op-plugin:基于昇腾Ascend Extension for PyTorch的算子插件项目

Star0
1
代码
代码
Issues
Pull Requests2
项目讨论
Wiki
分析
Star0
1

OpPlugin of Ascend Extension for PyTorch

分支39Tags2
文件最后提交记录最后更新时间
.gitee
update torch link
3 年前
ci
add torchnpugen package Co-authored-by: liangsongwei<liangsongwei@huawei.com> # message auto-generated for no-merge-commit merge: !4109 merge master_compile into master add torchnpugen package Created-by: liangsongwei Commit-by: liangsongwei Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: 修改codegen目录包为torchnpugen，确保编译执行OK **Special notes for your reviewers**: ![image.png](https://raw.gitcode.com/user-images/assets/7403085/79994eb4-3614-406a-b96d-7187674e8243/image.png 'image.png') See merge request: Ascend/op-plugin!4109
7 个月前
codegen
add torchnpugen package Co-authored-by: liangsongwei<liangsongwei@huawei.com> # message auto-generated for no-merge-commit merge: !4109 merge master_compile into master add torchnpugen package Created-by: liangsongwei Commit-by: liangsongwei Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: 修改codegen目录包为torchnpugen，确保编译执行OK **Special notes for your reviewers**: ![image.png](https://raw.gitcode.com/user-images/assets/7403085/79994eb4-3614-406a-b96d-7187674e8243/image.png 'image.png') See merge request: Ascend/op-plugin!4109
7 个月前
docs
fused_infer_attention_score_v2 资料修改 Co-authored-by: wuhaogl<wuhao305@h-partners.com> # message auto-generated for no-merge-commit merge: !4105 merge master into master fused_infer_attention_score_v2 资料修改 Created-by: wuhaogl Commit-by: wuhaogl Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: **Special notes for your reviewers**: See merge request: Ascend/op-plugin!4105
7 个月前
examples
[feature] compile adapter
7 个月前
op_plugin
add DynamicDualLevelMxQuant op interface Co-authored-by: wang-siyang3<wangsiyang3@huawei.com> # message auto-generated for no-merge-commit merge: !4000 merge dev into master add DynamicDualLevelMxQuant op interface Created-by: clwsy1 Commit-by: wang-siyang3 Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: **Special notes for your reviewers**: See merge request: Ascend/op-plugin!4000
7 个月前
test
add DynamicDualLevelMxQuant op interface Co-authored-by: wang-siyang3<wangsiyang3@huawei.com> # message auto-generated for no-merge-commit merge: !4000 merge dev into master add DynamicDualLevelMxQuant op interface Created-by: clwsy1 Commit-by: wang-siyang3 Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: **Special notes for your reviewers**: See merge request: Ascend/op-plugin!4000
7 个月前
torchnpugen
add torchnpugen package Co-authored-by: liangsongwei<liangsongwei@huawei.com> # message auto-generated for no-merge-commit merge: !4109 merge master_compile into master add torchnpugen package Created-by: liangsongwei Commit-by: liangsongwei Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: 修改codegen目录包为torchnpugen，确保编译执行OK **Special notes for your reviewers**: ![image.png](https://raw.gitcode.com/user-images/assets/7403085/79994eb4-3614-406a-b96d-7187674e8243/image.png 'image.png') See merge request: Ascend/op-plugin!4109
7 个月前
.gitignore
init submodule
3 年前
LICENSE
update LICENSE. Signed-off-by: weili10 <liwei386@huawei.com>
2 年前
OWNERS
Update URL (gitcode) Co-authored-by: dilililiwhy<why.wuhuanyu@huawei.com> # message auto-generated for no-merge-commit merge: merge update_url into master Update URL (gitcode) Created-by: dilililiwhy Commit-by: dilililiwhy Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: **Special notes for your reviewers**: See merge request: Ascend/op-plugin!3165
11 个月前
README.md
add v2r11 Co-authored-by: dilililiwhy<why.wuhuanyu@huawei.com> # message auto-generated for no-merge-commit merge: !4046 merge add_v2r11 into master Checkout v2r11 Created-by: dilililiwhy Commit-by: dilililiwhy Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: checkout v2r11 for pta upgrade **Special notes for your reviewers**: See merge request: Ascend/op-plugin!4046
7 个月前
SECURITYNOTE.md
readme update Co-authored-by: LQ1206<liuqian164@h-partners.com> # message auto-generated for no-merge-commit merge: merge master into master Readme update Created-by: LQ1206 Commit-by: LQ1206 Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: **Special notes for your reviewers**: See merge request: Ascend/op-plugin!3191
11 个月前
Third_Party_Open_Source_Software_Notice
!2388 [master] Update the referenced PyTorch version to v2.3.1. Merge pull request !2388 from shaoyf/master_notice
1 年前
gencode.sh
add torchnpugen package Co-authored-by: liangsongwei<liangsongwei@huawei.com> # message auto-generated for no-merge-commit merge: !4109 merge master_compile into master add torchnpugen package Created-by: liangsongwei Commit-by: liangsongwei Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: 修改codegen目录包为torchnpugen，确保编译执行OK **Special notes for your reviewers**: ![image.png](https://raw.gitcode.com/user-images/assets/7403085/79994eb4-3614-406a-b96d-7187674e8243/image.png 'image.png') See merge request: Ascend/op-plugin!4109
7 个月前

