_STATIC_PATH)
--api-prefix PREFIX prefix path the server serves from, without the trailing slash
(default: )
(env: LLAMA_ARG_API_PREFIX)
--no-webui Disable the Web UI (default: enabled)
(env: LLAMA_ARG_NO_WEBUI)
--embedding, --embeddings restrict to only support embedding use case; use only with dedicated
embedding models (default: disabled)
(env: LLAMA_ARG_EMBEDDINGS)
--reranking, --rerank enable reranking endpoint on server (default: disabled)
(env: LLAMA_ARG_RERANKING)
--api-key KEY API key to use for authentication (default: none)
(env: LLAMA_API_KEY)
--api-key-file FNAME path to file containing API keys (default: none)
--ssl-key-file FNAME path to file a PEM-encoded SSL private key
(env: LLAMA_ARG_SSL_KEY_FILE)
--ssl-cert-file FNAME path to file a PEM-encoded SSL certificate
(env: LLAMA_ARG_SSL_CERT_FILE)
--chat-template-kwargs STRING sets additional params for the json template parser
(env: LLAMA_CHAT_TEMPLATE_KWARGS)
-to, --timeout N server read/write timeout in seconds (default: 600)
(env: LLAMA_ARG_TIMEOUT)
--threads-http N number of threads used to process HTTP requests (default: -1)
(env: LLAMA_ARG_THREADS_HTTP)
--cache-reuse N min chunk size to attempt reusing from the cache via KV shifting
(default: 0)
[(card)](https://ggml.ai/f0.png)
(env: LLAMA_ARG_CACHE_REUSE)
--metrics enable prometheus compatible metrics endpoint (default: disabled)
(env: LLAMA_ARG_ENDPOINT_METRICS)
--props enable changing global properties via POST /props (default: disabled)
(env: LLAMA_ARG_ENDPOINT_PROPS)
--slots enable slots monitoring endpoint (default: enabled)
(env: LLAMA_ARG_ENDPOINT_SLOTS)
--no-slots disables slots monitoring endpoint
(env: LLAMA_ARG_NO_ENDPOINT_SLOTS)
--slot-save-path PATH path to save slot kv cache (default: disabled)
--jinja use jinja template for chat (default: disabled)
(env: LLAMA_ARG_JINJA)
--reasoning-format FORMAT controls whether thought tags are allowed and/or extracted from the
response, and in which format they're returned; one of:
- none: leaves thoughts unparsed in `message.content`
- deepseek: puts thoughts in `message.reasoning_content`
- deepseek-legacy: keeps `<think>` tags in `message.content` while
also populating `message.reasoning_content`
(default: auto)
(env: LLAMA_ARG_THINK)
--reasoning-budget N controls the amount of thinking allowed; currently only one of: -1 for
unrestricted thinking budget, or 0 to disable thinking (default: -1)
(env: LLAMA_ARG_THINK_BUDGET)
--chat-template JINJA_TEMPLATE set custom jinja chat template (default: template taken from model's
metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE)
--chat-template-file JINJA_TEMPLATE_FILE
set custom jinja chat template file (default: template taken from
model's metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE_FILE)
--no-prefill-assistant whether to prefill the assistant's response if the last message is an
assistant message (default: prefill enabled)
when this flag is set, if the last message is an assistant message
then it will be treated as a full message and not prefilled

(env: LLAMA_ARG_NO_PREFILL_ASSISTANT)
-sps, --slot-prompt-similarity SIMILARITY
how much the prompt of a request must match the prompt of a slot in
order to use that slot (default: 0.10, 0.0 = disabled)
--lora-init-without-apply load LoRA adapters without applying them (apply later via POST
/lora-adapters) (default: disabled)
-td, --threads-draft N number of threads to use during generation (default: same as
--threads)
-tbd, --threads-batch-draft N number of threads to use during batch and prompt processing (default:
same as --threads-draft)
--draft-max, --draft, --draft-n N number of tokens to draft for speculative decoding (default: 16)
(env: LLAMA_ARG_DRAFT_MAX)
--draft-min, --draft-n-min N minimum number of draft tokens to use for speculative decoding
(default: 0)
(env: LLAMA_ARG_DRAFT_MIN)
--draft-p-min P minimum speculative decoding probability (greedy) (default: 0.8)
(env: LLAMA_ARG_DRAFT_P_MIN)
-cd, --ctx-size-draft N size of the prompt context for the draft model (default: 0, 0 = loaded
from model)
(env: LLAMA_ARG_CTX_SIZE_DRAFT)
-devd, --device-draft <dev1,dev2,..> comma-separated list of devices to use for offloading the draft model
(none = don't offload)
use --list-devices to see a list of available devices
-ngld, --gpu-layers-draft, --n-gpu-layers-draft N
number of layers to store in VRAM for the draft model
(env: LLAMA_ARG_N_GPU_LAYERS_DRAFT)
-md, --model-draft FNAME draft model for speculative decoding (default: unused)
(env: LLAMA_ARG_MODEL_DRAFT)
--spec-replace TARGET DRAFT translate the string in TARGET into DRAFT if the draft model and main
model are not compatible
-mv, --model-vocoder FNAME vocoder model for audio generation (default: unused)
--tts-use-guide-tokens Use guide tokens to improve TTS word recall
--embd-gemma-default use default EmbeddingGemma model (note: can download weights from the
internet)
--fim-qwen-1.5b-default use default Qwen 2.5 Coder 1.5B (note: can download weights from the
internet)
--fim-qwen-3b-default use default Qwen 2.5 Coder 3B (note: can download weights from the
internet)
--fim-qwen-7b-default use default Qwen 2.5 Coder 7B (note: can download weights from the
internet)
--fim-qwen-7b-spec use Qwen 2.5 Coder 7B + 0.5B draft for speculative decoding (note: can
download weights from the internet)
--fim-qwen-14b-spec use Qwen 2.5 Coder 14B + 0.5B draft for speculative decoding (note:
can download weights from the internet)
--fim-qwen-30b-default use default Qwen 3 Coder 30B A3B Instruct (note: can download weights
from the internet)
--gpt-oss-20b-default use gpt-oss-20b (note: can download weights from the internet)
--gpt-oss-120b-default use gpt-oss-120b (note: can download weights from the internet)
--vision-gemma-4b-default use Gemma 3 4B QAT (note: can download weights from the internet)
--vision-gemma-12b-default use Gemma 3 12B QAT (note: can download weights from the internet)
</syntaxhighlight>
=== Test ===
=== Bekannte Probleme ===
<syntaxhighlight lang="bash" line="1">
Error 1 [ 34%] Linking HIP shared library ../../../bin/libggml-hip.so [ 34%] Built target ggml-hip gmake[1]: *** [CMakeFiles/Makefile2:1775: ggml/src/CMakeFiles/ggml-cpu.dir/all] Error 2 gmake: *** [Makefile:146: all] Error 2
</syntaxhighlight>
Lösung:
<syntaxhighlight lang="bash" line="1">
GCC 13.3.0 konnte damit nicht umgehen, aber GCC 14.2.0 hat bessere C++-Standard-Compliance und kann diesen Header-Konflikt korrekt auflösen – deshalb funktioniert die Kompilierung jetzt mit der neueren Compiler-Version.
sudo update-alternatives --remove gcc /usr/bin/gcc-13
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-14 100
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-13 50
sudo update-alternatives --config gcc
# Wähle gcc-14

sudo update-alternatives --remove g++ /usr/bin/g++-13
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-14 100
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-13 50
sudo update-alternatives --config g++
# Wähle g++-14
</syntaxhighlight>
=== Nützliche Links ===
* [https://github.com/ggml-org/llama.cpp Offizielles llama.cpp Repository]
* [https://github.com/ggml-org/llama.cpp/blob/master/examples/server/README.md llama-server Dokumentation]
* [https://github.com/ggml-org/llama.cpp/discussions llama.cpp Discussions]
* [https://huggingface.co/models?library=gguf GGUF Modelle auf Hugging Face]
* [https://docs.rocm.com/ ROCm Dokumentation]
* [https://vulkan.lunarg.com/doc/sdk Vulkan SDK Dokumentation]

Llama.cpp

2026-01-07T14:32:53Z

Hendrik: /* Vulkan Build */

=== Beschreibung ===
llama.cpp ist eine C/C++ Implementierung für Inference von Large Language Models (LLMs). Es unterstützt verschiedene Backends (CPU, Vulkan, ROCm, CUDA) und ermöglicht das Ausführen von quantisierten Modellen im GGUF-Format.

=== Download ===
<syntaxhighlight lang="bash" line="1">
git clone https://github.com/ggml-org/llama.cpp
cd llama.cpp
</syntaxhighlight>

=== Installation ===
==== Vulkan Build ====
<syntaxhighlight lang="bash" line="1">
sudo dnf install libcurl-devel -y
</syntaxhighlight>
[https://vulkan.lunarg.com/sdk/home Vulkan SDK] herunterladen
entpacken, ins Verzeichnis wechseln und source setup-env.sh ausführen. Dann ins Verzeichnis wechseln wo llama.cpp installiert werden soll.
Dort dann
<syntaxhighlight lang="bash" line="1">
source /home/hendrik/Programme/Vulkan_SDK/1.4.335.0/setup-env.sh
cd /home/hendrik/Programme/llama.cpp/
rm -R build-vulkan
cmake -B build-vulkan -DGGML_VULKAN=1
cmake --build build-vulkan --config Release -- -j $(nproc)
</syntaxhighlight>

==== ROCm Build (getestet mit ROCm 6.4.3 und 7.1.0 ====
gfx906 = Mi 50 Support 
gfx1100 = 7900 XTX Support

<syntaxhighlight lang="bash" line="1">
Copy & paste all the commands from the quick install https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/quick-start.html
Before rebooting to complete the install, download the 6.4 rocblas from the AUR: https://archlinux.org/packages/extra/x86_64/rocblas/
Extract it
Copy all tensor files that contain gfx906 in rocblas-6.4.3-3-x86_64.pkg/opt/rocm/lib/rocblas/library to /opt/rocm/lib/rocblas/library
Reboot
Check if it worked by running sudo update-alternatives --display rocm
</syntaxhighlight>
# ROCm Umgebung einrichten (optional falls Fehler auftreten)
<syntaxhighlight lang="bash" line="1">
echo 'export LD_LIBRARY_PATH=/opt/rocm/lib:$LD_LIBRARY_PATH' >> ~/.bashrc
echo 'export HSA_OVERRIDE_GFX_VERSION=9.0.6' >> ~/.bashrc # Für MI50/MI60
source ~/.bashrc
</syntaxhighlight>
# llama.cpp kompilieren
<syntaxhighlight lang="bash" line="1">
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
cmake -B build-rocm -DGGML_HIP=ON -DAMDGPU_TARGETS=gfx906 -DGGML_HIP_ROCWMMA_FATTN=ON \
-DCMAKE_BUILD_TYPE=Release
cmake --build build-rocm --config Release -- -j 8
</syntaxhighlight>

Für mehrere GPU-Architekturen gleichzeitig:
<syntaxhighlight lang="bash" line="1">
# Für MI50 (gfx906) und RX 7900 XTX (gfx1100)
HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
cmake -S . -B build -DGGML_HIP=ON -DAMDGPU_TARGETS="gfx906;gfx1100" \
-DCMAKE_BUILD_TYPE=Release && \
cmake --build build --config Release -- -j 8
</syntaxhighlight>

=== Konfiguration ===

==== llama-server als systemd Service einrichten ====

'''Service-Datei erstellen:'''
<syntaxhighlight lang="bash" line="1">
sudo nano /etc/systemd/system/llama-server.service
</syntaxhighlight>

'''Service-Datei Inhalt (Multi-GPU mit ROCm):'''
<syntaxhighlight lang="ini" line="1">
[Unit]
Description=Llama.cpp ROCm Multi-GPU Server
After=network.target

[Service]
Type=simple
User=username
Group=username
WorkingDirectory=/home/username/llama.cpp/build/bin

# Multi-GPU Konfiguration
Environment="HIP_VISIBLE_DEVICES=0,1"
Environment="HSA_OVERRIDE_GFX_VERSION=9.0.6"
Environment="PATH=/opt/rocm/bin:/usr/local/bin:/usr/bin:/bin"
Environment="LD_LIBRARY_PATH=/opt/rocm/lib"

# Server mit optimalen Multi-GPU Einstellungen
ExecStart=/home/username/llama.cpp/build/bin/llama-server \
-m /home/username/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja

Restart=always
RestartSec=10
LimitNOFILE=65535
LimitMEMLOCK=infinity
StandardOutput=journal
StandardError=journal
SyslogIdentifier=llama-server

[Install]
WantedBy=multi-user.target
</syntaxhighlight>

'''Service aktivieren und starten:'''
<syntaxhighlight lang="bash" line="1">
# Service neu laden
sudo systemctl daemon-reload

# Service aktivieren (Auto-Start beim Boot)
sudo systemctl enable llama-server

# Service starten
sudo systemctl start llama-server

# Status prüfen
sudo systemctl status llama-server

# Logs anzeigen
sudo journalctl -u llama-server -f
</syntaxhighlight>

'''Service Management:'''
<syntaxhighlight lang="bash" line="1">
# Service stoppen
sudo systemctl stop llama-server

