Dockerfile

# Use a lightweight PyTorch image with CUDA support
FROM pytorch/pytorch:2.0.0-cuda11.7-cudnn8-runtime

# Set working directory
WORKDIR /code

# Install dependencies
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

# Reduce memory usage (optional)
ENV TRANSFORMERS_NO_ADVISORY_WARNINGS=1
ENV HF_HOME="/code/hf_cache"

# Expose API port
EXPOSE 7860

# Run FastAPI
CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "7860"]