Dockerfile

# ── Base image ─────────────────────────────────────────────────────────────────
# Plain CUDA runtime — torch installed explicitly so its version is
# guaranteed to align with bitsandbytes (avoids the conda torch mismatch).
FROM nvidia/cuda:12.1.1-cudnn8-runtime-ubuntu22.04

# ── System deps ────────────────────────────────────────────────────────────────
ENV DEBIAN_FRONTEND=noninteractive
RUN apt-get update && apt-get install -y --no-install-recommends \
        python3.10 \
        python3.10-dev \
        python3-pip \
        git \
        curl \
    && ln -sf python3.10 /usr/bin/python \
    && ln -sf python3.10 /usr/bin/python3 \
    && rm -rf /var/lib/apt/lists/*


# ── Install remaining Python deps ──────────────────────────────────────────────
COPY requirements.txt /tmp/requirements.txt
RUN pip install --no-cache-dir -r /tmp/requirements.txt

# ── App code ───────────────────────────────────────────────────────────────────
WORKDIR /app
COPY app.py /app/app.py

# ── LoRA adapter weights ───────────────────────────────────────────────────────
# Mount your saved adapter at /app/adapter at runtime:
#   docker run --gpus all -v /local/path/gemma-math-weakness:/app/adapter ...
# Or COPY them in at build time (larger image):
# COPY gemma-math-weakness/ /app/adapter/

# ── Environment variables ──────────────────────────────────────────────────────
ENV HF_TOKEN=""
ENV MODEL_NAME="google/gemma-4-E2B-it"
ENV HF_HOME="/app/.cache/huggingface"
ENV TOKENIZERS_PARALLELISM=false

# ── Expose API port ────────────────────────────────────────────────────────────
EXPOSE 7860

# ── Run ────────────────────────────────────────────────────────────────────────
CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860", "--workers", "1"]