# OpenAI-compatible server for cloudunity/stealth-rifle, served by llama.cpp.
# Exposes /v1/chat/completions, /v1/completions, /v1/models, /health.
FROM ghcr.io/ggml-org/llama.cpp:server

# Writable caches (Spaces containers have a read-only home for the default user).
ENV LLAMA_CACHE=/tmp/llama-cache \
    HF_HOME=/tmp/hf \
    XDG_CACHE_HOME=/tmp/cache

# HF Spaces routes external traffic to app_port (7860). The GGUF is pulled from
# the public model repo on first boot and cached in /tmp.
CMD ["-hf", "bartowski/Qwen2.5-3B-Instruct-GGUF:Q4_K_M", \
     "--host", "0.0.0.0", "--port", "7860", \
     "--threads", "2", "--ctx-size", "24576", \
     "--chat-template", "chatml", \
     "--alias", "stealth-rifle"]