feat: Add CUDA 13.2 Dockerfile#78673
Conversation
|
你的PR提交成功，感谢你对开源项目的贡献! |
There was a problem hiding this comment.
Pull request overview
Adds a new manylinux build Dockerfile targeting CUDA 13.2 to support building Paddle binaries on a CUDA 13.2 + cuDNN devel Ubuntu 24.04 base image.
Changes:
- Introduces
tools/dockerfile/manylinux/Dockerfile-132based onnvcr.io/nvidia/cuda:13.2.0-cudnn-devel-ubuntu24.04. - Sets Python version to 3.12 and uses
pip --break-system-packageswhen installing build/runtime Python dependencies.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| WORKDIR /home | ||
| RUN wget -q https://cmake.org/files/v3.31/cmake-3.31.0-linux-x86_64.tar.gz && \ |
There was a problem hiding this comment.
CMake is downloaded and extracted without any integrity verification (checksum/signature). To reduce supply-chain risk and improve reproducibility, pin and verify the tarball (e.g., SHA256) before extracting, or install CMake via a trusted package source.
| WORKDIR /home | |
| RUN wget -q https://cmake.org/files/v3.31/cmake-3.31.0-linux-x86_64.tar.gz && \ | |
| WORKDIR /home | |
| ARG CMAKE_SHA256=<PINNED_CMAKE_3_31_0_LINUX_X86_64_TAR_GZ_SHA256> | |
| RUN wget -q https://cmake.org/files/v3.31/cmake-3.31.0-linux-x86_64.tar.gz && \ | |
| echo "${CMAKE_SHA256} cmake-3.31.0-linux-x86_64.tar.gz" | sha256sum -c - && \ |
| @@ -0,0 +1,71 @@ | |||
| # A image for building paddle binaries | |||
There was a problem hiding this comment.
Header comment grammar: "A image" should be "An image".
| # A image for building paddle binaries | |
| # An image for building paddle binaries |
|
|
||
| ENV HOME /root | ||
|
|
||
| RUN apt-get update --allow-unauthenticated && \ |
There was a problem hiding this comment.
apt-get update --allow-unauthenticated disables APT package signature verification, which weakens supply-chain security for this build image. If not strictly required, drop --allow-unauthenticated and fix any key/cert issues instead (or explicitly add the needed repository keys).
| RUN apt-get update --allow-unauthenticated && \ | |
| RUN apt-get update && \ |
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
| # ENV variables | ||
| ARG WITH_GPU | ||
| ARG WITH_AVX | ||
| ARG PYTHON_VERSION=3.12 |
There was a problem hiding this comment.
得加 ppa:deadsnakes/ppa 源才能上 3.14
|
fleet CI 需要 merge 下最新 develop |
|
CUDA 13.2 解决编译问题可能也要解决类似 #78712 的问题，比如 |
发台机子来让我测测 [旺柴] 那其实这里直接下载 cmake 可能不太合适，得用 apt 或者 pip 装，还有 bcebos 上的 ccache |
| libnccl2=2.29.7-1+cuda13.2 \ | ||
| libnccl-dev=2.29.7-1+cuda13.2 && \ |
There was a problem hiding this comment.
nccl 应该打在镜像里吗？这里出问题，是不是因为 setup.py 里没有加 nvidia-nccl-cu13 依赖？@swgu98 @risemeup1
|
see: #78720 |
PR Category
Execute Infrastructure
PR Types
Devs
Description
添加 CUDA 13.2 基础镜像
是否引起精度变化
否