# Service neustarten
sudo systemctl restart llama-server

# Service deaktivieren
sudo systemctl disable llama-server

==== Manuelle Server-Starts ====
'''Multi-GPU (ROCm) - Optimal:'''
<syntaxhighlight lang="bash" line="1">
HIP_VISIBLE_DEVICES=0,1 ./llama-server \
-m ~/models/model.gguf \
--split-mode row \
--tensor-split 0.5,0.5 \
-ngl 99 \
-fa 1 \
--host 0.0.0.0 \
--port 8080 \
-c 32768 \
-b 2048 \
-ub 2048 \
--threads 8 \
--parallel 1 \
--jinja
</syntaxhighlight>

=== Update ===
<syntaxhighlight lang="bash" line="1">
cd ~/llama.cpp
git pull
cmake --build build --config Release -- -j $(nproc)

# Service neu starten falls aktiv
sudo systemctl restart llama-server
</syntaxhighlight>
=== Befehlszeilenargumente ===
llama-server
<syntaxhighlight lang="bash" line="1">
-h, --help, --usage print usage and exit
--version show version and build info
-cl, --cache-list show list of models in cache
--completion-bash print source-able bash completion script for llama.cpp
--verbose-prompt print a verbose prompt before generation (default: false)
-t, --threads N number of CPU threads to use during generation (default: -1)
(env: LLAMA_ARG_THREADS)
-tb, --threads-batch N number of threads to use during batch and prompt processing (default:
same as --threads)
-C, --cpu-mask M CPU affinity mask: arbitrarily long hex. Complements cpu-range
(default: "")
-Cr, --cpu-range lo-hi range of CPUs for affinity. Complements --cpu-mask
--cpu-strict <0|1> use strict CPU placement (default: 0)
--prio N set process/thread priority : low(-1), normal(0), medium(1), high(2),
realtime(3) (default: 0)
--poll <0...100> use polling level to wait for work (0 - no polling, default: 50)
-Cb, --cpu-mask-batch M CPU affinity mask: arbitrarily long hex. Complements cpu-range-batch
(default: same as --cpu-mask)
-Crb, --cpu-range-batch lo-hi ranges of CPUs for affinity. Complements --cpu-mask-batch
--cpu-strict-batch <0|1> use strict CPU placement (default: same as --cpu-strict)
--prio-batch N set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime
(default: 0)
--poll-batch <0|1> use polling to wait for work (default: same as --poll)
-c, --ctx-size N size of the prompt context (default: 4096, 0 = loaded from model)
(env: LLAMA_ARG_CTX_SIZE)
-n, --predict, --n-predict N number of tokens to predict (default: -1, -1 = infinity)
(env: LLAMA_ARG_N_PREDICT)
-b, --batch-size N logical maximum batch size (default: 2048)
(env: LLAMA_ARG_BATCH)
-ub, --ubatch-size N physical maximum batch size (default: 512)
(env: LLAMA_ARG_UBATCH)
--keep N number of tokens to keep from the initial prompt (default: 0, -1 =
all)
--swa-full use full-size SWA cache (default: false)
[(more
info)](https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
(env: LLAMA_ARG_SWA_FULL)
--kv-unified, -kvu use single unified KV buffer for the KV cache of all sequences
(default: false)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/14363)
(env: LLAMA_ARG_KV_SPLIT)
-fa, --flash-attn [on|off|auto] set Flash Attention use ('on', 'off', or 'auto', default: 'auto')
(env: LLAMA_ARG_FLASH_ATTN)
--no-perf disable internal libllama performance timings (default: false)
(env: LLAMA_ARG_NO_PERF)
-e, --escape process escapes sequences (\n, \r, \t, \', \", \\) (default: true)
--no-escape do not process escape sequences
--rope-scaling {none,linear,yarn} RoPE frequency scaling method, defaults to linear unless specified by
the model
(env: LLAMA_ARG_ROPE_SCALING_TYPE)
--rope-scale N RoPE context scaling factor, expands context by a factor of N
(env: LLAMA_ARG_ROPE_SCALE)
--rope-freq-base N RoPE base frequency, used by NTK-aware scaling (default: loaded from
model)
(env: LLAMA_ARG_ROPE_FREQ_BASE)
--rope-freq-scale N RoPE frequency scaling factor, expands context by a factor of 1/N
(env: LLAMA_ARG_ROPE_FREQ_SCALE)
--yarn-orig-ctx N YaRN: original context size of model (default: 0 = model training
context size)
(env: LLAMA_ARG_YARN_ORIG_CTX)
--yarn-ext-factor N YaRN: extrapolation mix factor (default: -1.0, 0.0 = full
interpolation)
(env: LLAMA_ARG_YARN_EXT_FACTOR)
--yarn-attn-factor N YaRN: scale sqrt(t) or attention magnitude (default: -1.0)
(env: LLAMA_ARG_YARN_ATTN_FACTOR)
--yarn-beta-slow N YaRN: high correction dim or alpha (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_SLOW)
--yarn-beta-fast N YaRN: low correction dim or beta (default: -1.0)
(env: LLAMA_ARG_YARN_BETA_FAST)
-nkvo, --no-kv-offload disable KV offload
(env: LLAMA_ARG_NO_KV_OFFLOAD)
-nr, --no-repack disable weight repacking
(env: LLAMA_ARG_NO_REPACK)
--no-host bypass host buffer allowing extra buffers to be used
(env: LLAMA_ARG_NO_HOST)
-ctk, --cache-type-k TYPE KV cache data type for K
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K)
-ctv, --cache-type-v TYPE KV cache data type for V
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V)
-dt, --defrag-thold N KV cache defragmentation threshold (DEPRECATED)
(env: LLAMA_ARG_DEFRAG_THOLD)
-np, --parallel N number of parallel sequences to decode (default: 1)
(env: LLAMA_ARG_N_PARALLEL)
--mlock force system to keep model in RAM rather than swapping or compressing
(env: LLAMA_ARG_MLOCK)
--no-mmap do not memory-map model (slower load but may reduce pageouts if not
using mlock)
(env: LLAMA_ARG_NO_MMAP)
--numa TYPE attempt optimizations that help on some NUMA systems
- distribute: spread execution evenly over all nodes
- isolate: only spawn threads on CPUs on the node that execution
started on
- numactl: use the CPU map provided by numactl
if run without this previously, it is recommended to drop the system
page cache before using this
see https://github.com/ggml-org/llama.cpp/issues/1437
(env: LLAMA_ARG_NUMA)
-dev, --device <dev1,dev2,..> comma-separated list of devices to use for offloading (none = don't
offload)
use --list-devices to see a list of available devices
(env: LLAMA_ARG_DEVICE)
--list-devices print list of available devices and exit
--override-tensor, -ot <tensor name pattern>=<buffer type>,...
override tensor buffer type
--cpu-moe, -cmoe keep all Mixture of Experts (MoE) weights in the CPU
(env: LLAMA_ARG_CPU_MOE)
--n-cpu-moe, -ncmoe N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU
(env: LLAMA_ARG_N_CPU_MOE)
-ngl, --gpu-layers, --n-gpu-layers N max. number of layers to store in VRAM (default: -1)
(env: LLAMA_ARG_N_GPU_LAYERS)
-sm, --split-mode {none,layer,row} how to split the model across multiple GPUs, one of:
- none: use one GPU only
- layer (default): split layers and KV across GPUs
- row: split rows across GPUs
(env: LLAMA_ARG_SPLIT_MODE)
-ts, --tensor-split N0,N1,N2,... fraction of the model to offload to each GPU, comma-separated list of
proportions, e.g. 3,1
(env: LLAMA_ARG_TENSOR_SPLIT)
-mg, --main-gpu INDEX the GPU to use for the model (with split-mode = none), or for
intermediate results and KV (with split-mode = row) (default: 0)
(env: LLAMA_ARG_MAIN_GPU)
--check-tensors check model tensor data for invalid values (default: false)
--override-kv KEY=TYPE:VALUE advanced option to override model metadata by key. may be specified
multiple times.
types: int, float, bool, str. example: --override-kv
tokenizer.ggml.add_bos_token=bool:false
--no-op-offload disable offloading host tensor operations to device (default: false)
--lora FNAME path to LoRA adapter (can be repeated to use multiple adapters)
--lora-scaled FNAME SCALE path to LoRA adapter with user defined scaling (can be repeated to use
multiple adapters)
--control-vector FNAME add a control vector
note: this argument can be repeated to add multiple control vectors
--control-vector-scaled FNAME SCALE add a control vector with user defined scaling SCALE
note: this argument can be repeated to add multiple scaled control
vectors
--control-vector-layer-range START END
layer range to apply the control vector(s) to, start and end inclusive
-m, --model FNAME model path (default: `models/$filename` with filename from `--hf-file`
or `--model-url` if set, otherwise models/7B/ggml-model-f16.gguf)
(env: LLAMA_ARG_MODEL)
-mu, --model-url MODEL_URL model download url (default: unused)
(env: LLAMA_ARG_MODEL_URL)
-dr, --docker-repo [<repo>/]<model>[:quant]
Docker Hub model repository. repo is optional, default to ai/. quant
is optional, default to :latest.
example: gemma3
(default: unused)
(env: LLAMA_ARG_DOCKER_REPO)
-hf, -hfr, --hf-repo <user>/<model>[:quant]
Hugging Face model repository; quant is optional, case-insensitive,
default to Q4_K_M, or falls back to the first file in the repo if
Q4_K_M doesn't exist.
mmproj is also downloaded automatically if available. to disable, add
--no-mmproj
example: unsloth/phi-4-GGUF:q4_k_m
(default: unused)
(env: LLAMA_ARG_HF_REPO)
-hfd, -hfrd, --hf-repo-draft <user>/<model>[:quant]
Same as --hf-repo, but for the draft model (default: unused)
(env: LLAMA_ARG_HFD_REPO)
-hff, --hf-file FILE Hugging Face model file. If specified, it will override the quant in
--hf-repo (default: unused)
(env: LLAMA_ARG_HF_FILE)
-hfv, -hfrv, --hf-repo-v <user>/<model>[:quant]
Hugging Face model repository for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_REPO_V)
-hffv, --hf-file-v FILE Hugging Face model file for the vocoder model (default: unused)
(env: LLAMA_ARG_HF_FILE_V)
-hft, --hf-token TOKEN Hugging Face access token (default: value from HF_TOKEN environment
variable)
(env: HF_TOKEN)
--log-disable Log disable
--log-file FNAME Log to file
--log-colors [on|off|auto] Set colored logging ('on', 'off', or 'auto', default: 'auto')
'auto' enables colors when output is to a terminal
(env: LLAMA_LOG_COLORS)
-v, --verbose, --log-verbose Set verbosity level to infinity (i.e. log all messages, useful for
debugging)
--offline Offline mode: forces use of cache, prevents network access
(env: LLAMA_OFFLINE)
-lv, --verbosity, --log-verbosity N Set the verbosity threshold. Messages with a higher verbosity will be
ignored.
(env: LLAMA_LOG_VERBOSITY)
--log-prefix Enable prefix in log messages
(env: LLAMA_LOG_PREFIX)
--log-timestamps Enable timestamps in log messages
(env: LLAMA_LOG_TIMESTAMPS)
-ctkd, --cache-type-k-draft TYPE KV cache data type for K for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K_DRAFT)
-ctvd, --cache-type-v-draft TYPE KV cache data type for V for the draft model
allowed values: f32, f16, bf16, q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
(default: f16)
(env: LLAMA_ARG_CACHE_TYPE_V_DRAFT)

