39 lines
1.2 KiB
Docker
39 lines
1.2 KiB
Docker
# PyTorch 2.6.0 + CUDA 12.6 + cuDNN9
|
|
FROM pytorch/pytorch:2.6.0-cuda12.6-cudnn9-devel
|
|
|
|
# 기본 환경 변수 설정
|
|
ENV DEBIAN_FRONTEND=noninteractive \
|
|
HF_HOME=/workspace/.cache/huggingface \
|
|
CUDA_HOME=/usr/local/cuda \
|
|
LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH} \
|
|
PIP_DISABLE_PIP_VERSION_CHECK=1 \
|
|
PYTHONUNBUFFERED=1 \
|
|
TRITON_PTXAS_PATH=/usr/local/cuda/bin/ptxas \
|
|
TORCH_CUDA_ARCH_LIST="8.0"
|
|
|
|
WORKDIR /workspace
|
|
|
|
# 필수 빌드 도구 설치
|
|
RUN apt-get update && apt-get install -y --no-install-recommends \
|
|
git build-essential ninja-build \
|
|
&& rm -rf /var/lib/apt/lists/*
|
|
|
|
# pip 업그레이드
|
|
RUN python -m pip install -U pip setuptools wheel
|
|
|
|
# 기존 라이브러리 제거 및 특정 버전 재설치
|
|
RUN pip uninstall -y vllm torch torchvision torchaudio triton flash-attn || true
|
|
RUN pip install torch==2.6.0 torchvision==0.21.0 torchaudio==2.6.0
|
|
|
|
# 프로젝트 의존성 설치
|
|
COPY requirements.txt /tmp/requirements.txt
|
|
RUN pip install -r /tmp/requirements.txt
|
|
|
|
# vLLM 특정 버전 설치
|
|
RUN pip install vllm==0.8.5
|
|
|
|
# FlashAttention 소스에서 빌드하여 설치
|
|
RUN pip cache purge && \
|
|
pip install --no-cache-dir --no-build-isolation --no-binary=flash-attn flash-attn==2.7.3
|
|
|
|
WORKDIR /workspace |