Dockerfile

FROM nvidia/cuda:12.1.1-base-ubuntu22.04  

# System dependencies  
RUN apt update && apt install -y python3 python3-pip git  

# PyTorch & Transformers (GPU Optimized)  
RUN pip install torch torchvision torchaudio transformers accelerate bitsandbytes  

# Hugging Face Libraries  
RUN pip install fastapi uvicorn peft auto-gptq  

# Working directory set  
WORKDIR /app  
COPY . /app  

# Expose API port  
EXPOSE 7860  

# Start chatbot  
CMD ["python3", "server.py"]