OpPlugin

简介

本项目开发了Ascend Extension for Pytorch（torch_npu）算子插件，为使用PyTorch框架的开发者提供便捷的NPU算子库调用能力。 OpPlugin算子插件的编译、使用依赖昇腾Ascend Extension for PyTorch。因此，在编译OpPlugin之前，需要了解、安装昇腾PyTorch。使用手册可参考昇腾社区Ascend Extension for Pytorch。

版本配套表

op-plugin仓旨在为torch_npu提供运行所需要的算子适配文件，两个仓的对应关系如下：

op-plugin分支 对应Ascend Extension for PyTorch版本
master 主线版本，如v2.1.0等
7.1.0 7.1.0版本，如v2.1.0-7.1.0等
7.0.0 7.0.0版本，如v2.1.0-7.0.0等
6.0.0 6.0.0版本，如v2.1.0-6.0.0等
6.0.rc3 6.0.rc3版本，如v2.1.0-6.0.rc3等
6.0.rc2 6.0.rc2版本，如v2.1.0-6.0.rc2等
6.0.rc1 6.0.rc1版本，如v2.1.0-6.0.rc1等
5.0.0 5.0.0版本，如v2.1.0-5.0.0等
5.0.rc3 5.0.rc3版本，如v2.1.0-5.0.rc3等

安装OpPlugin

发布包安装

暂未正式发布

源码编译安装

安装依赖

安装时需要安装系统依赖及官方PyTorch框架，建议使用torch_npu提供的docker镜像进行编译，依赖安装和镜像使用指导可参考Ascend Extension for Pytorch。

操作步骤

  1. 配置CANN环境变量脚本。

    source <CANN软件安装目录>/<CANN软件路径>/set_env.sh

    环境变量脚本的默认路径一般为：/usr/local/Ascend/ascend-toolkit/set_env.sh，其中ascend-toolkit路径取决于安装的CANN软件名称。

  2. 编译生成插件的二进制安装包。

    下载对应OpPlugin版本分支代码，进入插件根目录，以v2.1为例。

    git clone --branch master https://gitcode.com/ascend/op-plugin.git
cd op-plugin

    执行编译构建，当前支持torch_npu 2.1/2.3/2.4/2.5/2.6/2.7/2.8/master版本。

    bash ci/build.sh --python=3.8 --pytorch=v2.1.0

    须知：
    请注意，编译时gcc版本遵循如下约束：

    • ARM架构下推荐使用gcc 10.2版本
    • X86架构下推荐使用gcc 9.3.1; PyTorch 2.6.0及之后版本的编译推荐使用gcc 11.2.1
    参数 取值范围 说明 缺省值 备注
    PyTorch v2.1.0, v2.3.1, v2.4.0, v2.5.1, v2.6.0, v2.7.1, v2.8.0, master 指定编译过程中使用的PyTorch版本 master master对应的PyTorch版本为2.9
    Python 3.8, 3.9, 3.10, 3.11 指定编译过程中使用的Python版本 3.8 编译基于PyTorch2.5及后续版本时，不支持Python3.8
    PyTorch Python GCC
    v2.1.0 3.8, 3.9, 3.10, 3.11 ARM: 10.2 x86: 9.3
    v2.6.0 3.9, 3.10, 3.11 ARM: 11.2 x86: 9.3
    v2.7.1 3.9, 3.10, 3.11 11.2
    v2.8.0 3.9, 3.10, 3.11 13.3
    v2.9.0 3.10, 3.11 13.3
    v2.10.0 3.10, 3.11 13.3
    master(2.11) 3.10, 3.11 13.3

    编译过程中，会在插件根目录新建build文件夹，并下载torch_npu对应版本的源码，协同编译。 若build/pytorch目录存在，则编译op-plugin时，不再重复下载torch_npu源码。如需下载所依赖的最新torch_npu源码，删除build/pytorch目录即可。

  3. 完成编译后，安装dist目录下生成的插件torch_npu包，如果使用非root用户安装，需要在命令后加--user。

    pip3 install --upgrade dist/torch_npu-{torch_npu_version}-{Python_version}-{arch}.whl
# 实际执行时需要根据生成的whl包名称进行替换，其中{torch_npu_version}表示编译的torch_npu版本，{Python_version} 为所使用的 Python 版本，{arch} 则代表目标架构。
# 典型的whl包名类似：torch_npu-2.1.0.post13-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl

硬件配套

昇腾训练设备包含以下型号，都可作为PyTorch模型的训练环境。

产品系列 产品型号
Atlas 训练系列产品 Atlas 800 训练服务器（型号：9000）
Atlas 800 训练服务器（型号：9010）
Atlas 900 PoD（型号：9000）
Atlas 300T 训练卡（型号：9000）
Atlas 300T Pro 训练卡（型号：9000）
Atlas A2 训练系列产品 Atlas 800T A2 训练服务器
Atlas 900 A2 PoD 集群基础单元
Atlas 200T A2 Box16 异构子框
Atlas A3 训练系列产品 Atlas 800T A3 训练服务器
Atlas 900 A3 SuperPoD 超节点

昇腾推理设备包含以下型号，都可作为大模型的推理环境。

产品系列 产品型号
Atlas 800I A2推理产品 Atlas 800I A2 推理服务器

生命周期

op-plugin仓依赖torch_npu运行，生命周期请参考torch_npu中的PyTorch版本维护策略。

安全声明

OpPlugin安全声明

项目介绍

OpPlugin of Ascend Extension for PyTorch

Python1.82 K提交数
定制我的领域
README
规则集