add train_amp doc
HydrogenSulfate
parent
7494441082
commit
5f1c265a69
|
|
@ -0,0 +1,115 @@
|
|||
# Linux GPU/CPU 混合精度训练推理测试
|
||||
|
||||
Linux GPU/CPU 混合精度训练推理测试的主程序为`test_train_inference_python.sh`，可以测试基于Python的模型混合精度(默认O2)训练、评估、推理等基本功能。
|
||||
|
||||
## 1. 测试结论汇总
|
||||
|
||||
- 训练相关：
|
||||
|
||||
| 算法名称 | 模型名称 | 单机单卡 | 单机多卡 |
|
||||
| :-------------: | :-------------------------------------: | :----------: | :----------: |
|
||||
| MobileNetV3 | MobileNetV3_large_x1_0 | 混合精度训练 | 混合精度训练 |
|
||||
| PP-ShiTu | GeneralRecognition_PPLCNet_x2_5 | 混合精度训练 | 混合精度训练 |
|
||||
| PPHGNet | PPHGNet_small | 混合精度训练 | 混合精度训练 |
|
||||
| PPHGNet | PPHGNet_tiny | 混合精度训练 | 混合精度训练 |
|
||||
| PPLCNet | PPLCNet_x0_25 | 混合精度训练 | 混合精度训练 |
|
||||
| PPLCNet | PPLCNet_x0_35 | 混合精度训练 | 混合精度训练 |
|
||||
| PPLCNet | PPLCNet_x0_5 | 混合精度训练 | 混合精度训练 |
|
||||
| PPLCNet | PPLCNet_x0_75 | 混合精度训练 | 混合精度训练 |
|
||||
| PPLCNet | PPLCNet_x1_0 | 混合精度训练 | 混合精度训练 |
|
||||
| PPLCNet | PPLCNet_x1_5 | 混合精度训练 | 混合精度训练 |
|
||||
| PPLCNet | PPLCNet_x2_0 | 混合精度训练 | 混合精度训练 |
|
||||
| PPLCNet | PPLCNet_x2_5 | 混合精度训练 | 混合精度训练 |
|
||||
| PPLCNetV2 | PPLCNetV2_base | 混合精度训练 | 混合精度训练 |
|
||||
| ResNet | ResNet50 | 混合精度训练 | 混合精度训练 |
|
||||
| ResNet | ResNet50_vd | 混合精度训练 | 混合精度训练 |
|
||||
| SwinTransformer | SwinTransformer_tiny_patch4_window7_224 | 混合精度训练 | 混合精度训练 |
|
||||
|
||||
- 推理相关：
|
||||
|
||||
| 算法名称 | 模型名称 | device_CPU | device_GPU | batchsize |
|
||||
| :-------------: | :-------------------------------------: | :--------: | :--------: | :-------: |
|
||||
| MobileNetV3 | MobileNetV3_large_x1_0 | 支持 | 支持 | 1 |
|
||||
| PP-ShiTu | GeneralRecognition_PPLCNet_x2_5 | 支持 | 支持 | 1 |
|
||||
| PPHGNet | PPHGNet_small | 支持 | 支持 | 1 |
|
||||
| PPHGNet | PPHGNet_tiny | 支持 | 支持 | 1 |
|
||||
| PPLCNet | PPLCNet_x0_25 | 支持 | 支持 | 1 |
|
||||
| PPLCNet | PPLCNet_x0_35 | 支持 | 支持 | 1 |
|
||||
| PPLCNet | PPLCNet_x0_5 | 支持 | 支持 | 1 |
|
||||
| PPLCNet | PPLCNet_x0_75 | 支持 | 支持 | 1 |
|
||||
| PPLCNet | PPLCNet_x1_0 | 支持 | 支持 | 1 |
|
||||
| PPLCNet | PPLCNet_x1_5 | 支持 | 支持 | 1 |
|
||||
| PPLCNet | PPLCNet_x2_0 | 支持 | 支持 | 1 |
|
||||
| PPLCNet | PPLCNet_x2_5 | 支持 | 支持 | 1 |
|
||||
| PPLCNetV2 | PPLCNetV2_base | 支持 | 支持 | 1 |
|
||||
| ResNet | ResNet50 | 支持 | 支持 | 1 |
|
||||
| ResNet | ResNet50_vd | 支持 | 支持 | 1 |
|
||||
| SwinTransformer | SwinTransformer_tiny_patch4_window7_224 | 支持 | 支持 | 1 |
|
||||
|
||||
## 2. 测试流程
|
||||
|
||||
以下测试流程以 PPLCNet_x1_0 模型为例。
|
||||
|
||||
### 2.1 准备环境
|
||||
|
||||
- 安装PaddlePaddle：如果您已经安装了2.2或者以上版本的paddlepaddle，那么无需运行下面的命令安装paddlepaddle。
|
||||
```bash
|
||||
# 需要安装2.2及以上版本的Paddle
|
||||
# 安装GPU版本的Paddle
|
||||
python3.7 -m pip install paddlepaddle-gpu==2.2.0
|
||||
# 安装CPU版本的Paddle
|
||||
python3.7 -m pip install paddlepaddle==2.2.0
|
||||
```
|
||||
|
||||
