28 lines
906 B
Docker
28 lines
906 B
Docker
FROM nvidia/cuda:12.1.0-runtime-ubuntu22.04 AS builder
|
|
|
|
# Установка зависимостей
|
|
|
|
RUN apt update && apt-get install -y \
|
|
python3.10 pip git curl wget nano sudo apt-utils && \
|
|
rm -rf /var/lib/apt/lists/*
|
|
|
|
# Установка RKLLM Toolkit
|
|
RUN git clone https://github.com/airockchip/rknn-llm
|
|
|
|
# Удаляем защиту PEP 668
|
|
RUN rm -f /usr/lib/python3.*/EXTERNALLY-MANAGED || true
|
|
|
|
RUN pip3 install torch==2.6.0 numpy==1.26.4
|
|
|
|
# 2. Установите auto_gptq БЕЗ CUDA (игнорируя суффиксы версий)
|
|
RUN pip install auto-gptq==0.7.1 --no-build-isolation
|
|
|
|
# Установка Python-пакета
|
|
RUN pip3 install /rknn-llm/rkllm-toolkit/packages/rkllm_toolkit-1.2.3-cp310-cp310-linux_x86_64.whl
|
|
|
|
# Клонирование модели
|
|
WORKDIR /work
|
|
VOLUME ["/models", "/output"]
|
|
|
|
# Точка входа для компиляции
|
|
CMD ["bash"] |