Dockerfile

# syntax=docker/dockerfile:1
FROM python:3.10-slim

ENV DEBIAN_FRONTEND=noninteractive \
    PYTHONUNBUFFERED=1 \
    PYTHONDONTWRITEBYTECODE=1 \
    PORT=7860

WORKDIR /code

RUN apt-get update && apt-get install -y --no-install-recommends \
    build-essential \
    curl \
    libopenblas-dev \
    libomp-dev \
    && rm -rf /var/lib/apt/lists/*

# Install CPU-only PyTorch first so pip does not pull CUDA stacks (~3GB+) into the image (OOM on HF builders).
COPY requirements.txt .
RUN pip install --no-cache-dir --upgrade pip \
    && pip install --no-cache-dir torch --index-url https://download.pytorch.org/whl/cpu \
    && pip install --no-cache-dir -r requirements.txt \
    && pip install --no-cache-dir torch --index-url https://download.pytorch.org/whl/cpu --force-reinstall --no-deps \
    && pip install --no-cache-dir huggingface-hub sentencepiece

ENV HF_HOME=/models/huggingface \
    TRANSFORMERS_CACHE=/models/huggingface \
    HUGGINGFACE_HUB_CACHE=/models/huggingface \
    HF_HUB_CACHE=/models/huggingface

RUN mkdir -p /models/huggingface && chmod -R 777 /models/huggingface

# Low thread counts during build reduce peak RSS on small HF build VMs.
ENV OMP_NUM_THREADS=2 \
    MKL_NUM_THREADS=2 \
    NUMEXPR_NUM_THREADS=2

# Hugging Face Hub token (optional). On HF Spaces: add secret HF_TOKEN or HUGGING_FACE_HUB_TOKEN and enable
# “available during build” / Docker build args so this reaches the builder (runtime secrets alone may not apply).
ARG HF_TOKEN=
ARG HUGGING_FACE_HUB_TOKEN=
ENV HF_TOKEN=${HF_TOKEN}
ENV HUGGING_FACE_HUB_TOKEN=${HUGGING_FACE_HUB_TOKEN}

COPY . .

RUN python scripts/docker_build_assets.py

EXPOSE 7860

CMD ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "7860", "--workers", "1", "--timeout-keep-alive", "30"]