- 安装PaddleSlim
|
||||
```bash
|
||||
python3.7 -m pip install paddleslim==2.2.0
|
||||
```
|
||||
|
||||
- 安装依赖
|
||||
```bash
|
||||
python3.7 -m pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple
|
||||
```
|
||||
|
||||
- 安装AutoLog（规范化日志输出工具）
|
||||
```bash
|
||||
python3.7 -m pip install https://paddleocr.bj.bcebos.com/libs/auto_log-1.2.0-py3-none-any.whl
|
||||
```
|
||||
|
||||
### 2.2 准备数据和模型
|
||||
|
||||
```bash
|
||||
bash test_tipc/prepare.sh test_tipc/configs/PPLCNet/PPLCNet_x1_0_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt lite_train_lite_infer
|
||||
```
|
||||
### 2.3 功能测试
|
||||
|
||||
测试方法如下所示，希望测试不同的模型文件，只需更换为自己的参数配置文件，即可完成对应模型的测试。
|
||||
|
||||
```bash
|
||||
bash test_tipc/test_train_inference_python.sh ${your_params_file} lite_train_lite_infer
|
||||
```
|
||||
|
||||
以`PPLCNet_x1_0`的`Linux GPU/CPU 混合精度训练推理测试`为例，命令如下所示。
|
||||
|
||||
```bash
|
||||
bash test_tipc/test_train_inference_python.sh test_tipc/configs/PPLCNet/PPLCNet_x1_0_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt lite_train_lite_infer
|
||||
```
|
||||
|
||||
输出结果如下，表示命令运行成功。
|
||||
|
||||
```log
|
||||
Run successfully with command - PPLCNet_x1_0 - python3.7 tools/train.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x1_0.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2 -o Optimizer.multi_precision=True -o Global.eval_during_train=False -o Global.device=gpu -o Global.output_dir=./test_tipc/output/PPLCNet_x1_0/lite_train_lite_infer/amp_train_gpus_0_autocast_null -o Global.epochs=2 -o DataLoader.Train.sampler.batch_size=8 !
|
||||
Run successfully with command - PPLCNet_x1_0 - python3.7 tools/eval.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x1_0.yaml -o Global.pretrained_model=./test_tipc/output/PPLCNet_x1_0/lite_train_lite_infer/amp_train_gpus_0_autocast_null/PPLCNet_x1_0/latest -o Global.device=gpu !
|
||||
Run successfully with command - PPLCNet_x1_0 - python3.7 tools/export_model.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x1_0.yaml -o Global.pretrained_model=./test_tipc/output/PPLCNet_x1_0/lite_train_lite_infer/amp_train_gpus_0_autocast_null/PPLCNet_x1_0/latest -o Global.save_inference_dir=./test_tipc/output/PPLCNet_x1_0/lite_train_lite_infer/amp_train_gpus_0_autocast_null!
|
||||
Run successfully with command - PPLCNet_x1_0 - python3.7 python/predict_cls.py -c configs/inference_cls.yaml -o Global.use_gpu=True -o Global.use_tensorrt=False -o Global.use_fp16=False -o Global.inference_model_dir=.././test_tipc/output/PPLCNet_x1_0/lite_train_lite_infer/amp_train_gpus_0_autocast_null -o Global.batch_size=1 -o Global.infer_imgs=../dataset/ILSVRC2012/val -o Global.benchmark=False > .././test_tipc/output/PPLCNet_x1_0/lite_train_lite_infer/infer_gpu_usetrt_False_precision_False_batchsize_1.log 2>&1 !
