Dockerfile

# GPU-ready base with PyTorch + CUDA 12.1
FROM pytorch/pytorch:2.3.1-cuda12.1-cudnn8-runtime

ENV DEBIAN_FRONTEND=noninteractive \
    PIP_NO_CACHE_DIR=1 \
    HF_HUB_ENABLE_HF_TRANSFER=1 \
    UVICORN_WORKERS=1 \
    PORT=7860 \
    HOME=/app \
    XDG_CACHE_HOME=/app/.cache \
    HF_HOME=/app/.cache/huggingface \
    HF_HUB_CACHE=/app/.cache/huggingface/hub \
    TRANSFORMERS_CACHE=/app/.cache/huggingface/transformers \
    HF_DATASETS_CACHE=/app/.cache/huggingface/datasets \
    TORCH_HOME=/app/.cache/torch \
    PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:128

RUN apt-get update && apt-get install -y --no-install-recommends \
    python3-pip python3-dev git build-essential && \
    ln -s /usr/bin/python3 /usr/bin/python && \
    pip install --upgrade pip && \
    rm -rf /var/lib/apt/lists/*

# Make caches writable
RUN mkdir -p /app/.cache/huggingface/hub /app/.cache/huggingface/transformers \
    /app/.cache/huggingface/datasets /app/.cache/torch && \
    chmod -R 777 /app/.cache

WORKDIR /app
COPY requirements.txt /app/requirements.txt

# IMPORTANT: torch is already installed in the base image — don't overwrite it.
# Remove "torch" from requirements.txt to avoid pulling a CPU wheel by mistake.
RUN sed -i '/^torch==/d' /app/requirements.txt && pip install -r /app/requirements.txt

COPY app.py /app/app.py
EXPOSE 7860
CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]