----- sampling params -----

--samplers SAMPLERS samplers that will be used for generation in the order, separated by
';'
(default:
penalties;dry;top_n_sigma;top_k;typ_p;top_p;min_p;xtc;temperature)
-s, --seed SEED RNG seed (default: -1, use random seed for -1)
--sampling-seq, --sampler-seq SEQUENCE
simplified sequence for samplers that will be used (default:
edskypmxt)
--ignore-eos ignore end of stream token and continue generating (implies
--logit-bias EOS-inf)
--temp N temperature (default: 0.8)
--top-k N top-k sampling (default: 40, 0 = disabled)
--top-p N top-p sampling (default: 0.9, 1.0 = disabled)
--min-p N min-p sampling (default: 0.1, 0.0 = disabled)
--top-nsigma N top-n-sigma sampling (default: -1.0, -1.0 = disabled)
--xtc-probability N xtc probability (default: 0.0, 0.0 = disabled)
--xtc-threshold N xtc threshold (default: 0.1, 1.0 = disabled)
--typical N locally typical sampling, parameter p (default: 1.0, 1.0 = disabled)
--repeat-last-n N last n tokens to consider for penalize (default: 64, 0 = disabled, -1
= ctx_size)
--repeat-penalty N penalize repeat sequence of tokens (default: 1.0, 1.0 = disabled)
--presence-penalty N repeat alpha presence penalty (default: 0.0, 0.0 = disabled)
--frequency-penalty N repeat alpha frequency penalty (default: 0.0, 0.0 = disabled)
--dry-multiplier N set DRY sampling multiplier (default: 0.0, 0.0 = disabled)
--dry-base N set DRY sampling base value (default: 1.75)
--dry-allowed-length N set allowed length for DRY sampling (default: 2)
--dry-penalty-last-n N set DRY penalty for the last n tokens (default: -1, 0 = disable, -1 =
context size)
--dry-sequence-breaker STRING add sequence breaker for DRY sampling, clearing out default breakers
('\n', ':', '"', '*') in the process; use "none" to not use any
sequence breakers
--dynatemp-range N dynamic temperature range (default: 0.0, 0.0 = disabled)
--dynatemp-exp N dynamic temperature exponent (default: 1.0)
--mirostat N use Mirostat sampling.
Top K, Nucleus and Locally Typical samplers are ignored if used.
(default: 0, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)
--mirostat-lr N Mirostat learning rate, parameter eta (default: 0.1)
--mirostat-ent N Mirostat target entropy, parameter tau (default: 5.0)
-l, --logit-bias TOKEN_ID(+/-)BIAS modifies the likelihood of token appearing in the completion,
i.e. `--logit-bias 15043+1` to increase likelihood of token ' Hello',
or `--logit-bias 15043-1` to decrease likelihood of token ' Hello'
--grammar GRAMMAR BNF-like grammar to constrain generations (see samples in grammars/
dir) (default: '')
--grammar-file FNAME file to read grammar from
-j, --json-schema SCHEMA JSON schema to constrain generations (https://json-schema.org/), e.g.
`{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead
-jf, --json-schema-file FILE File containing a JSON schema to constrain generations
(https://json-schema.org/), e.g. `{}` for any JSON object
For schemas w/ external $refs, use --grammar +
example/json_schema_to_grammar.py instead

----- example-specific params -----

--ctx-checkpoints, --swa-checkpoints N
max number of context checkpoints to create per slot (default: 8)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/15293)
(env: LLAMA_ARG_CTX_CHECKPOINTS)
--cache-ram, -cram N set the maximum cache size in MiB (default: 8192, -1 - no limit, 0 -
disable)
[(more info)](https://github.com/ggml-org/llama.cpp/pull/16391)
(env: LLAMA_ARG_CACHE_RAM)
--no-context-shift disables context shift on infinite text generation (default: enabled)
(env: LLAMA_ARG_NO_CONTEXT_SHIFT)
--context-shift enables context shift on infinite text generation (default: disabled)
(env: LLAMA_ARG_CONTEXT_SHIFT)
-r, --reverse-prompt PROMPT halt generation at PROMPT, return control in interactive mode
-sp, --special special tokens output enabled (default: false)
--no-warmup skip warming up the model with an empty run
--spm-infill use Suffix/Prefix/Middle pattern for infill (instead of
Prefix/Suffix/Middle) as some models prefer this. (default: disabled)
--pooling {none,mean,cls,last,rank} pooling type for embeddings, use model default if unspecified
(env: LLAMA_ARG_POOLING)
-cb, --cont-batching enable continuous batching (a.k.a dynamic batching) (default: enabled)
(env: LLAMA_ARG_CONT_BATCHING)
-nocb, --no-cont-batching disable continuous batching
(env: LLAMA_ARG_NO_CONT_BATCHING)
--mmproj FILE path to a multimodal projector file. see tools/mtmd/README.md
note: if -hf is used, this argument can be omitted
(env: LLAMA_ARG_MMPROJ)
--mmproj-url URL URL to a multimodal projector file. see tools/mtmd/README.md
(env: LLAMA_ARG_MMPROJ_URL)
--no-mmproj explicitly disable multimodal projector, useful when using -hf
(env: LLAMA_ARG_NO_MMPROJ)
--no-mmproj-offload do not offload multimodal projector to GPU
(env: LLAMA_ARG_NO_MMPROJ_OFFLOAD)
--image-min-tokens N minimum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MIN_TOKENS)
--image-max-tokens N maximum number of tokens each image can take, only used by vision
models with dynamic resolution (default: read from model)
(env: LLAMA_ARG_IMAGE_MAX_TOKENS)
--override-tensor-draft, -otd <tensor name pattern>=<buffer type>,...
override tensor buffer type for draft model
--cpu-moe-draft, -cmoed keep all Mixture of Experts (MoE) weights in the CPU for the draft
model
(env: LLAMA_ARG_CPU_MOE_DRAFT)
--n-cpu-moe-draft, -ncmoed N keep the Mixture of Experts (MoE) weights of the first N layers in the
CPU for the draft model
(env: LLAMA_ARG_N_CPU_MOE_DRAFT)
-a, --alias STRING set alias for model name (to be used by REST API)
(env: LLAMA_ARG_ALIAS)
--host HOST ip address to listen, or bind to an UNIX socket if the address ends
with .sock (default: 127.0.0.1)
(env: LLAMA_ARG_HOST)
--port PORT port to listen (default: 8080)
(env: LLAMA_ARG_PORT)
--path PATH path to serve static files from (default: )
(env: LLAMA_ARG_STATIC_PATH)
--api-prefix PREFIX prefix path the server serves from, without the trailing slash
(default: )
(env: LLAMA_ARG_API_PREFIX)
--no-webui Disable the Web UI (default: enabled)
(env: LLAMA_ARG_NO_WEBUI)
--embedding, --embeddings restrict to only support embedding use case; use only with dedicated
embedding models (default: disabled)
(env: LLAMA_ARG_EMBEDDINGS)
--reranking, --rerank enable reranking endpoint on server (default: disabled)
(env: LLAMA_ARG_RERANKING)
--api-key KEY API key to use for authentication (default: none)
(env: LLAMA_API_KEY)
--api-key-file FNAME path to file containing API keys (default: none)
--ssl-key-file FNAME path to file a PEM-encoded SSL private key
(env: LLAMA_ARG_SSL_KEY_FILE)
--ssl-cert-file FNAME path to file a PEM-encoded SSL certificate
(env: LLAMA_ARG_SSL_CERT_FILE)
--chat-template-kwargs STRING sets additional params for the json template parser
(env: LLAMA_CHAT_TEMPLATE_KWARGS)
-to, --timeout N server read/write timeout in seconds (default: 600)
(env: LLAMA_ARG_TIMEOUT)
--threads-http N number of threads used to process HTTP requests (default: -1)
(env: LLAMA_ARG_THREADS_HTTP)
--cache-reuse N min chunk size to attempt reusing from the cache via KV shifting
(default: 0)
[(card)](https://ggml.ai/f0.png)
(env: LLAMA_ARG_CACHE_REUSE)
--metrics enable prometheus compatible metrics endpoint (default: disabled)
(env: LLAMA_ARG_ENDPOINT_METRICS)
--props enable changing global properties via POST /props (default: disabled)
(env: LLAMA_ARG_ENDPOINT_PROPS)
--slots enable slots monitoring endpoint (default: enabled)
(env: LLAMA_ARG_ENDPOINT_SLOTS)
--no-slots disables slots monitoring endpoint
(env: LLAMA_ARG_NO_ENDPOINT_SLOTS)
--slot-save-path PATH path to save slot kv cache (default: disabled)
--jinja use jinja template for chat (default: disabled)
(env: LLAMA_ARG_JINJA)
--reasoning-format FORMAT controls whether thought tags are allowed and/or extracted from the
response, and in which format they're returned; one of:
- none: leaves thoughts unparsed in `message.content`
- deepseek: puts thoughts in `message.reasoning_content`
- deepseek-legacy: keeps `<think>` tags in `message.content` while
also populating `message.reasoning_content`
(default: auto)
(env: LLAMA_ARG_THINK)
--reasoning-budget N controls the amount of thinking allowed; currently only one of: -1 for
unrestricted thinking budget, or 0 to disable thinking (default: -1)
(env: LLAMA_ARG_THINK_BUDGET)
--chat-template JINJA_TEMPLATE set custom jinja chat template (default: template taken from model's
metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE)
--chat-template-file JINJA_TEMPLATE_FILE
set custom jinja chat template file (default: template taken from
model's metadata)
if suffix/prefix are specified, template will be disabled
only commonly used templates are accepted (unless --jinja is set
before this flag):
list of built-in templates:
bailing, bailing-think, bailing2, chatglm3, chatglm4, chatml,
command-r, deepseek, deepseek2, deepseek3, exaone3, exaone4, falcon3,
gemma, gigachat, glmedge, gpt-oss, granite, grok-2, hunyuan-dense,
hunyuan-moe, kimi-k2, llama2, llama2-sys, llama2-sys-bos,
llama2-sys-strip, llama3, llama4, megrez, minicpm, mistral-v1,
mistral-v3, mistral-v3-tekken, mistral-v7, mistral-v7-tekken, monarch,
openchat, orion, pangu-embedded, phi3, phi4, rwkv-world, seed_oss,
smolvlm, vicuna, vicuna-orca, yandex, zephyr
(env: LLAMA_ARG_CHAT_TEMPLATE_FILE)
--no-prefill-assistant whether to prefill the assistant's response if the last message is an
assistant message (default: prefill enabled)
when this flag is set, if the last message is an assistant message
then it will be treated as a full message and not prefilled

(env: LLAMA_ARG_NO_PREFILL_ASSISTANT)
-sps, --slot-prompt-similarity SIMILARITY
how much the prompt of a request must match the prompt of a slot in
order to use that slot (default: 0.10, 0.0 = disabled)
--lora-init-without-apply load LoRA adapters without applying them (apply later via POST
/lora-adapters) (default: disabled)
-td, --threads-draft N number of threads to use during generation (default: same as
--threads)
-tbd, --threads-batch-draft N number of threads to use during batch and prompt processing (default:
same as --threads-draft)
--draft-max, --draft, --draft-n N number of tokens to draft for speculative decoding (default: 16)
(env: LLAMA_ARG_DRAFT_MAX)
--draft-min, --draft-n-min N minimum number of draft tokens to use for speculative decoding
(default: 0)
(env: LLAMA_ARG_DRAFT_MIN)
--draft-p-min P minimum speculative decoding probability (greedy) (default: 0.8)
(env: LLAMA_ARG_DRAFT_P_MIN)
-cd, --ctx-size-draft N size of the prompt context for the draft model (default: 0, 0 = loaded
from model)
(env: LLAMA_ARG_CTX_SIZE_DRAFT)
-devd, --device-draft <dev1,dev2,..> comma-separated list of devices to use for offloading the draft model
(none = don't offload)
use --list-devices to see a list of available devices
-ngld, --gpu-layers-draft, --n-gpu-layers-draft N
number of layers to store in VRAM for the draft model
(env: LLAMA_ARG_N_GPU_LAYERS_DRAFT)
-md, --model-draft FNAME draft model for speculative decoding (default: unused)
(env: LLAMA_ARG_MODEL_DRAFT)
--spec-replace TARGET DRAFT translate the string in TARGET into DRAFT if the draft model and main
model are not compatible
-mv, --model-vocoder FNAME vocoder model for audio generation (default: unused)
--tts-use-guide-tokens Use guide tokens to improve TTS word recall
--embd-gemma-default use default EmbeddingGemma model (note: can download weights from the
internet)
--fim-qwen-1.5b-default use default Qwen 2.5 Coder 1.5B (note: can download weights from the
internet)
--fim-qwen-3b-default use default Qwen 2.5 Coder 3B (note: can download weights from the
internet)
--fim-qwen-7b-default use default Qwen 2.5 Coder 7B (note: can download weights from the
internet)
--fim-qwen-7b-spec use Qwen 2.5 Coder 7B + 0.5B draft for speculative decoding (note: can
download weights from the internet)
--fim-qwen-14b-spec use Qwen 2.5 Coder 14B + 0.5B draft for speculative decoding (note:
can download weights from the internet)
--fim-qwen-30b-default use default Qwen 3 Coder 30B A3B Instruct (note: can download weights
from the internet)
--gpt-oss-20b-default use gpt-oss-20b (note: can download weights from the internet)
--gpt-oss-120b-default use gpt-oss-120b (note: can download weights from the internet)
--vision-gemma-4b-default use Gemma 3 4B QAT (note: can download weights from the internet)
--vision-gemma-12b-default use Gemma 3 12B QAT (note: can download weights from the internet)
</syntaxhighlight>
=== Test ===
=== Bekannte Probleme ===
<syntaxhighlight lang="bash" line="1">
Error 1 [ 34%] Linking HIP shared library ../../../bin/libggml-hip.so [ 34%] Built target ggml-hip gmake[1]: *** [CMakeFiles/Makefile2:1775: ggml/src/CMakeFiles/ggml-cpu.dir/all] Error 2 gmake: *** [Makefile:146: all] Error 2
</syntaxhighlight>
Lösung:
<syntaxhighlight lang="bash" line="1">
GCC 13.3.0 konnte damit nicht umgehen, aber GCC 14.2.0 hat bessere C++-Standard-Compliance und kann diesen Header-Konflikt korrekt auflösen – deshalb funktioniert die Kompilierung jetzt mit der neueren Compiler-Version.
sudo update-alternatives --remove gcc /usr/bin/gcc-13
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-14 100
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-13 50
sudo update-alternatives --config gcc
# Wähle gcc-14

sudo update-alternatives --remove g++ /usr/bin/g++-13
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-14 100
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-13 50
sudo update-alternatives --config g++
# Wähle g++-14
</syntaxhighlight>
=== Nützliche Links ===
* [https://github.com/ggml-org/llama.cpp Offizielles llama.cpp Repository]
* [https://github.com/ggml-org/llama.cpp/blob/master/examples/server/README.md llama-server Dokumentation]
* [https://github.com/ggml-org/llama.cpp/discussions llama.cpp Discussions]
* [https://huggingface.co/models?library=gguf GGUF Modelle auf Hugging Face]
* [https://docs.rocm.com/ ROCm Dokumentation]
* [https://vulkan.lunarg.com/doc/sdk Vulkan SDK Dokumentation]

