Dockerfile

# force rebuild

FROM nvidia/cuda:11.8.0-cudnn8-runtime-ubuntu22.04

ENV DEBIAN_FRONTEND=noninteractive
ENV OMP_NUM_THREADS=1
ENV PYTHONUNBUFFERED=1

RUN apt-get update && apt-get install -y \
    python3-pip git wget && \
    rm -rf /var/lib/apt/lists/*

WORKDIR /app
COPY . /app

RUN pip3 install --upgrade pip


# Install CUDA torch
RUN pip3 install torch --index-url https://download.pytorch.org/whl/cu118

# Install PEFT dev commit (your adapter requires this)
RUN pip3 install --no-cache-dir git+https://github.com/huggingface/peft.git@eaecab993365da1c939347155a961bc00168d672

# Install accelerate >= 0.34.0 (required for clear_device_cache)
RUN pip3 install --no-cache-dir "accelerate>=0.34.0"

# FIX: Install transformers >= 4.38.0
RUN pip3 install --no-cache-dir "transformers>=4.38.0" safetensors flask
RUN pip3 install bitsandbytes


EXPOSE 7806

CMD ["python3", "app.py"]