diff --git a/docs/en/PyTorch Installation Guide/PyTorch Installation Guide.md b/docs/en/PyTorch Installation Guide/PyTorch Installation Guide.md index 28542efee4d9c8dda4afe59ab9c20562dc31699c..333f7c8924c59786f9e714bc027d73d1fbed8d28 100644 --- a/docs/en/PyTorch Installation Guide/PyTorch Installation Guide.md +++ b/docs/en/PyTorch Installation Guide/PyTorch Installation Guide.md @@ -8,7 +8,7 @@ - [References](#referencesmd) - [Installing CMake](#installing-cmakemd) - [How Do I Install GCC 7.3.0?](#how-do-i-install-gcc-7-3-0md) - - [What Do I Do If "torch 1.5.0xxxx" and "torchvision" Do Not Match When torch-\*.whl Is Installed?](#what-do-i-do-if-torch-1-5-0xxxx-and-torchvision-do-not-match-when-torch--whl-is-installedmd) + - [What to Do If "torch 1.5.0xxxx" and "torchvision" Do Not Match When torch-\*.whl Is Installed?](#what-to-do-if-torch-1-5-0xxxx-and-torchvision-do-not-match-when-torch--whl-is-installedmd)

Overview

When setting up the environment for PyTorch model development and running, you can manually build and install the modules adapted to the PyTorch framework on a server. @@ -33,10 +33,11 @@ When setting up the environment for PyTorch model development and running, you c #### Prerequisites -- The development or operating environment of CANN has been installed. For details, see the _CANN Software Installation Guide_. +- The development or operating environment of CANN has been installed. For details, see the _CANN Software Installation Guide_. - CMake 3.12.0 or later has been installed. For details about how to install CMake, see [Installing CMake](#installing-cmakemd). - GCC 7.3.0 or later has been installed. For details about how to install and use GCC 7.3.0, see [How Do I Install GCC 7.3.0?](#how-do-i-install-gcc-7-3-0md). -- Python 3.7.5 or 3.8 has been installed. +- Python 3.7.5, 3.8, or 3.9 has been installed. +- Note that PyTorch 1.5 does not support Python 3.9 build and installation. Only Torch 1.8.1 supports Python 3.9 build and installation. - The Patch and Git tools have been installed in the environment. To install the tools for Ubuntu and CentOS, run the following commands: - Ubuntu @@ -70,10 +71,13 @@ When setting up the environment for PyTorch model development and running, you c 3. Obtain the PyTorch source code. - 1. Run the following command to obtain the PyTorch source code adapted to Ascend AI Processors: + 1. Run the following command to obtain the PyTorch source code adapted to Ascend AI Processors and switch to the required branch: ``` git clone https://gitee.com/ascend/pytorch.git + # By default, the masterf branch is used. If other branches are required, run the git checkout command to switch to that branch. + # git checkout -b 2.0.3.tr5 remotes/origin/2.0.3.tr5 + ``` The directory structure of the downloaded source code is as follows: @@ -106,7 +110,7 @@ When setting up the environment for PyTorch model development and running, you c git clone -b v1.8.1 --depth=1 https://github.com/pytorch/pytorch.git ``` - 3. Run the following commands to go to the native PyTorch code directory **pytorch** and obtain the PyTorch passive dependency code: + 3. Go to the native PyTorch code directory **pytorch** and obtain the PyTorch passive dependency code. ``` cd pytorch @@ -143,6 +147,9 @@ When setting up the environment for PyTorch model development and running, you c bash build.sh --python=3.7 or bash build.sh --python=3.8 + or + bash build.sh --python=3.9 # PyTorch 1.5 does not support build and installation using Python 3.9. + ``` Specify the Python version in the environment for build. The generated binary package is stored in the current dist directory **pytorch/pytorch/dist**. @@ -179,7 +186,7 @@ After the software packages are installed, configure environment variables to us export HCCL_WHITELIST_DISABLE=1 # Disable the HCCL trustlist. # Scenario 2: Multi-node scenario export HCCL_WHITELIST_DISABLE=1 # Disable the HCCL trustlist. - export HCCL_IF_IP="1.1.1.1" # 1.1.1.1 is the NIC IP address of the host. Change it based on the site requirements. Ensure that the NIC IP addresses in use can communicate with each other in the cluster. + export HCCL_IF_IP="1.1.1.1" # Replace 1.1.1.1 with the actual NIC IP address of the host. Ensure that the NIC IP addresses in use can communicate with each other in the cluster. ``` 3. \(Optional\) Configure function or performance environment variables in the NPU scenario. The variables are disabled by default. @@ -338,7 +345,7 @@ After the software packages are installed, configure environment variables to us apex │ ├─patch # Directory of the patch adapted to Ascend AI Processors │ ├─npu.patch - │ ├─scripts # Build and create a directory. + │ ├─scripts # Build and creation directory │ ├─gen.sh │ ├─src # Source code directory │ ├─tests # Directory for storing test cases @@ -358,7 +365,7 @@ After the software packages are installed, configure environment variables to us │ ├─apex # Directory for storing the native Apex code │ ├─patch # Directory of the patch adapted to Ascend AI Processors │ ├─npu.patch - │ ├─scripts # Build and create a directory. + │ ├─scripts # Build and creation directory │ ├─gen.sh │ ├─src # Source code directory │ ├─tests # Directory for storing test cases @@ -384,7 +391,7 @@ After the software packages are installed, configure environment variables to us The full code adapted to Ascend AI Processors is generated in the **apex/apex** directory. - 2. Go to the full code directory **apex/apex**, and compile and generate the binary installation package of Apex. + 2. Go to the full code directory **apex/apex**, and build and generate the binary installation package of Apex. ``` cd ../apex @@ -414,12 +421,12 @@ After the software packages are installed, configure environment variables to us - **[How Do I Install GCC 7.3.0?](#how-do-i-install-gcc-7-3-0md)** -- **[What Do I Do If "torch 1.5.0xxxx" and "torchvision" Do Not Match When torch-\*.whl Is Installed?](#what-do-i-do-if-torch-1-5-0xxxx-and-torchvision-do-not-match-when-torch--whl-is-installedmd)** +- **[What to Do If "torch 1.5.0xxxx" and "torchvision" Do Not Match When torch-\*.whl Is Installed?](#what-to-do-if-torch-1-5-0xxxx-and-torchvision-do-not-match-when-torch--whl-is-installedmd)**