Oracle

2026-01-06T09:48:34Z

Hendrik: /* SAP* kopieren (000 nach 200) */

=== Beschreibung ===

=== Download ===

=== Installation ===

=== Konfiguration ===
==== SAP* löschen ====
sqlplus / as sysdba
<syntaxhighlight lang="sql" line="1">
DELETE FROM SAPSR3.USR02 WHERE MANDT = '000' AND BNAME = 'SAP*';
COMMIT;
</syntaxhighlight>
==== SAP* kopieren (000 nach 200) ====
<syntaxhighlight lang="oracle11" line="1">
INSERT INTO SAPSR3.USR02 (
MANDT, BNAME, BCODE, GLTGV, GLTGB, USTYP,
CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT,
TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2,
BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4,
BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN,
CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE,
RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE,
PWDSALTEDHASH, SECURITY_POLICY
)
SELECT
'200', BNAME, BCODE, GLTGV, GLTGB, USTYP,
CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT,
TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2,
BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4,
BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN,
CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE,
RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE,
PWDSALTEDHASH, SECURITY_POLICY
FROM SAPSR3.USR02
WHERE MANDT = '000' AND BNAME = 'SAP*';
COMMIT;
</syntaxhighlight>

==== Tablespaces und Files anzeigen lassen ====
<syntaxhighlight lang="sql" line="1">
SELECT file_name, bytes/1024/1024 AS size_mb FROM dba_data_files WHERE tablespace_name = 'PSAPSR3' ORDER BY file_name;
</syntaxhighlight>
====Tablespace anlegen====
CREATE TABLESPACE PSAPSR3740X DATAFILE '/oracle/PZ0/sapdata1/sr3740x_1' SIZE 10M;

= Erweiterung ohne ASM =

== Tablespace-Belegung ermitteln ==

Belegung aller Tablespaces in Prozent ermitteln:

<syntaxhighlight lang="sql">
with MBYTES as(
select tablespace_name,
SUM(bytes)/1024/1024 as Akt,
SUM(decode(maxbytes,0,bytes,maxbytes))/1024/1024 as ZMax
from dba_data_files
group by tablespace_name
),
MFREE as(
select tablespace_name,
SUM(bytes)/1024/1024 as Free
from dba_free_space
group by tablespace_name
)
select MBYTES.tablespace_name,
ROUND((AKT-FREE)/ZMAX*100,2) as BelegtProz
from MBYTES, MFREE
where MFREE.tablespace_name = MBYTES.tablespace_name
order by ROUND((AKT-FREE)/ZMAX*100,2) desc;
</syntaxhighlight>

'''Beispiel-Ergebnis:'''
<pre>
TABLESPACE_NAME BELEGTPROZ
------------------------------ ----------
PSAPSR3 89.43
PSAPSR3702 33.32
PSAPSR3USR 20.26
PSAPUNDO 6.55
SYSTEM 5.05
SYSAUX 4.05
</pre>

== Verfügbaren Speicherplatz ermitteln ==

Im Betriebssystem ermitteln, in welchem data-Laufwerk am meisten Platz für die Erweiterung vorhanden ist:

<syntaxhighlight lang="bash">
df -h | grep oracle
</syntaxhighlight>

'''Beispiel-Ausgabe:'''
<pre>
p11data01/data01 98G 68G 30G 70% /oracle/P11/data01
p11data02/data02 215G 172G 43G 81% /oracle/P11/data02
p11data03/data03 98G 51G 47G 52% /oracle/P11/data03
</pre>

== An der Datenbank anmelden ==

<syntaxhighlight lang="sql">
sqlplus / as sysdba
</syntaxhighlight>

=== Ausgabe formatieren ===

Für eine übersichtlichere Darstellung:

<syntaxhighlight lang="sql">
set pages 9999
set lin 180
col file_name for a60
</syntaxhighlight>

== Datafiles anzeigen ==

Datafiles zum Tablespace anzeigen:

<syntaxhighlight lang="sql">
select file_name, bytes/1024/1024
from dba_data_files
where tablespace_name = '<TABLESPACE>';
</syntaxhighlight>

{{Hinweis|Der Name des Tablespaces ist in der E-Mail von Icinga zu finden.}}

'''Beispiel-Ausgabe:'''
<pre>
FILE_NAME BYTES/1024/1024
------------------------------------------------------------ ---------------
/oracle/P11/data01/beispiel_data_01.dbf 20480
/oracle/P11/data02/beispiel_data_02.dbf 20480
/oracle/P11/data03/beispiel_data_03.dbf 30720
/oracle/P11/data02/beispiel_data_04.dbf 20480
</pre>

== Datafile erweitern ==

=== Existierendes Datafile vergrößern ===

Wenn ein Datafile erweitert werden kann:

<syntaxhighlight lang="sql">
alter database datafile '/oracle/P11/data02/beispiel_04.dbf' resize 30G;
</syntaxhighlight>

{{Hinweis|Der Prozess dauert ca. 1 GB pro Minute.}}

'''Überwachung auf OS-Ebene:'''

<syntaxhighlight lang="bash">
du -h <FILENAME>
</syntaxhighlight>

'''Kontrolle:'''

<syntaxhighlight lang="sql">
select file_name, bytes/1024/1024
from dba_data_files
where tablespace_name = '<TABLESPACE>';
</syntaxhighlight>

'''Beispiel nach Erweiterung:'''
<pre>
FILE_NAME BYTES/1024/1024
------------------------------------------------------------ ---------------
/oracle/P11/data01/beispiel_data_01.dbf 20480
/oracle/P11/data02/beispiel_data_02.dbf 20480
/oracle/P11/data03/beispiel_data_03.dbf 30720
/oracle/P11/data02/beispiel_data_04.dbf 30720
</pre>

=== Neues Datafile anlegen ===

Falls kein existierendes Datafile erweitert werden kann:

<syntaxhighlight lang="sql">
alter tablespace BEISPIEL add datafile '/oracle/P11/data03/beispiel.dbf' size 100M;
</syntaxhighlight>

{{Hinweis|Der Prozess dauert ca. 1 GB pro Minute.}}

'''Überwachung auf OS-Ebene:'''

<syntaxhighlight lang="bash">
du -h <FILENAME>
</syntaxhighlight>

'''Kontrolle:'''

<syntaxhighlight lang="sql">
select file_name, bytes/1024/1024
from dba_data_files
where tablespace_name = '<TABLESPACE>';
</syntaxhighlight>

'''Beispiel nach Anlegen:'''
<pre>
FILE_NAME BYTES/1024/1024
------------------------------------------------------------ ---------------
/oracle/P11/data01/beispiel_data_01.dbf 20480
/oracle/P11/data02/beispiel_data_02.dbf 20480
/oracle/P11/data03/beispiel_data_03.dbf 30720
/oracle/P11/data02/beispiel_data_04.dbf 20480
/oracle/P11/data03/beispiel_data_05.dbf 100
</pre>

== Tempfile erweitern ==

'''Tempfiles anzeigen:'''

<syntaxhighlight lang="sql">
select file_name, bytes/1024/1024
from dba_temp_files;
</syntaxhighlight>

'''Tempfile hinzufügen:'''

<syntaxhighlight lang="sql">
alter tablespace <TABLESPACE> add tempfile '+DATA' size 25G;
</syntaxhighlight>

====Tablespace mit neuer Datei erweitern====
alter tablespace PSAPSR3740X add datafile '/oracle/PZ0/sapdata1/sr3740x_2' size 30G;
====Backupmodus einstellen====
Wird benötigt um ggf. die sapdata online kopieren zu können
<syntaxhighlight lang="bash" line="1">
SQL> ALTER DATABASE BEGIN BACKUP;
To find if database or any tablespace is in backup mode, the status in V$BACKUP is ACTIVE
SQL> select * from v$backup;
SQL> ALTER DATABASE END BACKUP;
</syntaxhighlight>
==== SYSTEM Passwort ändern ====
sqlplus / as sysdba
ALTER USER system IDENTIFIED BY "<PASSWORT>";
=== PSAPUNDO - Undo Tablespace ===
<syntaxhighlight lang="sql" line="1">
SELECT file_name,
tablespace_name,
autoextensible,
bytes/1024/1024 AS size_mb,
maxbytes/1024/1024 AS max_size_mb
FROM dba_data_files
WHERE tablespace_name = 'PSAPUNDO';

ALTER DATABASE DATAFILE '/oracle/R08/sapdata1/undo_1/undo.data1' AUTOEXTEND ON NEXT 100M MAXSIZE 20G;

</syntaxhighlight>

=== Update ===

=== Test ===
==== Freien Speicherplatz anzeigen ====
<syntaxhighlight lang="bash" line="1">
select a.tablespace_name, a.free_bytes*100/b.total_bytes as "Free(%)"
from
(select tablespace_name, sum(bytes) free_bytes
from dba_free_space
group by tablespace_name) a,
(select tablespace_name, sum(bytes) total_bytes
from dba_data_files
group by tablespace_name) b
where a.tablespace_name=b.tablespace_name;
</syntaxhighlight>
=== Codeschnipsel ===

=== Nützliche Links ===

Oracle

2026-01-06T09:48:13Z

Hendrik: /* SAP* kopieren (000 nach 200) */

=== Beschreibung ===

=== Download ===

=== Installation ===

=== Konfiguration ===
==== SAP* löschen ====
sqlplus / as sysdba
<syntaxhighlight lang="sql" line="1">
DELETE FROM SAPSR3.USR02 WHERE MANDT = '000' AND BNAME = 'SAP*';
COMMIT;
</syntaxhighlight>
==== SAP* kopieren (000 nach 200) ====
<syntaxhighlight lang="oracle11" line="1">
INSERT INTO SAPSR3.USR02 (
MANDT, BNAME, BCODE, GLTGV, GLTGB, USTYP,
CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT,
TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2,
BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4,
BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN,
CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE,
RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE,
PWDSALTEDHASH, SECURITY_POLICY
)
SELECT
'200', BNAME, BCODE, GLTGV, GLTGB, USTYP,
CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT,
TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2,
BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4,
BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN,
CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE,
RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE,
PWDSALTEDHASH, SECURITY_POLICY
FROM SAPSR3.USR02
WHERE MANDT = '000' AND BNAME = 'SAP*';
</syntaxhighlight>
COMMIT;

==== Tablespaces und Files anzeigen lassen ====
<syntaxhighlight lang="sql" line="1">
SELECT file_name, bytes/1024/1024 AS size_mb FROM dba_data_files WHERE tablespace_name = 'PSAPSR3' ORDER BY file_name;
</syntaxhighlight>
====Tablespace anlegen====
CREATE TABLESPACE PSAPSR3740X DATAFILE '/oracle/PZ0/sapdata1/sr3740x_1' SIZE 10M;

= Erweiterung ohne ASM =

== Tablespace-Belegung ermitteln ==

Belegung aller Tablespaces in Prozent ermitteln:

<syntaxhighlight lang="sql">
with MBYTES as(
select tablespace_name,
SUM(bytes)/1024/1024 as Akt,
SUM(decode(maxbytes,0,bytes,maxbytes))/1024/1024 as ZMax
from dba_data_files
group by tablespace_name
),
MFREE as(
select tablespace_name,
SUM(bytes)/1024/1024 as Free
from dba_free_space
group by tablespace_name
)
select MBYTES.tablespace_name,
ROUND((AKT-FREE)/ZMAX*100,2) as BelegtProz
from MBYTES, MFREE
where MFREE.tablespace_name = MBYTES.tablespace_name
order by ROUND((AKT-FREE)/ZMAX*100,2) desc;
</syntaxhighlight>

'''Beispiel-Ergebnis:'''
<pre>
TABLESPACE_NAME BELEGTPROZ
------------------------------ ----------
PSAPSR3 89.43
PSAPSR3702 33.32
PSAPSR3USR 20.26
PSAPUNDO 6.55
SYSTEM 5.05
SYSAUX 4.05
</pre>

== Verfügbaren Speicherplatz ermitteln ==

Im Betriebssystem ermitteln, in welchem data-Laufwerk am meisten Platz für die Erweiterung vorhanden ist:

<syntaxhighlight lang="bash">
df -h | grep oracle
</syntaxhighlight>

'''Beispiel-Ausgabe:'''
<pre>
p11data01/data01 98G 68G 30G 70% /oracle/P11/data01
p11data02/data02 215G 172G 43G 81% /oracle/P11/data02
p11data03/data03 98G 51G 47G 52% /oracle/P11/data03
</pre>

== An der Datenbank anmelden ==

<syntaxhighlight lang="sql">
sqlplus / as sysdba
</syntaxhighlight>

=== Ausgabe formatieren ===

Für eine übersichtlichere Darstellung:

<syntaxhighlight lang="sql">
set pages 9999
set lin 180
col file_name for a60
</syntaxhighlight>

== Datafiles anzeigen ==

Datafiles zum Tablespace anzeigen:

<syntaxhighlight lang="sql">
select file_name, bytes/1024/1024
from dba_data_files
where tablespace_name = '<TABLESPACE>';
</syntaxhighlight>

{{Hinweis|Der Name des Tablespaces ist in der E-Mail von Icinga zu finden.}}

'''Beispiel-Ausgabe:'''
<pre>
FILE_NAME BYTES/1024/1024
------------------------------------------------------------ ---------------
/oracle/P11/data01/beispiel_data_01.dbf 20480
/oracle/P11/data02/beispiel_data_02.dbf 20480
/oracle/P11/data03/beispiel_data_03.dbf 30720
/oracle/P11/data02/beispiel_data_04.dbf 20480
</pre>

== Datafile erweitern ==

=== Existierendes Datafile vergrößern ===

Wenn ein Datafile erweitert werden kann:

<syntaxhighlight lang="sql">
alter database datafile '/oracle/P11/data02/beispiel_04.dbf' resize 30G;
</syntaxhighlight>

{{Hinweis|Der Prozess dauert ca. 1 GB pro Minute.}}

'''Überwachung auf OS-Ebene:'''

<syntaxhighlight lang="bash">
du -h <FILENAME>
</syntaxhighlight>

'''Kontrolle:'''

<syntaxhighlight lang="sql">
select file_name, bytes/1024/1024
from dba_data_files
where tablespace_name = '<TABLESPACE>';
</syntaxhighlight>

'''Beispiel nach Erweiterung:'''
<pre>
FILE_NAME BYTES/1024/1024
------------------------------------------------------------ ---------------
/oracle/P11/data01/beispiel_data_01.dbf 20480
/oracle/P11/data02/beispiel_data_02.dbf 20480
/oracle/P11/data03/beispiel_data_03.dbf 30720
/oracle/P11/data02/beispiel_data_04.dbf 30720
</pre>

=== Neues Datafile anlegen ===

Falls kein existierendes Datafile erweitert werden kann:

<syntaxhighlight lang="sql">
alter tablespace BEISPIEL add datafile '/oracle/P11/data03/beispiel.dbf' size 100M;
</syntaxhighlight>

{{Hinweis|Der Prozess dauert ca. 1 GB pro Minute.}}

'''Überwachung auf OS-Ebene:'''

<syntaxhighlight lang="bash">
du -h <FILENAME>
</syntaxhighlight>

'''Kontrolle:'''

<syntaxhighlight lang="sql">
select file_name, bytes/1024/1024
from dba_data_files
where tablespace_name = '<TABLESPACE>';
</syntaxhighlight>

'''Beispiel nach Anlegen:'''
<pre>
FILE_NAME BYTES/1024/1024
------------------------------------------------------------ ---------------
/oracle/P11/data01/beispiel_data_01.dbf 20480
/oracle/P11/data02/beispiel_data_02.dbf 20480
/oracle/P11/data03/beispiel_data_03.dbf 30720
/oracle/P11/data02/beispiel_data_04.dbf 20480
/oracle/P11/data03/beispiel_data_05.dbf 100
</pre>

== Tempfile erweitern ==

'''Tempfiles anzeigen:'''

<syntaxhighlight lang="sql">
select file_name, bytes/1024/1024
from dba_temp_files;
</syntaxhighlight>

'''Tempfile hinzufügen:'''

<syntaxhighlight lang="sql">
alter tablespace <TABLESPACE> add tempfile '+DATA' size 25G;
</syntaxhighlight>

====Tablespace mit neuer Datei erweitern====
alter tablespace PSAPSR3740X add datafile '/oracle/PZ0/sapdata1/sr3740x_2' size 30G;
====Backupmodus einstellen====
Wird benötigt um ggf. die sapdata online kopieren zu können
<syntaxhighlight lang="bash" line="1">
SQL> ALTER DATABASE BEGIN BACKUP;
To find if database or any tablespace is in backup mode, the status in V$BACKUP is ACTIVE
SQL> select * from v$backup;
SQL> ALTER DATABASE END BACKUP;
</syntaxhighlight>
==== SYSTEM Passwort ändern ====
sqlplus / as sysdba
ALTER USER system IDENTIFIED BY "<PASSWORT>";
=== PSAPUNDO - Undo Tablespace ===
<syntaxhighlight lang="sql" line="1">
SELECT file_name,
tablespace_name,
autoextensible,
bytes/1024/1024 AS size_mb,
maxbytes/1024/1024 AS max_size_mb
FROM dba_data_files
WHERE tablespace_name = 'PSAPUNDO';

ALTER DATABASE DATAFILE '/oracle/R08/sapdata1/undo_1/undo.data1' AUTOEXTEND ON NEXT 100M MAXSIZE 20G;

</syntaxhighlight>

=== Update ===

=== Test ===
==== Freien Speicherplatz anzeigen ====
<syntaxhighlight lang="bash" line="1">
select a.tablespace_name, a.free_bytes*100/b.total_bytes as "Free(%)"
from
(select tablespace_name, sum(bytes) free_bytes
from dba_free_space
group by tablespace_name) a,
(select tablespace_name, sum(bytes) total_bytes
from dba_data_files
group by tablespace_name) b
where a.tablespace_name=b.tablespace_name;
</syntaxhighlight>
=== Codeschnipsel ===

=== Nützliche Links ===

Oracle

2026-01-06T09:47:47Z

Hendrik: /* SAP* löschen */

=== Beschreibung ===

=== Download ===

=== Installation ===

=== Konfiguration ===
==== SAP* löschen ====
sqlplus / as sysdba
<syntaxhighlight lang="sql" line="1">
DELETE FROM SAPSR3.USR02 WHERE MANDT = '000' AND BNAME = 'SAP*';
COMMIT;
</syntaxhighlight>
==== SAP* kopieren (000 nach 200) ====
<syntaxhighlight lang="sql" line="1">
INSERT INTO SAPSR3.USR02 (
MANDT, BNAME, BCODE, GLTGV, GLTGB, USTYP,
CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT,
TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2,
BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4,
BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN,
CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE,
RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE,
PWDSALTEDHASH, SECURITY_POLICY
)
SELECT
'200', BNAME, BCODE, GLTGV, GLTGB, USTYP,
CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT,
TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2,
BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4,
BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN,
CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE,
RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE,
PWDSALTEDHASH, SECURITY_POLICY
FROM SAPSR3.USR02
WHERE MANDT = '000' AND BNAME = 'SAP*';
</syntaxhighlight>
COMMIT;

==== Tablespaces und Files anzeigen lassen ====
<syntaxhighlight lang="sql" line="1">
SELECT file_name, bytes/1024/1024 AS size_mb FROM dba_data_files WHERE tablespace_name = 'PSAPSR3' ORDER BY file_name;
</syntaxhighlight>
====Tablespace anlegen====
CREATE TABLESPACE PSAPSR3740X DATAFILE '/oracle/PZ0/sapdata1/sr3740x_1' SIZE 10M;

= Erweiterung ohne ASM =

== Tablespace-Belegung ermitteln ==

Belegung aller Tablespaces in Prozent ermitteln:

<syntaxhighlight lang="sql">
with MBYTES as(
select tablespace_name,
SUM(bytes)/1024/1024 as Akt,
SUM(decode(maxbytes,0,bytes,maxbytes))/1024/1024 as ZMax
from dba_data_files
group by tablespace_name
),
MFREE as(
select tablespace_name,
SUM(bytes)/1024/1024 as Free
from dba_free_space
group by tablespace_name
)
select MBYTES.tablespace_name,
ROUND((AKT-FREE)/ZMAX*100,2) as BelegtProz
from MBYTES, MFREE
where MFREE.tablespace_name = MBYTES.tablespace_name
order by ROUND((AKT-FREE)/ZMAX*100,2) desc;
</syntaxhighlight>

'''Beispiel-Ergebnis:'''
<pre>
TABLESPACE_NAME BELEGTPROZ
------------------------------ ----------
PSAPSR3 89.43
PSAPSR3702 33.32
PSAPSR3USR 20.26
PSAPUNDO 6.55
SYSTEM 5.05
SYSAUX 4.05
</pre>

== Verfügbaren Speicherplatz ermitteln ==

Im Betriebssystem ermitteln, in welchem data-Laufwerk am meisten Platz für die Erweiterung vorhanden ist:

<syntaxhighlight lang="bash">
df -h | grep oracle
</syntaxhighlight>

'''Beispiel-Ausgabe:'''
<pre>
p11data01/data01 98G 68G 30G 70% /oracle/P11/data01
p11data02/data02 215G 172G 43G 81% /oracle/P11/data02
p11data03/data03 98G 51G 47G 52% /oracle/P11/data03
</pre>

== An der Datenbank anmelden ==

<syntaxhighlight lang="sql">
sqlplus / as sysdba
</syntaxhighlight>

=== Ausgabe formatieren ===

Für eine übersichtlichere Darstellung:

<syntaxhighlight lang="sql">
set pages 9999
set lin 180
col file_name for a60
</syntaxhighlight>

== Datafiles anzeigen ==

Datafiles zum Tablespace anzeigen:

<syntaxhighlight lang="sql">
select file_name, bytes/1024/1024
from dba_data_files
where tablespace_name = '<TABLESPACE>';
</syntaxhighlight>

{{Hinweis|Der Name des Tablespaces ist in der E-Mail von Icinga zu finden.}}

'''Beispiel-Ausgabe:'''
<pre>
FILE_NAME BYTES/1024/1024
------------------------------------------------------------ ---------------
/oracle/P11/data01/beispiel_data_01.dbf 20480
/oracle/P11/data02/beispiel_data_02.dbf 20480
/oracle/P11/data03/beispiel_data_03.dbf 30720
/oracle/P11/data02/beispiel_data_04.dbf 20480
</pre>

== Datafile erweitern ==

=== Existierendes Datafile vergrößern ===

Wenn ein Datafile erweitert werden kann:

<syntaxhighlight lang="sql">
alter database datafile '/oracle/P11/data02/beispiel_04.dbf' resize 30G;
</syntaxhighlight>

{{Hinweis|Der Prozess dauert ca. 1 GB pro Minute.}}

'''Überwachung auf OS-Ebene:'''

<syntaxhighlight lang="bash">
du -h <FILENAME>
</syntaxhighlight>

'''Kontrolle:'''

<syntaxhighlight lang="sql">
select file_name, bytes/1024/1024
from dba_data_files
where tablespace_name = '<TABLESPACE>';
</syntaxhighlight>

'''Beispiel nach Erweiterung:'''
<pre>
FILE_NAME BYTES/1024/1024
------------------------------------------------------------ ---------------
/oracle/P11/data01/beispiel_data_01.dbf 20480
/oracle/P11/data02/beispiel_data_02.dbf 20480
/oracle/P11/data03/beispiel_data_03.dbf 30720
/oracle/P11/data02/beispiel_data_04.dbf 30720
</pre>

=== Neues Datafile anlegen ===

Falls kein existierendes Datafile erweitert werden kann:

<syntaxhighlight lang="sql">
alter tablespace BEISPIEL add datafile '/oracle/P11/data03/beispiel.dbf' size 100M;
</syntaxhighlight>

{{Hinweis|Der Prozess dauert ca. 1 GB pro Minute.}}

'''Überwachung auf OS-Ebene:'''

<syntaxhighlight lang="bash">
du -h <FILENAME>
</syntaxhighlight>

'''Kontrolle:'''

<syntaxhighlight lang="sql">
select file_name, bytes/1024/1024
from dba_data_files
where tablespace_name = '<TABLESPACE>';
</syntaxhighlight>

'''Beispiel nach Anlegen:'''
<pre>
FILE_NAME BYTES/1024/1024
------------------------------------------------------------ ---------------
/oracle/P11/data01/beispiel_data_01.dbf 20480
/oracle/P11/data02/beispiel_data_02.dbf 20480
/oracle/P11/data03/beispiel_data_03.dbf 30720
/oracle/P11/data02/beispiel_data_04.dbf 20480
/oracle/P11/data03/beispiel_data_05.dbf 100
</pre>

== Tempfile erweitern ==

'''Tempfiles anzeigen:'''

<syntaxhighlight lang="sql">
select file_name, bytes/1024/1024
from dba_temp_files;
</syntaxhighlight>

'''Tempfile hinzufügen:'''

<syntaxhighlight lang="sql">
alter tablespace <TABLESPACE> add tempfile '+DATA' size 25G;
</syntaxhighlight>

