Dockerfile

FROM mambaorg/micromamba:1.5.3

# 1. Root setup for permissions (Standard HF Space requirement)
USER root
WORKDIR /app
RUN chown -R mambauser:mambauser /app

# 2. Switch to the non-root user
USER mambauser

# 3. Initialize the shell
RUN micromamba shell init --shell=bash --root-prefix=~/micromamba

# 4. INSTALL PACKAGES VIA MICROMAMBA (FAST)
# - llama-cpp-python: Pre-compiled binary (no build time)
# - gcc/make: Installed so your AI can "install anything" later
# - python=3.10: Stable version
RUN micromamba install -y -n base \
    python=3.10 \
    pip \
    llama-cpp-python \
    gcc_linux-64 \
    gxx_linux-64 \
    make \
    cmake \
    -c conda-forge && \
    micromamba clean --all --yes

# 5. Copy requirements
COPY --chown=mambauser:mambauser requirements.txt /app/requirements.txt

# 6. Install Pip packages (Gradio 5+)
ARG MAMBA_DOCKERFILE_ACTIVATE=1
RUN pip install --no-cache-dir -r requirements.txt

# 7. Download the Model (Qwen 2.5 1.5B - Fast CPU)
RUN python -c "from huggingface_hub import hf_hub_download; hf_hub_download(repo_id='Qwen/Qwen2.5-1.5B-Instruct-GGUF', filename='qwen2.5-1.5b-instruct-q4_k_m.gguf', local_dir='./models')"

# 8. Copy the application code
COPY --chown=mambauser:mambauser . /app

# 9. Expose the single allowed port
EXPOSE 7860

# 10. Start the app
CMD ["python", "main.py"]