From aafe6822e652bf2c780b6198800de37c49bd0605 Mon Sep 17 00:00:00 2001 From: wuyulong11 Date: Thu, 23 Nov 2023 15:34:13 +0800 Subject: [PATCH] =?UTF-8?q?=E3=80=90=E4=BF=AE=E6=94=B9=E4=BF=A1=E6=81=AF?= =?UTF-8?q?=E3=80=91=E3=80=90tbplugin=E3=80=91=E3=80=90=E9=9C=80=E6=B1=82?= =?UTF-8?q?=E3=80=91tensorboard=E7=A6=BB=E7=BA=BF=E7=89=88=E6=9C=AC?= =?UTF-8?q?=E9=80=82=E9=85=8D=EF=BC=8C=E8=B5=84=E6=96=99=E4=BF=AE=E6=94=B9?= =?UTF-8?q?=20=E3=80=90=E4=BF=AE=E6=94=B9=E4=BA=BA=E3=80=91=20wuyulong=203?= =?UTF-8?q?0031080?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- .../tensorboard-plugins/tb_plugin/README.md | 35 +++++++++---------- 1 file changed, 16 insertions(+), 19 deletions(-) diff --git a/plugins/tensorboard-plugins/tb_plugin/README.md b/plugins/tensorboard-plugins/tb_plugin/README.md index 7f50fc20e..0418cd8bb 100644 --- a/plugins/tensorboard-plugins/tb_plugin/README.md +++ b/plugins/tensorboard-plugins/tb_plugin/README.md @@ -1,43 +1,40 @@ # PyTorch Profiler TensorBoard NPU Plugin ### 介绍 -此工具是PyTorch profiling数据以及可视化的TensorBoard的插件。 +此工具是PyTorch profiling数据以及可视化的TensorBoard的插件。 \ 它支持将Ascend平台采集、解析的Pytorch Profiling数据可视化呈现,也兼容GPU数据采集、解析可视化,现已支持PyTorch 2.0GPU版本的profiling数据可视化。同时集成了精度比对的功能,支持查看loss曲线和比对两个网络的loss收敛趋势。 ### 快速安装说明 * 相关依赖: pandas >= 1.0.0 ,tensorboard >= 2.11.0,protobuf <= 3.20.3 * 安装方式 - 1. pip安装(推荐) + 1. pip安装(推荐) \ 现本插件已经上传到pypi社区,用户可在python环境下直接通过以下pip指令进行安装: `pip install torch-tb-profiler-ascend` 2. 插件离线方式安装 + * 插件下载地址 \ + https://mindstudio-sample.obs.cn-north-4.myhuaweicloud.com/torch-tb-profiler-ascend/v0.4.0.2/torch_tb_profiler_ascend-0.4.0.2-py3-none-any.whl - * 插件下载地址 \ - 正式版:https://mindstudio-sample.obs.cn-north-4.myhuaweicloud.com/torch-tb-profiler-ascend/v0.4.0/torch_tb_profiler_ascend-0.4.0-py3-none-any.whl \ - 离线版:https://mindstudio-sample.obs.cn-north-4.myhuaweicloud.com/torch-tb-profiler-ascend/v0.4.0/offline/torch_tb_profiler_ascend-0.4.0-py3-none-any.whl + * 插件形式为whl包,使用指令安装(此处{version}为whl包实际版本) - * 插件形式为whl包,使用指令安装 - - `pip install torch-tb-profiler_npu_0.4.0_py3_none_any.whl` + `pip install torch-tb-profiler_ascend_{version}_py3_none_any.whl` 3. 从源代码安装 + * 从仓库下载源码: - * 从仓库下载源码: - - `git clone https://gitee.com/ascend/att.git` + `git clone https://gitee.com/ascend/att.git` - * 进入目录 `/plugins/tensorboard_plugins/tb_plugin` 下. + * 进入目录 `/plugins/tensorboard_plugins/tb_plugin` 下. - * 执行安装命令: - `pip install .` - * 构建whl包 - - `python setup.py build_fe sdist bdist_wheel` \ - 注意: build_fe步骤需要安装yarn和Node.js环境 - - `python setup.py sdist bdist_wheel` + * 执行安装命令: + `pip install .` + * 构建whl包 + - `python setup.py build_fe sdist bdist_wheel` \ + 注意: build_fe步骤需要安装yarn和Node.js环境 + - `python setup.py sdist bdist_wheel` - 在 `/tb_plugins/profiling/tb_plugin/dist` 目录下取出whl包,使用方式1进行安装 + 在 `/tb_plugins/profiling/tb_plugin/dist` 目录下取出whl包,使用方式2进行安装 ### 解析数据说明 -- Gitee