====Tablespace mit neuer Datei erweitern====
alter tablespace PSAPSR3740X add datafile '/oracle/PZ0/sapdata1/sr3740x_2' size 30G;
====Backupmodus einstellen====
Wird benötigt um ggf. die sapdata online kopieren zu können
<syntaxhighlight lang="bash" line="1">
SQL> ALTER DATABASE BEGIN BACKUP;
To find if database or any tablespace is in backup mode, the status in V$BACKUP is ACTIVE
SQL> select * from v$backup;
SQL> ALTER DATABASE END BACKUP;
</syntaxhighlight>
==== SYSTEM Passwort ändern ====
sqlplus / as sysdba
ALTER USER system IDENTIFIED BY "<PASSWORT>";
=== PSAPUNDO - Undo Tablespace ===
<syntaxhighlight lang="sql" line="1">
SELECT file_name,
tablespace_name,
autoextensible,
bytes/1024/1024 AS size_mb,
maxbytes/1024/1024 AS max_size_mb
FROM dba_data_files
WHERE tablespace_name = 'PSAPUNDO';

ALTER DATABASE DATAFILE '/oracle/R08/sapdata1/undo_1/undo.data1' AUTOEXTEND ON NEXT 100M MAXSIZE 20G;

</syntaxhighlight>

=== Update ===

=== Test ===
==== Freien Speicherplatz anzeigen ====
<syntaxhighlight lang="bash" line="1">
select a.tablespace_name, a.free_bytes*100/b.total_bytes as "Free(%)"
from
(select tablespace_name, sum(bytes) free_bytes
from dba_free_space
group by tablespace_name) a,
(select tablespace_name, sum(bytes) total_bytes
from dba_data_files
group by tablespace_name) b
where a.tablespace_name=b.tablespace_name;
</syntaxhighlight>
=== Codeschnipsel ===

=== Nützliche Links ===

Oracle

2026-01-06T09:47:40Z

Hendrik: /* Konfiguration */

=== Beschreibung ===

=== Download ===

=== Installation ===

=== Konfiguration ===
==== SAP* löschen ====
sqlplus / as sysdba
<syntaxhighlight lang="sql" line="1">
DELETE FROM SAPSR3.USR02 WHERE MANDT = '000' AND BNAME = 'SAP*';
COMMIT;
</syntaxhighlight>
<syntaxhighlight lang="sql" line="1">
==== SAP* kopieren (000 nach 200) ====
INSERT INTO SAPSR3.USR02 (
MANDT, BNAME, BCODE, GLTGV, GLTGB, USTYP,
CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT,
TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2,
BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4,
BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN,
CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE,
RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE,
PWDSALTEDHASH, SECURITY_POLICY
)
SELECT
'200', BNAME, BCODE, GLTGV, GLTGB, USTYP,
CLASS, LOCNT, UFLAG, ACCNT, ANAME, ERDAT,
TRDAT, LTIME, OCOD1, BCDA1, CODV1, OCOD2,
BCDA2, CODV2, OCOD3, BCDA3, CODV3, OCOD4,
BCDA4, CODV4, OCOD5, BCDA5, CODV5, VERSN,
CODVN, TZONE, ZBVMASTER, PASSCODE, PWDCHGDATE, PWDSTATE,
RESERVED, PWDHISTORY, PWDLGNDATE, PWDSETDATE, PWDINITIAL, PWDLOCKDATE,
PWDSALTEDHASH, SECURITY_POLICY
FROM SAPSR3.USR02
WHERE MANDT = '000' AND BNAME = 'SAP*';
</syntaxhighlight>
COMMIT;
==== Tablespaces und Files anzeigen lassen ====
<syntaxhighlight lang="sql" line="1">
SELECT file_name, bytes/1024/1024 AS size_mb FROM dba_data_files WHERE tablespace_name = 'PSAPSR3' ORDER BY file_name;
</syntaxhighlight>
====Tablespace anlegen====
CREATE TABLESPACE PSAPSR3740X DATAFILE '/oracle/PZ0/sapdata1/sr3740x_1' SIZE 10M;

= Erweiterung ohne ASM =

== Tablespace-Belegung ermitteln ==

Belegung aller Tablespaces in Prozent ermitteln:

<syntaxhighlight lang="sql">
with MBYTES as(
select tablespace_name,
SUM(bytes)/1024/1024 as Akt,
SUM(decode(maxbytes,0,bytes,maxbytes))/1024/1024 as ZMax
from dba_data_files
group by tablespace_name
),
MFREE as(
select tablespace_name,
SUM(bytes)/1024/1024 as Free
from dba_free_space
group by tablespace_name
)
select MBYTES.tablespace_name,
ROUND((AKT-FREE)/ZMAX*100,2) as BelegtProz
from MBYTES, MFREE
where MFREE.tablespace_name = MBYTES.tablespace_name
order by ROUND((AKT-FREE)/ZMAX*100,2) desc;
</syntaxhighlight>

'''Beispiel-Ergebnis:'''
<pre>
TABLESPACE_NAME BELEGTPROZ
------------------------------ ----------
PSAPSR3 89.43
PSAPSR3702 33.32
PSAPSR3USR 20.26
PSAPUNDO 6.55
SYSTEM 5.05
SYSAUX 4.05
</pre>

== Verfügbaren Speicherplatz ermitteln ==

Im Betriebssystem ermitteln, in welchem data-Laufwerk am meisten Platz für die Erweiterung vorhanden ist:

<syntaxhighlight lang="bash">
df -h | grep oracle
</syntaxhighlight>

'''Beispiel-Ausgabe:'''
<pre>
p11data01/data01 98G 68G 30G 70% /oracle/P11/data01
p11data02/data02 215G 172G 43G 81% /oracle/P11/data02
p11data03/data03 98G 51G 47G 52% /oracle/P11/data03
</pre>

== An der Datenbank anmelden ==

<syntaxhighlight lang="sql">
sqlplus / as sysdba
</syntaxhighlight>

=== Ausgabe formatieren ===

Für eine übersichtlichere Darstellung:

<syntaxhighlight lang="sql">
set pages 9999
set lin 180
col file_name for a60
</syntaxhighlight>

== Datafiles anzeigen ==

Datafiles zum Tablespace anzeigen:

<syntaxhighlight lang="sql">
select file_name, bytes/1024/1024
from dba_data_files
where tablespace_name = '<TABLESPACE>';
</syntaxhighlight>

{{Hinweis|Der Name des Tablespaces ist in der E-Mail von Icinga zu finden.}}

'''Beispiel-Ausgabe:'''
<pre>
FILE_NAME BYTES/1024/1024
------------------------------------------------------------ ---------------
/oracle/P11/data01/beispiel_data_01.dbf 20480
/oracle/P11/data02/beispiel_data_02.dbf 20480
/oracle/P11/data03/beispiel_data_03.dbf 30720
/oracle/P11/data02/beispiel_data_04.dbf 20480
</pre>

== Datafile erweitern ==

=== Existierendes Datafile vergrößern ===

Wenn ein Datafile erweitert werden kann:

<syntaxhighlight lang="sql">
alter database datafile '/oracle/P11/data02/beispiel_04.dbf' resize 30G;
</syntaxhighlight>

{{Hinweis|Der Prozess dauert ca. 1 GB pro Minute.}}

'''Überwachung auf OS-Ebene:'''

<syntaxhighlight lang="bash">
du -h <FILENAME>
</syntaxhighlight>

'''Kontrolle:'''

<syntaxhighlight lang="sql">
select file_name, bytes/1024/1024
from dba_data_files
where tablespace_name = '<TABLESPACE>';
</syntaxhighlight>

'''Beispiel nach Erweiterung:'''
<pre>
FILE_NAME BYTES/1024/1024
------------------------------------------------------------ ---------------
/oracle/P11/data01/beispiel_data_01.dbf 20480
/oracle/P11/data02/beispiel_data_02.dbf 20480
/oracle/P11/data03/beispiel_data_03.dbf 30720
/oracle/P11/data02/beispiel_data_04.dbf 30720
</pre>

=== Neues Datafile anlegen ===

Falls kein existierendes Datafile erweitert werden kann:

<syntaxhighlight lang="sql">
alter tablespace BEISPIEL add datafile '/oracle/P11/data03/beispiel.dbf' size 100M;
</syntaxhighlight>

{{Hinweis|Der Prozess dauert ca. 1 GB pro Minute.}}

'''Überwachung auf OS-Ebene:'''

<syntaxhighlight lang="bash">
du -h <FILENAME>
</syntaxhighlight>

'''Kontrolle:'''

<syntaxhighlight lang="sql">
select file_name, bytes/1024/1024
from dba_data_files
where tablespace_name = '<TABLESPACE>';
</syntaxhighlight>

'''Beispiel nach Anlegen:'''
<pre>
FILE_NAME BYTES/1024/1024
------------------------------------------------------------ ---------------
/oracle/P11/data01/beispiel_data_01.dbf 20480
/oracle/P11/data02/beispiel_data_02.dbf 20480
/oracle/P11/data03/beispiel_data_03.dbf 30720
/oracle/P11/data02/beispiel_data_04.dbf 20480
/oracle/P11/data03/beispiel_data_05.dbf 100
</pre>

== Tempfile erweitern ==

'''Tempfiles anzeigen:'''

<syntaxhighlight lang="sql">
select file_name, bytes/1024/1024
from dba_temp_files;
</syntaxhighlight>

'''Tempfile hinzufügen:'''

<syntaxhighlight lang="sql">
alter tablespace <TABLESPACE> add tempfile '+DATA' size 25G;
</syntaxhighlight>

====Tablespace mit neuer Datei erweitern====
alter tablespace PSAPSR3740X add datafile '/oracle/PZ0/sapdata1/sr3740x_2' size 30G;
====Backupmodus einstellen====
Wird benötigt um ggf. die sapdata online kopieren zu können
<syntaxhighlight lang="bash" line="1">
SQL> ALTER DATABASE BEGIN BACKUP;
To find if database or any tablespace is in backup mode, the status in V$BACKUP is ACTIVE
SQL> select * from v$backup;
SQL> ALTER DATABASE END BACKUP;
</syntaxhighlight>
==== SYSTEM Passwort ändern ====
sqlplus / as sysdba
ALTER USER system IDENTIFIED BY "<PASSWORT>";
=== PSAPUNDO - Undo Tablespace ===
<syntaxhighlight lang="sql" line="1">
SELECT file_name,
tablespace_name,
autoextensible,
bytes/1024/1024 AS size_mb,
maxbytes/1024/1024 AS max_size_mb
FROM dba_data_files
WHERE tablespace_name = 'PSAPUNDO';

ALTER DATABASE DATAFILE '/oracle/R08/sapdata1/undo_1/undo.data1' AUTOEXTEND ON NEXT 100M MAXSIZE 20G;

</syntaxhighlight>

=== Update ===

=== Test ===
==== Freien Speicherplatz anzeigen ====
<syntaxhighlight lang="bash" line="1">
select a.tablespace_name, a.free_bytes*100/b.total_bytes as "Free(%)"
from
(select tablespace_name, sum(bytes) free_bytes
from dba_free_space
group by tablespace_name) a,
(select tablespace_name, sum(bytes) total_bytes
from dba_data_files
group by tablespace_name) b
where a.tablespace_name=b.tablespace_name;
</syntaxhighlight>
=== Codeschnipsel ===

=== Nützliche Links ===

Wireguard einrichten

2025-12-29T12:39:47Z

Hendrik: /* Wireguard Client */

=== Beschreibung ===

=== Download ===

=== Installation ===
apt install wireguard -y

=== Konfiguration ===
====Server====
Keys erzeugen
<syntaxhighlight lang="bash" line="1">
wg genkey | sudo tee /etc/wireguard/priv.key
chmod go= /etc/wireguard/priv.key

cat /etc/wireguard/priv.key | wg pubkey | sudo tee >> /etc/wireguard/pub.key
nano /etc/wireguard/wg0.conf
</syntaxhighlight>
Inhalt:
<syntaxhighlight lang="bash" line="1">
[Interface]
Address = 10.0.5.1/24
ListenPort = 51820
# Use your own private key, from /etc/wireguard/privatekey
PrivateKey = XXX

#Routen einstellen damit Clients kommunizieren können
PostUp = iptables -t nat -A POSTROUTING -o eth0 -j MASQUERADE; iptables -A FORWARD -i wg0 -o wg0 -j ACCEPT
PostDown = iptables -t nat -D POSTROUTING -o eth0 -j MASQUERADE

[Peer]
# Client 1
PublicKey = XXX
# VPN client's IP address in the VPN
AllowedIPs = 10.0.5.2/32

</syntaxhighlight>
<syntaxhighlight lang="bash" line="1">
echo "net.ipv4.ip_forward=1" >> /etc/sysctl.conf && sysctl -p && sysctl -w net.ipv4.ip_forward=1
</syntaxhighlight>
wg-quick up wg0

