FROM runpod/pytorch:1.5.0-cu1320-torch2130-ubuntu2404
WORKDIR /root

RUN mkdir llama.cpp
COPY llama.cpp ./llama.cpp/
RUN git clone https://github.com/silveroxides/convert_to_quant && cd convert_to_quant && pip install -r requirements.txt && pip install .

RUN mkdir nvfp4_gguf_quant
WORKDIR /root/nvfp4_gguf_quant
COPY --exclude=__pycache__ nvfp4gguf ./nvfp4gguf
COPY --chmod=0644 gguf_to_nvfp4_gguf.py hf_to_nvfp4_gguf.py hf_to_nvfp4_safetensors.py requirements.txt safetensors_to_nvfp4_gguf.py .

RUN pip install -r requirements.txt
COPY --chmod=0644 docker/README.txt /workspace/README.txt
COPY --chmod=0755 docker/hf_to_nvfp4_gguf docker/hf_to_nvfp4_safetensors docker/gguf_to_nvfp4_gguf docker/safetensors_to_nvfp4_gguf /usr/local/bin/

RUN apt remove -y python3-cryptography && pip install heretic-llm

WORKDIR /workspace

