Dockerfile

# FastAPI inference Space for Committed.
# This is the DEPLOY Dockerfile, living in the HF Space repo (not the in-repo
# src/committed/serving/Dockerfile, which is the reference container definition).
# It installs the `committed` package straight from GitHub (single source of
# truth, per ADR 0047) and runs the FastAPI app. No model is baked: the engine
# pulls the fine-tuned GGUF from the Hub at runtime via its defaults (ADR 0048)
# or the COMMITTED_MODEL_* Space Variables.

FROM python:3.11-slim

# llama-cpp-python compiles native code at install time, so a C/C++ toolchain
# is required in the build image.
RUN apt-get update \
    && apt-get install -y --no-install-recommends build-essential git \
    && rm -rf /var/lib/apt/lists/*

# HF Spaces run as a non-root user (uid 1000). Create it and own the workdir so
# the model cache under HF_HOME is writable at runtime.
RUN useradd -m -u 1000 user
USER user
ENV HOME=/home/user \
    PATH=/home/user/.local/bin:$PATH \
    HF_HOME=/home/user/.cache/huggingface

WORKDIR /home/user/app

# Install the package and its required (serve-minimal) deps from GitHub.
# Prebuilt CPU wheels for llama-cpp-python avoid a slow source compile on the
# free CPU Space; the extra index is where those wheels are published.
RUN pip install --no-cache-dir \
    --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu \
    "committed @ git+https://github.com/marzoukbaig14/Committed.git"

# HF Spaces route public traffic to port 7860.
EXPOSE 7860

# Launch the FastAPI app. The model loads once at startup (lifespan); first boot
# downloads the GGUF from the Hub into HF_HOME, then it is cached.
CMD ["uvicorn", "committed.serving.api:app", "--host", "0.0.0.0", "--port", "7860"]