|
||||
Run successfully with command - PPLCNet_x1_0 - python3.7 python/predict_cls.py -c configs/inference_cls.yaml -o Global.use_gpu=False -o Global.enable_mkldnn=False -o Global.cpu_num_threads=6 -o Global.inference_model_dir=.././test_tipc/output/PPLCNet_x1_0/lite_train_lite_infer/amp_train_gpus_0_autocast_null -o Global.batch_size=1 -o Global.infer_imgs=../dataset/ILSVRC2012/val -o Global.benchmark=False > .././test_tipc/output/PPLCNet_x1_0/lite_train_lite_infer/infer_cpu_usemkldnn_False_threads_6_batchsize_1.log 2>&1 !
|
||||
Run successfully with command - PPLCNet_x1_0 - python3.7 -m paddle.distributed.launch --gpus=0,1 tools/train.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x1_0.yaml -o Global.seed=1234 -o DataLoader.Train.sampler.shuffle=False -o DataLoader.Train.loader.num_workers=0 -o DataLoader.Train.loader.use_shared_memory=False -o AMP.scale_loss=65536 -o AMP.use_dynamic_loss_scaling=True -o AMP.level=O2 -o Optimizer.multi_precision=True -o Global.eval_during_train=False -o Global.device=gpu -o Global.output_dir=./test_tipc/output/PPLCNet_x1_0/lite_train_lite_infer/amp_train_gpus_0,1_autocast_null -o Global.epochs=2 -o DataLoader.Train.sampler.batch_size=8 !
|
||||
Run successfully with command - PPLCNet_x1_0 - python3.7 tools/eval.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x1_0.yaml -o Global.pretrained_model=./test_tipc/output/PPLCNet_x1_0/lite_train_lite_infer/amp_train_gpus_0,1_autocast_null/PPLCNet_x1_0/latest -o Global.device=gpu !
|
||||
Run successfully with command - PPLCNet_x1_0 - python3.7 tools/export_model.py -c ppcls/configs/ImageNet/PPLCNet/PPLCNet_x1_0.yaml -o Global.pretrained_model=./test_tipc/output/PPLCNet_x1_0/lite_train_lite_infer/amp_train_gpus_0,1_autocast_null/PPLCNet_x1_0/latest -o Global.save_inference_dir=./test_tipc/output/PPLCNet_x1_0/lite_train_lite_infer/amp_train_gpus_0,1_autocast_null!
|
||||
Run successfully with command - PPLCNet_x1_0 - python3.7 python/predict_cls.py -c configs/inference_cls.yaml -o Global.use_gpu=True -o Global.use_tensorrt=False -o Global.use_fp16=False -o Global.inference_model_dir=.././test_tipc/output/PPLCNet_x1_0/lite_train_lite_infer/amp_train_gpus_0,1_autocast_null -o Global.batch_size=1 -o Global.infer_imgs=../dataset/ILSVRC2012/val -o Global.benchmark=False > .././test_tipc/output/PPLCNet_x1_0/lite_train_lite_infer/infer_gpu_usetrt_False_precision_False_batchsize_1.log 2>&1 !
|
||||
Run successfully with command - PPLCNet_x1_0 - python3.7 python/predict_cls.py -c configs/inference_cls.yaml -o Global.use_gpu=False -o Global.enable_mkldnn=False -o Global.cpu_num_threads=6 -o Global.inference_model_dir=.././test_tipc/output/PPLCNet_x1_0/lite_train_lite_infer/amp_train_gpus_0,1_autocast_null -o Global.batch_size=1 -o Global.infer_imgs=../dataset/ILSVRC2012/val -o Global.benchmark=False > .././test_tipc/output/PPLCNet_x1_0/lite_train_lite_infer/infer_cpu_usemkldnn_False_threads_6_batchsize_1.log 2>&1 !
|
||||
```
|
||||
|
||||
该信息可以在运行log中查看，以`PPLCNet_x1_0`为例，log位置在`./test_tipc/output/PPLCNet_x1_0/lite_train_lite_infer/results_python.log`。
|
||||
|
||||
如果运行失败，也会在终端中输出运行失败的日志信息以及对应的运行命令。可以基于该命令，分析运行失败的原因。
|
||||
|
|
@ -49,7 +49,7 @@ Linux GPU/CPU KL离线量化推理测试的主程序为`test_ptq_inference_pytho
|
|||
|
||||
## 2. 测试流程
|
||||
|
||||
一下测试流程以 MobileNetV3_large_x1_0 模型为例。
|
||||
以下测试流程以 MobileNetV3_large_x1_0 模型为例。
|
||||
|
||||
### 2.1 准备环境
|
||||
|
||||
|
|
|
|||
Loading…
Reference in New Issue