bdi_podman_serverconf/Services/llamacpp/llamacpp.container

[Container]
ContainerName=llamacpp
Image=localhost/llamacpp:vulkan-amd64
#AutoUpdate=registry
Network=internal.network
PublishPort=8090:8090

# Production - Lemonade usa Hugging Face Hub per i modelli
Volume=/srv/containers/aitools/models/hf:/root/.cache/huggingface/hub
# Vecchia cartella
#Volume=/srv/containers/aitools/models:/root/.cache/llama.cpp

# ROCm tuning
AddDevice=/dev/dri/renderD128
PodmanArgs=--group-add=keep-groups --ipc=host
SecurityLabelType=container_runtime_t

Environment=LLAMA_ARG_HOST=0.0.0.0
Environment=LLAMA_ARG_PORT=8090
Environment=LLAMA_ARG_NO_MMAP=true
Environment=LLAMA_ARG_CTX_SIZE=131072
Environment=LLAMA_ARG_HF_REPO=unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF:Q2_K

# MTP
Environment=LLAMA_ARG_SPEC_TYPE=draft-mtp
Environment=LLAMA_ARG_SPEC_DRAFT_N_MAX=6

# HF
Environment=HF_HOME=/root/.cache/huggingface
Environment=HF_TOKEN=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

# API Key — protezione accesso al server (da impostare anche in Odoo > AI > API Key)
# Decommentare per abilitare l'autenticazione
#Environment=LLAMA_API_KEY=la-tua-chiave-segreta

# Model alias — nome con cui il server risponde alle richieste (campo "model" nell'API)
# Se non impostato, llama-server usa il nome del file .gguf come alias
#Environment=LLAMA_ARG_ALIAS=chat

# Override Boot
#Entrypoint=/usr/bin/sleep
#Exec=infinity

[Service]
Restart=on-failure
TimeoutStartSec=15m

[Install]
WantedBy=multi-user.target default.target