Installing CMake

-Procedure for upgrading CMake to 3.12.1 +The following describes how to install CMake 3.12.1. 1. Obtain the CMake software package. @@ -447,8 +454,7 @@ Procedure for upgrading CMake to 3.12.1 ln -s /usr/local/cmake/bin/cmake /usr/bin/cmake ``` -5. Run the following command to check whether CMake has been installed: - +5. Check whether CMake has been installed. ``` cmake --version ``` @@ -525,7 +531,7 @@ Perform the following steps as the **root** user. 5. Set the environment variable. - Training must be performed in the compilation environment with GCC upgraded. If you want to run training, configure the following environment variable in your training script: + Training must be performed in the compilation environment with GCC upgraded. Therefore, configure the following environment variable in your training script: ``` export LD_LIBRARY_PATH=${install_path}/lib64:${LD_LIBRARY_PATH} @@ -537,11 +543,11 @@ Perform the following steps as the **root** user. >Skip this step if you do not need to use the compilation environment with GCC upgraded. -

What Do I Do If "torch 1.5.0xxxx" and "torchvision" Do Not Match When torch-\*.whl Is Installed?

+

What to Do If "torch 1.5.0xxxx" and "torchvision" Do Not Match When torch-\*.whl Is Installed?

#### Symptom -During the installation of **torch-**_\*_**.whl**, the message "ERROR: torchvision 0.6.0 has requirement torch==1.5.0, but you'll have torch 1.5.0a0+1977093 which is incompatible" " is displayed. +During the installation of **torch-**_\*_**.whl**, the message "ERROR: torchvision 0.6.0 has requirement torch==1.5.0, but you'll have torch 1.5.0a0+1977093 which is incompatible" is displayed. ![](figures/en-us_image_0000001190081735.png) diff --git a/docs/en/PyTorch Online Inference Guide/PyTorch Online Inference Guide.md b/docs/en/PyTorch Online Inference Guide/PyTorch Online Inference Guide.md index fccfe6ab1811a7325772b63b6e640c99fc47e6ff..a3d28aff597a55f41c28d61e8e044dd7799e2275 100644 --- a/docs/en/PyTorch Online Inference Guide/PyTorch Online Inference Guide.md +++ b/docs/en/PyTorch Online Inference Guide/PyTorch Online Inference Guide.md @@ -57,10 +57,10 @@ The following are the environment variables required for starting the inference export PATH=/usr/local/python3.7.5/bin:$PATH export LD_LIBRARY_PATH=/usr/local/python3.7.5/lib:$LD_LIBRARY_PATH -# Sets the logical ID of a processor. +# Set the logical ID of a processor. export ASCEND_DEVICE_ID=0 -# Outputs log information. Replace it as required. +# Output log information. Replace it as required. export ASCEND_SLOG_PRINT_TO_STDOUT=1 export ASCEND_GLOBAL_LOG_LEVEL=0 @@ -416,11 +416,11 @@ The following uses the ResNet-50 model as an example to describe how to perform 2. Edit the inference script. - Create a model script file **resnet50\_infer\_for\_pytorch.py** and write code by referring to [Sample Code](). + Create a model script file **resnet50\_infer\_for\_pytorch.py** and write code. For how to write the code, see [Sample Code](). 3. Run inference. - Set environment variables by referring to [Environment Variable Configuration](#environment-variable-configurationmd) and then run the following command: + Set environment variables (for how to set them, see [Environment Variable Configuration](#environment-variable-configurationmd)) and then run the following command: ``` python3 pytorch-resnet50-apex.py --data /data/imagenet \ @@ -491,13 +491,13 @@ However, the mixed precision training is limited by the precision range expresse #### Initializing the Mixed Precision Model -1. To use the mixed precision module Apex, you need to import the amp module from the Apex library as follows: +1. To use the mixed precision module Apex, import the amp module from the Apex library. ``` from apex import amp ``` -2. After the amp module is imported, you need to initialize it so that it can modify the model, optimizer, and PyTorch internal functions. The initialization code is as follows: +2. Initialize the amp module so that it can modify the model, optimizer, and PyTorch internal functions. ``` model, optimizer = amp.initialize(model, optimizer) @@ -585,7 +585,7 @@ Perform the following steps as the **root** user. 5. Set the environment variable. - The build environment after GCC upgrade is required for training. Therefore, you need to configure the following environment variable in the training script: + Training must be performed in the compilation environment with GCC upgraded. Therefore, configure the following environment variable in the training script: ``` export LD_LIBRARY_PATH=${install_path}/lib64:${LD_LIBRARY_PATH} @@ -594,6 +594,6 @@ Perform the following steps as the **root** user. **$\{install\_path\}** indicates the GCC 7.3.0 installation path configured in [3.](#en-us_topic_0000001146754749_en-us_topic_0000001072593337_l75d31a2874534a2092e80a5f865b46f0). In