hf-space-rag/Dockerfile

# CPU + FREE-tier RAG Space (Docker SDK).
#   Embeddings : BAAI/bge-small-en-v1.5  (fastembed / ONNX, no torch)
#   Vector DB  : FAISS (in-memory)
#   LLM        : Qwen2.5-1.5B-Instruct   (llama.cpp GGUF, fast on CPU)
#   API        : OpenAI-compatible  ->  /v1/chat/completions  (+ web UI at /)
#
# Everything is CPU-only. No GPU, no paid hardware required.
# Models are baked into the image so the Space cold-starts instantly.

FROM python:3.11-slim

# build-essential + cmake so llama-cpp-python can compile if no prebuilt wheel
# matches (python:slim has no compiler otherwise -> CMake "gcc not found").
RUN apt-get update && apt-get install -y --no-install-recommends \
        build-essential cmake git curl ca-certificates && rm -rf /var/lib/apt/lists/*

# Hugging Face Spaces run the container as non-root UID 1000.
RUN useradd -m -u 1000 user
USER user
ENV HOME=/home/user \
    PATH=/home/user/.local/bin:$PATH
WORKDIR /home/user/app

# --- Python deps ---
# GGML_NATIVE=OFF -> portable CPU binary (no -march=native), so it can't crash
# with "illegal instruction" if the build CPU != the runtime CPU.
COPY --chown=user requirements.txt .
ENV CMAKE_ARGS="-DGGML_NATIVE=OFF" FORCE_CMAKE=1
RUN pip install --no-cache-dir --user -r requirements.txt

# --- Model config ------------------------------------------------------------
ENV LLM_REPO=Qwen/Qwen2.5-1.5B-Instruct-GGUF \
    LLM_FILE=qwen2.5-1.5b-instruct-q4_k_m.gguf \
    MODEL_DIR=/home/user/models \
    EMBED_MODEL=BAAI/bge-small-en-v1.5 \
    FASTEMBED_CACHE=/home/user/.cache/fastembed \
    HF_HOME=/home/user/.cache/huggingface

# Bake the LLM (~1 GB) into the image -> instant cold start.
RUN python -c "from huggingface_hub import hf_hub_download; \
    hf_hub_download(repo_id='${LLM_REPO}', filename='${LLM_FILE}', local_dir='${MODEL_DIR}')"

# Bake the embedding model (ONNX ~130 MB) into the image too.
RUN python -c "import os; from fastembed import TextEmbedding; \
    TextEmbedding(os.environ['EMBED_MODEL'], cache_dir=os.environ['FASTEMBED_CACHE'])"

# --- App ---------------------------------------------------------------------
COPY --chown=user . .

# Runtime knobs (free tier = 2 vCPU).
ENV N_CTX=8192 \
    N_THREADS=2 \
    TOP_K=4 \
    DOCS_DIR=documents

EXPOSE 7860
CMD ["python", "-m", "uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]