[Unit]
Name=llamacpp

[Container]
ContainerName=llamacpp
Image=localhost/llamacpp:rocm-amd64
Network=internal.network

#PublishPort=8080:8080

# ROCm
AddDevice=/dev/kfd
AddDevice=/dev/dri
PodmanArgs=--userns=keep-id --group-add=keep-groups --ipc=host
SecurityLabelType=container_runtime_t

# ROCm tuning
#Environment=HSA_OVERRIDE_GFX_VERSION=11.5.1
#Environment=ROCR_VISIBLE_DEVICES=0
#Environment=GPU_TARGETS=gfx1151

# API Key
#Environment=LLAMA_API_KEY=""

# Models
Environment=GENERAL_FAST_MODEL="-m models/gemma-3-1b-it-Q5_K_M.gguf"
Environment=GENERAL_MODEL="-m models/gpt-oss-20b-Q4_K_M.gguf"

Environment=CHAT_MODEL="-m models/Qwen3-VL-30B-A3B-Q4_K_S.gguf"

Environment=CODER_MODEL="-m models/Qwen3-Coder-30B-A3B-Instruct-Q6_K.gguf"

Environment=EMBEDDING_FAST_MODEL="-m models/embeddinggemma-300M-Q8_0.gguf"
Environment=EMBEDDING_MODEL="-m models/bge-code-v1-q6_k.gguf"

Environment=GENERAL_CONTEXT_SIZE=262144
Environment=GENERAL_GPU_LAYERS=99
Environment=GENERAL_MAX_TOKENS=512

Environment=CODER_CONTEXT_SIZE=131072
Environment=CODER_GPU_LAYERS=99
Environment=CODER_MAX_TOKENS=512

# Mount points
Volume=/srv/containers/aitools/models/llamacpp:/home/ubuntu/.cache/llama.cpp
Volume=/srv/containers/aitools/models/llamacpp:/app/models
Volume=/srv/containers/aitools/llamacpp_config.yaml:/app/config.yaml

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

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