this example, the GCC 7.3.0 installation path is **/usr/local/linux\_gcc7.3.0/**. >![](public_sys-resources/icon-note.gif) **NOTE:** - >The environment variable needs to be configured only when you need to use the build environment after the GCC upgrade. + >Skip this step if you do not the compilation environment with GCC upgraded. diff --git a/docs/en/PyTorch Operator Development Guide/PyTorch Operator Development Guide.md b/docs/en/PyTorch Operator Development Guide/PyTorch Operator Development Guide.md index 78351f5798567a8c2ffbcb5c6fe592b9d8635816..48ddd588986a17becfa7645d7a5c905a05cb490b 100644 --- a/docs/en/PyTorch Operator Development Guide/PyTorch Operator Development Guide.md +++ b/docs/en/PyTorch Operator Development Guide/PyTorch Operator Development Guide.md @@ -444,13 +444,13 @@ You can develop an operator adaptation plugin to convert the formats of the inpu 3. Define the main adaptation function of the operator. - Determine the adaptation theme function for custom operators based on the dispatch function in the registered operator. + Determine the main adaptation function for custom operators based on the dispatch function in the registered operator. 4. Implement the main adaptation functions. - Implement the operator adaptation theme function and construct the corresponding input, output, and attributes based on the TBE operator prototype. + Implement the operator's main adaptation function and construct the corresponding input, output, and attributes based on the TBE operator prototype. -5. Use the **TORCH\_LIBRARY\_IMPL** macro to associate the operator description func in the **native\_functions.yaml** file generated during the operator registration. \(Only PyTorch 1.8.1 requires this step.\) +5. \(Only PyTorch 1.8.1 requires this step.\) Use the **TORCH\_LIBRARY\_IMPL** macro to associate the operator description func in the **native\_functions.yaml** file generated during the operator registration. **TORCH\_LIBRARY\_IMPL** is a macro provided by PyTorch for registered operator distribution. To use it, perform the following steps: @@ -618,7 +618,7 @@ The following uses the torch.add\(\) operator as an example to describe how to a } ``` -5. Use the **TORCH\_LIBRARY\_IMPL** macro to associate the registered operator. \(Only PyTorch 1.8.1 requires this step.\) +5. \(Only PyTorch 1.8.1 requires this step.\) Use the **TORCH\_LIBRARY\_IMPL** macro to associate the registered operator. ``` TORCH_LIBRARY_IMPL(aten, NPU, m) { @@ -827,7 +827,7 @@ pip3.7 install torchvision --no-deps #### Symptom -During the installation of **torch-**_\*_**.whl**, the message "ERROR: torchvision 0.6.0 has requirement torch==1.5.0, but you'll have torch 1.5.0a0+1977093 which is incompatible" " is displayed. +During the installation of **torch-**_\*_**.whl**, the message "ERROR: torchvision 0.6.0 has requirement torch==1.5.0, but you'll have torch 1.5.0a0+1977093 which is incompatible" is displayed. ![](figures/en-us_image_0000001144082048.png) @@ -900,7 +900,7 @@ The custom TBE operator has been developed and adapted to PyTorch. However, the There should be no error in this step. The log added in **add** should be displayed. If an error occurs, check the code to ensure that no newly developed code affects the test. - 3. The newly developed custom TBE operator is combined into CANN. Logs are added to the operator entry as the running identifier. + 3. Combine the newly developed custom TBE operator into CANN. Add logs to the operator entry as the running identifier. 4. After the compilation and installation of CANN are complete, call **python3.7.5 test\_add.py** to perform the test. >![](public_sys-resources/icon-note.gif) **NOTE:** @@ -1047,7 +1047,7 @@ The following describes how to upgrade CMake to 3.12.1. ln -s /usr/local/cmake/bin/cmake /usr/bin/cmake ``` -5. Run the following command to check whether CMake has been installed: +5. Check whether CMake has been installed. ``` cmake --version diff --git "a/docs/zh/\346\224\257\346\214\201ONNX\347\256\227\345\255\220\346\270\205\345\215\225/\346\224\257\346\214\201ONNX\347\256\227\345\255\220\346\270\205\345\215\225.md" "b/docs/zh/\346\224\257\346\214\201ONNX\347\256\227\345\255\220\346\270\205\345\215\225/\346\224\257\346\214\201ONNX\347\256\227\345\255\220\346\270\205\345\215\225.md" index 2c606fdd00673c1876bddcd3cc3e13de476e5a29..298c52f3bd607673ef2105d431b9d45706406139 100644 --- "a/docs/zh/\346\224\257\346\214\201ONNX\347\256\227\345\255\220\346\270\205\345\215\225/\346\224\257\346\214\201ONNX\347\256\227\345\255\220\346\270\205\345\215\225.md" +++ "b/docs/zh/\346\224\257\346\214\201ONNX\347\256\227\345\255\220\346\270\205\345\215\225/\346\224\257\346\214\201ONNX\347\256\227\345\255\220\346\270\205\345\215\225.md" @@ -220,68 +220,6 @@ y：一个tensor，数据类型和shape与输入一致 Opset v9/v10/v11/v12/v13 -

