# Use a pre-compiled official HF image for llama-cpp
FROM ghcr.io/huggingface/llama-cpp-python-huggingface-endpoint:latest

# Install downloader
RUN pip install --no-cache-dir huggingface_hub

# Model details - Verified Filename
ENV REPO_ID="tensorblock/WhiteRabbitNeo-2.5-Qwen-2.5-Coder-7B-GGUF"
ENV FILENAME="WhiteRabbitNeo-2.5-Qwen-2.5-Coder-7B-Q2_K.gguf"

# 1. Download at runtime (Runtime doesn't timeout)
# 2. Start server on HF port 7860
CMD python3 -c "from huggingface_hub import hf_hub_download; hf_hub_download(repo_id='$REPO_ID', filename='$FILENAME', local_dir='.')" && \
    python3 -m llama_cpp.server --model ./$FILENAME --host 0.0.0.0 --port 7860 --n_ctx 1024 --n_threads 2