=====Wireguard Client=====
<syntaxhighlight lang="bash" line="1">
apt update && apt install wireguard resolvconf -y
cd /etc/wireguard
umask 077
wg genkey | tee privatekey | wg pubkey > publickey
cat publickey
</syntaxhighlight>
Auf der OPNSense Firewall jetzt den Peer erzeugen:
1. Gehe zu VPN -> WireGuard -> Peers.
2. Erstelle einen neuen Peer für den ersten VPS (z.B. "VPS_195"):
* Enabled: Haken setzen.
* Public Key: Hier den Inhalt der publickey Datei vom VPS eintragen.
* Allowed IPs: 10.0.0.2/32 (WICHTIG: Hier nur die eine IP des VPS eintragen, mit /32. Das sagt der OPNsense: "Pakete für 10.0.0.2 schicke ich an diesen Tunnel").
* Keepalive: 25 (Sinnvoll, um die Verbindung durch NAT/Firewalls offen zu halten).
<syntaxhighlight lang="bash" line="1">
cat privatekey
nano /etc/wireguard/wg0.conf
Beispiel für VPS 1 (IP 10.0.0.2):
[Interface]
Address = 10.0.0.2/24
PrivateKey = <HIER_DEN_PRIVATE_KEY_VOM_VPS_EINFÜGEN>
# Optional: Wenn der VPS auch DNS von Zuhause nutzen soll
# DNS = 192.168.1.1

[Peer]
PublicKey = <HIER_DEN_PUBLIC_KEY_DER_OPNSENSE_EINFÜGEN>
# Der Endpoint muss per IPv6 erreichbar sein
Endpoint = xsarts.de:51820
# Welche IPs sollen durch den Tunnel gehen?
# 10.0.0.0/24 erlaubt Zugriff auf das Tunnel-Netzwerk.
# Wenn der VPS auch dein LAN (192.168.1.x) erreichen soll, füge es hinzu:
AllowedIPs = 10.0.0.0/24, 192.168.1.0/24
PersistentKeepalive = 25

#WoW Portweiterleitung ins HomeLab
# ---------- PostUp ----------
PostUp = sysctl -w net.ipv4.ip_forward=1
PostUp = iptables -t nat -A PREROUTING -p tcp --dport 8085 -j DNAT --to-destination 192.168.1.17:8085
PostUp = iptables -t nat -A PREROUTING -p tcp --dport 3724 -j DNAT --to-destination 192.168.1.17:3724
PostUp = iptables -t nat -A POSTROUTING -o %i -j MASQUERADE
PostUp = iptables -A FORWARD -i eth0 -o %i -p tcp -m multiport --dports 8085,3724 -m conntrack --ctstate NEW -j ACCEPT
PostUp = iptables -A FORWARD -m conntrack --ctstate ESTABLISHED,RELATED -j ACCEPT

# ---------- PostDown ----------
PostDown = iptables -t nat -D PREROUTING -p tcp --dport 8085 -j DNAT --to-destination 192.168.1.17:8085
PostDown = iptables -t nat -D PREROUTING -p tcp --dport 3724 -j DNAT --to-destination 192.168.1.17:3724
PostDown = iptables -t nat -D POSTROUTING -o %i -j MASQUERADE
PostDown = iptables -D FORWARD -i eth0 -o %i -p tcp -m multiport --dports 8085,3724 -m conntrack --ctstate NEW -j ACCEPT
PostDown = iptables -D FORWARD -m conntrack --ctstate ESTABLISHED,RELATED -j ACCEPT
PostDown = sysctl -w net.ipv4.ip_forward=0

[Peer]
PublicKey =
Endpoint = xsarts.de:51820
AllowedIPs = 192.168.1.0/24
</syntaxhighlight>
Wireguard starten und testen:
<syntaxhighlight lang="bash" line="1">
systemctl enable wg-quick@wg0 && systemctl start wg-quick@wg0 && ping 10.0.0.1
</syntaxhighlight>

=== Update ===

=== Test ===

=== Fehlerbehebung===
====Starten/Stoppen funktioniert nicht====
=====Lösung 1=====
<syntaxhighlight lang="bash" line="1">
auszufüllen
</syntaxhighlight>

=== Codeschnipsel ===

=== Nützliche Links ===

=== Codeschnipsel ===

=== Nützliche Links ===

Wireguard einrichten

2025-12-29T12:21:26Z

Hendrik: /* Wireguard Client */

=== Beschreibung ===

=== Download ===

=== Installation ===
apt install wireguard -y

=== Konfiguration ===
====Server====
Keys erzeugen
<syntaxhighlight lang="bash" line="1">
wg genkey | sudo tee /etc/wireguard/priv.key
chmod go= /etc/wireguard/priv.key

cat /etc/wireguard/priv.key | wg pubkey | sudo tee >> /etc/wireguard/pub.key
nano /etc/wireguard/wg0.conf
</syntaxhighlight>
Inhalt:
<syntaxhighlight lang="bash" line="1">
[Interface]
Address = 10.0.5.1/24
ListenPort = 51820
# Use your own private key, from /etc/wireguard/privatekey
PrivateKey = XXX

#Routen einstellen damit Clients kommunizieren können
PostUp = iptables -t nat -A POSTROUTING -o eth0 -j MASQUERADE; iptables -A FORWARD -i wg0 -o wg0 -j ACCEPT
PostDown = iptables -t nat -D POSTROUTING -o eth0 -j MASQUERADE

[Peer]
# Client 1
PublicKey = XXX
# VPN client's IP address in the VPN
AllowedIPs = 10.0.5.2/32

</syntaxhighlight>
<syntaxhighlight lang="bash" line="1">
echo "net.ipv4.ip_forward=1" >> /etc/sysctl.conf && sysctl -p && sysctl -w net.ipv4.ip_forward=1
</syntaxhighlight>
wg-quick up wg0

=====Wireguard Client=====
<syntaxhighlight lang="bash" line="1">
apt update && apt install wireguard resolvconf -y
cd /etc/wireguard
umask 077
cat publickey
</syntaxhighlight>
Auf der OPNSense Firewall jetzt den Peer erzeugen:
1. Gehe zu VPN -> WireGuard -> Peers.
2. Erstelle einen neuen Peer für den ersten VPS (z.B. "VPS_195"):
* Enabled: Haken setzen.
* Public Key: Hier den Inhalt der publickey Datei vom VPS eintragen.
* Allowed IPs: 10.0.0.2/32 (WICHTIG: Hier nur die eine IP des VPS eintragen, mit /32. Das sagt der OPNsense: "Pakete für 10.0.0.2 schicke ich an diesen Tunnel").
* Keepalive: 25 (Sinnvoll, um die Verbindung durch NAT/Firewalls offen zu halten).
<syntaxhighlight lang="bash" line="1">
cat privatekey
nano /etc/wireguard/wg0.conf
Beispiel für VPS 1 (IP 10.0.0.2):
[Interface]
Address = 10.0.0.2/24
PrivateKey = <HIER_DEN_PRIVATE_KEY_VOM_VPS_EINFÜGEN>
# Optional: Wenn der VPS auch DNS von Zuhause nutzen soll
# DNS = 192.168.1.1

[Peer]
PublicKey = <HIER_DEN_PUBLIC_KEY_DER_OPNSENSE_EINFÜGEN>
# Der Endpoint muss per IPv6 erreichbar sein
Endpoint = xsarts.de:51820
# Welche IPs sollen durch den Tunnel gehen?
# 10.0.0.0/24 erlaubt Zugriff auf das Tunnel-Netzwerk.
# Wenn der VPS auch dein LAN (192.168.1.x) erreichen soll, füge es hinzu:
AllowedIPs = 10.0.0.0/24, 192.168.1.0/24
PersistentKeepalive = 25

#WoW Portweiterleitung ins HomeLab
# ---------- PostUp ----------
PostUp = sysctl -w net.ipv4.ip_forward=1
PostUp = iptables -t nat -A PREROUTING -p tcp --dport 8085 -j DNAT --to-destination 192.168.1.17:8085
PostUp = iptables -t nat -A PREROUTING -p tcp --dport 3724 -j DNAT --to-destination 192.168.1.17:3724
PostUp = iptables -t nat -A POSTROUTING -o %i -j MASQUERADE
PostUp = iptables -A FORWARD -i eth0 -o %i -p tcp -m multiport --dports 8085,3724 -m conntrack --ctstate NEW -j ACCEPT
PostUp = iptables -A FORWARD -m conntrack --ctstate ESTABLISHED,RELATED -j ACCEPT

# ---------- PostDown ----------
PostDown = iptables -t nat -D PREROUTING -p tcp --dport 8085 -j DNAT --to-destination 192.168.1.17:8085
PostDown = iptables -t nat -D PREROUTING -p tcp --dport 3724 -j DNAT --to-destination 192.168.1.17:3724
PostDown = iptables -t nat -D POSTROUTING -o %i -j MASQUERADE
PostDown = iptables -D FORWARD -i eth0 -o %i -p tcp -m multiport --dports 8085,3724 -m conntrack --ctstate NEW -j ACCEPT
PostDown = iptables -D FORWARD -m conntrack --ctstate ESTABLISHED,RELATED -j ACCEPT
PostDown = sysctl -w net.ipv4.ip_forward=0

[Peer]
PublicKey =
Endpoint = xsarts.de:51820
AllowedIPs = 192.168.1.0/24
</syntaxhighlight>
Wireguard starten und testen:
<syntaxhighlight lang="bash" line="1">
systemctl enable wg-quick@wg0 && systemctl start wg-quick@wg0 && ping 10.0.0.1
</syntaxhighlight>

=== Update ===

=== Test ===

=== Fehlerbehebung===
====Starten/Stoppen funktioniert nicht====
=====Lösung 1=====
<syntaxhighlight lang="bash" line="1">
auszufüllen
</syntaxhighlight>

=== Codeschnipsel ===

=== Nützliche Links ===

=== Codeschnipsel ===

=== Nützliche Links ===

Icinga

2025-12-27T09:36:00Z

Hendrik: /* Agent/Client Installation */

=== Beschreibung ===

=== Download ===

=== Installation ===
==== Master Server ====
==== Agent/Client Installation ====
Am Agent/Client:
<syntaxhighlight lang="bash" line="1">
curl -sSL https://packages.icinga.com/icinga.key | apt-key add -
apt-add-repository "deb https://packages.icinga.com/$(lsb_release -cs) icinga-$(lsb_release -cs) main"
apt-get update
apt install monitoring-plugins -y
adduser icinga
sudo sed -i 's/^#\?\s*PasswordAuthentication\s\+no/PasswordAuthentication yes/' /etc/ssh/sshd_config && service ssh restart
</syntaxhighlight>
Vom Icinga-Server aus:
<syntaxhighlight lang="bash" line="1">
sudo -u nagios ssh-copy-id icinga@<remote-server>
</syntaxhighlight>
Falls der Befehl oben nicht verfügbar ist, kopiere den Inhalt von /var/lib/icinga2/.ssh/id_rsa.pub und füge ihn auf dem Remote-Server am Ende der Datei /home/icinga/.ssh/authorized_keys ein.

Am Agent/Client:
<syntaxhighlight lang="bash" line="1">
sudo sed -i 's/^#\?\s*PasswordAuthentication\s\+yes/PasswordAuthentication no/' /etc/ssh/sshd_config && service ssh restart
</syntaxhighlight>
Vom Icinga-Server aus:
<syntaxhighlight lang="bash" line="1">
sudo -u nagios /usr/bin/ssh -i /var/lib/nagios/.ssh/id_rsa icinga@<remote-server> "/usr/lib/nagios/plugins/check_disk -w 20% -c 10% -p /" #testen
</syntaxhighlight>
===== Emailversand =====
Siehe [[Postfix#Via_SMTP_Relay_Host|Postfix]]

==== Agent/Client ====
<syntaxhighlight lang="bash" line="1">
icinga2 node wizard
</syntaxhighlight>
Während der Konfiguration:

*Typ des Setups: Wähle agent.
*Master-Verbindung: Gib die IP/Hostname des Masters ein.
*Port: 5665 (Standardport für den Agent).
*Zertifikatsanforderung: Bestätige, dass der Agent ein Zertifikat vom Master anfordert.
*Zone: Stelle sicher, dass die Zone des Agents korrekt benannt ist.
*Endpunkt: Füge den Endpunkt des Masters hinzu.
Am Master, neuen Client hinzufügen:
<syntaxhighlight lang="bash" line="1">
icinga2 pki ticket --cn '<client hostname>' --salt '<salt>' #die Ausgabe dann beim Client eintragen
icinga2 ca list
</syntaxhighlight>
Client:
<syntaxhighlight lang="bash" line="1">
sudo systemctl restart icinga2
sudo systemctl enable icinga2
</syntaxhighlight>
Füge den Agent in der /etc/icinga2/zones.conf hinzu:
<syntaxhighlight lang="bash" line="1">
object Endpoint "vps-hostname" {
host = "IP-Adresse-des-VPS"
}

object Zone "vps-hostname" {
endpoints = [ "vps-hostname" ]
parent = "icinga"
}
</syntaxhighlight>
Füge den Host in der Datei /etc/icinga2/conf.d/hosts.conf hinzu:
<syntaxhighlight lang="bash" line="1">
object Host "vps-hostname" {
address = "IP-Adresse-des-VPS"
check_command = "hostalive"
vars.os = "Linux"
}
</syntaxhighlight>
Konfiguration neu laden:
<syntaxhighlight lang="bash" line="1">
icinga2 daemon -C && systemctl restart icinga2
</syntaxhighlight>