AdaptiveAvgPool2D

- -### 功能 - -对输入进行2d自适应平均池化计算 - -### 边界 - -【输入】 - -一个输入 - -x：一个tensor，数据类型：float16、float32 - -【属性】 - -一个属性： - -output\_size：int型数组，指定输出的hw的shape大小 - -【输出】 - -一个输出 - -y：一个tensor，数据类型：与x类型一致 - -### 支持的ONNX版本 - -自定义算子，无对应onnx版本 - -

AdaptiveMaxPool2D

- -### 功能 - -对输入进行2d自适应最大池化计算 - -### 边界 - -【输入】 - -一个输入 - -x：一个tensor，数据类型：float16、float32、float64 - -【属性】 - -一个属性： - -output\_size：int型数组，指定输出的hw的shape大小 - -【输出】 - -两个输出 - -y：一个tensor，数据类型：与x类型一致 - -argmax：一个tensor，数据类型：int32，int64 - -### 支持的ONNX版本 - -自定义算子，无对应onnx版本 -

Add

### 功能 @@ -306,36 +244,6 @@ C：一个张量，数据类型与A相同 Opset v8/v9/v10/v11/v12/v13 -

Addcmul

- -### 功能 - -元素级计算\(x1 \* x2\) \* value + input\_data - -### 边界 - -【输入】 - -四个输入 - -input\_data：一个tensor，数据类型：float16、float32、int32、int8、uint8 - -x1： 一个tensor，类型与input\_data相同 - -x2： 一个tensor，类型与input\_data相同 - -value： 一个tensor，类型与input\_data相同 - -【输出】 - -一个输出 - -y：一个tensor，数据类型：y与输入相同 - -### 支持的ONNX版本 - -自定义算子，无对应onnx版本 -

AffineGrid

### 功能 @@ -3569,38 +3477,6 @@ reshaped：一个张量 Opset v8/v9/v10/v11/v12/v13 -

ReverseSequence

- -### 功能 - -根据指定长度对batch序列进行排序 - -### 边界 - -【输入】 - -2个输入 - -x：tensor，rank \>= 2，数据类型：float16、float32 - -sequence\_lens：tensor，每个batch的指定长度，数据类型：int64 - -【输出】 - -一个输出 - -y：一个张量，和输入x同样的type和shape - -【属性】 - -batch\_axis：int，默认为1，含义：指定batch轴 - -time\_axis：int，默认为1，含义：指定time轴 - -### 支持的ONNX版本 - -Opset v10/v11/v12/v13 -

RoiExtractor

### 功能