=== Update ===

=== Test ===

=== Fehlerbehebung===
====Remote command execution failed: @@@@@@====
Bash Script auf Icinga Server erstellen:
<syntaxhighlight lang="bash" line="1">
#!/bin/bash

# Script zum Eintragen eines SSH-Host-Keys für den Nagios-Benutzer
# Verwendung: sudo ./add_known_host.sh npuls.de

HOST="$1"
USER="nagios"
SSH_DIR="/var/lib/$USER/.ssh"
KNOWN_HOSTS="$SSH_DIR/known_hosts"

if [[ -z "$HOST" ]]; then
echo "Verwendung: $0 <hostname oder IP>"
exit 1
fi

# Erstelle das .ssh-Verzeichnis, falls nicht vorhanden
if [[ ! -d "$SSH_DIR" ]]; then
echo "[INFO] Erstelle $SSH_DIR ..."
mkdir -p "$SSH_DIR"
chown "$USER:$USER" "$SSH_DIR"
chmod 700 "$SSH_DIR"
fi

# Stelle sicher, dass known_hosts existiert
if [[ ! -f "$KNOWN_HOSTS" ]]; then
echo "[INFO] Erstelle $KNOWN_HOSTS ..."
touch "$KNOWN_HOSTS"
chown "$USER:$USER" "$KNOWN_HOSTS"
chmod 600 "$KNOWN_HOSTS"
fi

# Füge den Host-Key hinzu, falls noch nicht vorhanden
echo "[INFO] Füge SSH-Host-Key für $HOST hinzu ..."
sudo -u "$USER" ssh-keyscan -H "$HOST" >> "$KNOWN_HOSTS"

echo "[OK] Host $HOST wurde zu $KNOWN_HOSTS hinzugefügt."
</syntaxhighlight>
Ausführen:
<syntaxhighlight lang="bash" line="1">
sudo chmod +x /usr/local/bin/add_known_host.sh && sudo /usr/local/bin/add_known_host.sh <IP oder Domain>
</syntaxhighlight>
====Remote command execution failed: bash: line 1: /usr/lib/nagios/plugins/check_mem.pl: No such file or directory====
<syntaxhighlight lang="bash" line="1">
wget https://raw.githubusercontent.com/justintime/nagios-plugins/master/check_mem/check_mem.pl -O /usr/lib/nagios/plugins/check_mem.pl
chmod +x /usr/lib/nagios/plugins/check_mem.pl
chown root:nagios /usr/lib/nagios/plugins/check_mem.pl
</syntaxhighlight>

=== Codeschnipsel ===

=== Nützliche Links ===

=== Codeschnipsel ===

=== Nützliche Links ===

Icinga

2025-12-27T09:35:41Z

Hendrik: /* Agent/Client Installation */

=== Beschreibung ===

=== Download ===

=== Installation ===
==== Master Server ====
==== Agent/Client Installation ====
Am Agent/Client:
<syntaxhighlight lang="bash" line="1">
curl -sSL https://packages.icinga.com/icinga.key | apt-key add -
apt-add-repository "deb https://packages.icinga.com/$(lsb_release -cs) icinga-$(lsb_release -cs) main"
apt-get update
apt install monitoring-plugins -y
adduser icinga
sudo sed -i 's/^#\?\s*PasswordAuthentication\s\+no/PasswordAuthentication yes/' /etc/ssh/sshd_config && service ssh restart
</syntaxhighlight>
Vom Icinga-Server aus:
<syntaxhighlight lang="bash" line="1">
sudo -u nagios ssh-copy-id icinga@<remote-server>
</syntaxhighlight>
Falls ssh-copy-id nicht geht (Manuelle Methode):

Falls der Befehl oben nicht verfügbar ist, kopiere den Inhalt von /var/lib/icinga2/.ssh/id_rsa.pub und füge ihn auf dem Remote-Server am Ende der Datei /home/icinga/.ssh/authorized_keys ein.

Am Agent/Client:
<syntaxhighlight lang="bash" line="1">
sudo sed -i 's/^#\?\s*PasswordAuthentication\s\+yes/PasswordAuthentication no/' /etc/ssh/sshd_config && service ssh restart
</syntaxhighlight>
Vom Icinga-Server aus:
<syntaxhighlight lang="bash" line="1">
sudo -u nagios /usr/bin/ssh -i /var/lib/nagios/.ssh/id_rsa icinga@<remote-server> "/usr/lib/nagios/plugins/check_disk -w 20% -c 10% -p /" #testen
</syntaxhighlight>
===== Emailversand =====
Siehe [[Postfix#Via_SMTP_Relay_Host|Postfix]]

==== Agent/Client ====
<syntaxhighlight lang="bash" line="1">
icinga2 node wizard
</syntaxhighlight>
Während der Konfiguration:

*Typ des Setups: Wähle agent.
*Master-Verbindung: Gib die IP/Hostname des Masters ein.
*Port: 5665 (Standardport für den Agent).
*Zertifikatsanforderung: Bestätige, dass der Agent ein Zertifikat vom Master anfordert.
*Zone: Stelle sicher, dass die Zone des Agents korrekt benannt ist.
*Endpunkt: Füge den Endpunkt des Masters hinzu.
Am Master, neuen Client hinzufügen:
<syntaxhighlight lang="bash" line="1">
icinga2 pki ticket --cn '<client hostname>' --salt '<salt>' #die Ausgabe dann beim Client eintragen
icinga2 ca list
</syntaxhighlight>
Client:
<syntaxhighlight lang="bash" line="1">
sudo systemctl restart icinga2
sudo systemctl enable icinga2
</syntaxhighlight>
Füge den Agent in der /etc/icinga2/zones.conf hinzu:
<syntaxhighlight lang="bash" line="1">
object Endpoint "vps-hostname" {
host = "IP-Adresse-des-VPS"
}

object Zone "vps-hostname" {
endpoints = [ "vps-hostname" ]
parent = "icinga"
}
</syntaxhighlight>
Füge den Host in der Datei /etc/icinga2/conf.d/hosts.conf hinzu:
<syntaxhighlight lang="bash" line="1">
object Host "vps-hostname" {
address = "IP-Adresse-des-VPS"
check_command = "hostalive"
vars.os = "Linux"
}
</syntaxhighlight>
Konfiguration neu laden:
<syntaxhighlight lang="bash" line="1">
icinga2 daemon -C && systemctl restart icinga2
</syntaxhighlight>

=== Update ===

=== Test ===

=== Fehlerbehebung===
====Remote command execution failed: @@@@@@====
Bash Script auf Icinga Server erstellen:
<syntaxhighlight lang="bash" line="1">
#!/bin/bash

# Script zum Eintragen eines SSH-Host-Keys für den Nagios-Benutzer
# Verwendung: sudo ./add_known_host.sh npuls.de

HOST="$1"
USER="nagios"
SSH_DIR="/var/lib/$USER/.ssh"
KNOWN_HOSTS="$SSH_DIR/known_hosts"

if [[ -z "$HOST" ]]; then
echo "Verwendung: $0 <hostname oder IP>"
exit 1
fi

# Erstelle das .ssh-Verzeichnis, falls nicht vorhanden
if [[ ! -d "$SSH_DIR" ]]; then
echo "[INFO] Erstelle $SSH_DIR ..."
mkdir -p "$SSH_DIR"
chown "$USER:$USER" "$SSH_DIR"
chmod 700 "$SSH_DIR"
fi

# Stelle sicher, dass known_hosts existiert
if [[ ! -f "$KNOWN_HOSTS" ]]; then
echo "[INFO] Erstelle $KNOWN_HOSTS ..."
touch "$KNOWN_HOSTS"
chown "$USER:$USER" "$KNOWN_HOSTS"
chmod 600 "$KNOWN_HOSTS"
fi

# Füge den Host-Key hinzu, falls noch nicht vorhanden
echo "[INFO] Füge SSH-Host-Key für $HOST hinzu ..."
sudo -u "$USER" ssh-keyscan -H "$HOST" >> "$KNOWN_HOSTS"

echo "[OK] Host $HOST wurde zu $KNOWN_HOSTS hinzugefügt."
</syntaxhighlight>
Ausführen:
<syntaxhighlight lang="bash" line="1">
sudo chmod +x /usr/local/bin/add_known_host.sh && sudo /usr/local/bin/add_known_host.sh <IP oder Domain>
</syntaxhighlight>
====Remote command execution failed: bash: line 1: /usr/lib/nagios/plugins/check_mem.pl: No such file or directory====
<syntaxhighlight lang="bash" line="1">
wget https://raw.githubusercontent.com/justintime/nagios-plugins/master/check_mem/check_mem.pl -O /usr/lib/nagios/plugins/check_mem.pl
chmod +x /usr/lib/nagios/plugins/check_mem.pl
chown root:nagios /usr/lib/nagios/plugins/check_mem.pl
</syntaxhighlight>

=== Codeschnipsel ===

=== Nützliche Links ===

=== Codeschnipsel ===

=== Nützliche Links ===

VLLm

2025-12-18T22:06:11Z

Hendrik: /* Download */

=== Beschreibung ===
Docker normal installieren

=== Download ===
Normal (ROCm)
<syntaxhighlight lang="bash" line="1">
docker pull rocm/vllm-dev:nightly
</syntaxhighlight>
gfx906
<syntaxhighlight lang="bash" line="1">
docker pull nalanzeyu/vllm-gfx906
</syntaxhighlight>

=== Ausführen ===
Variante 1:<syntaxhighlight lang="bash" line="1">
docker run -it --rm --shm-size=8g --device=/dev/kfd --device=/dev/dri \
--group-add video -p 8086:8000 \
-v /mnt/share/models:/models \
nalanzeyu/vllm-gfx906 \
vllm serve /models/Qwen3-Coder-30B-A3B-Instruct-AWQ-4bit --served-model-name Homelab --max-model-len 30000 --enable-auto-tool-choice --tool-call-parser hermes
</syntaxhighlight>Variante 2, getestet 18.12.2025:<syntaxhighlight lang="bash">
sudo docker run -it --rm --network=host \
--group-add=video --ipc=host --cap-add=SYS_PTRACE \
--security-opt seccomp=unconfined --device /dev/kfd \
--device /dev/dri \
-v /home/hendrik/.lmstudio/models/:/app/models \
-e HF_HOME="/app/models" \
-e HF_TOKEN="<TOKEN>" \
-e NCCL_P2P_DISABLE=1 \
-e VLLM_CUSTOM_OPS=all \
-e VLLM_ROCM_USE_AITER=0 \
-e SAFETENSORS_FAST_GPU=1 \
-e PYTORCH_TUNABLEOP_ENABLED=1
rocm/vllm-dev:nightly
</syntaxhighlight>Für gfx1201:<syntaxhighlight lang="bash">
sudo docker run -it --rm --network=host \
--group-add=video --ipc=host --cap-add=SYS_PTRACE \
--security-opt seccomp=unconfined --device /dev/kfd \
--device /dev/dri \
-v /home/hendrik/.lmstudio/models/:/app/models \
-e HF_HOME="/app/models" \
-e HF_TOKEN="<TOKEN>" \
-e NCCL_P2P_DISABLE=1 \
-e VLLM_CUSTOM_OPS=all \
-e VLLM_ROCM_USE_AITER=0 \
-e SAFETENSORS_FAST_GPU=1 \
-e PYTORCH_TUNABLEOP_ENABLED=1
kyuz0/vllm-therock-gfx1201
</syntaxhighlight>

Ohne Tensor Parallism:<syntaxhighlight lang="bash">
vllm serve Qwen/Qwen3-VL-8B-Thinking --served-model-name Homelab --max_model_len 4096 --enable-auto-tool-choice --tool-call-parser hermes --reasoning-parser qwen3
</syntaxhighlight>Mit:<syntaxhighlight lang="bash">
vllm serve Qwen/Qwen3-VL-8B-Thinking --served-model-name Homelab --tp 2 --max_model_len 4096 --enable-auto-tool-choice --tool-call-parser hermes --reasoning-parser qwen3
</syntaxhighlight>
Benchmark:
<syntaxhighlight lang="bash">
vllm bench serve --num-prompts 1 --dataset-name=random --input-len 512 --output-len 128 --model Qwen/Qwen3-4B-Instruct-2507-FP8
</syntaxhighlight>

=== Test ===
=== Bekannte Probleme ===
=== Nützliche Links ===
* [https://github.com/ggml-org/llama.cpp Offizielles llama.cpp Repository]
* [https://github.com/ggml-org/llama.cpp/blob/master/examples/server/README.md llama-server Dokumentation]
* [https://github.com/ggml-org/llama.cpp/discussions llama.cpp Discussions]
* [https://huggingface.co/models?library=gguf GGUF Modelle auf Hugging Face]
* [https://docs.rocm.com/ ROCm Dokumentation]
* [https://vulkan.lunarg.com/doc/sdk Vulkan SDK Dokumentation]