"testing/python/kernel/test_tilelang_int4_mma_matmul.py" did not exist on "64f17c2f369e612cc297d358f607307a615bbb59"
Commit 555464ee authored by unknown's avatar unknown
Browse files

修改mmseg train.md

parent a024b5d7
_base_ = '../faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py'
model = dict(
roi_head=dict(
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(
_delete_=True,
type='ModulatedDeformRoIPoolPack',
output_size=7,
output_channels=256),
out_channels=256,
featmap_strides=[4, 8, 16, 32])))
...@@ -23,15 +23,15 @@ pip3 install opencv-python ...@@ -23,15 +23,15 @@ pip3 install opencv-python
### 安装python依赖包 ### 安装python依赖包
```python ```python
pip3 install torch-1.10.0a0+gitcc7c9c7-cp36-cp36m-linux_x86_64.whl -i https://pypi.tuna.tsinghua.edu.cn/simple pip3 install torch-1.10.0a0+git2040069.dtk2210-cp37-cp37m-manylinux2014_x86_64.whl -i https://pypi.tuna.tsinghua.edu.cn/simple
pip3 install torchvision-0.10.0a0+300a8a4-cp36-cp36m-linux_x86_64.whl -i https://pypi.tuna.tsinghua.edu.cn/simple pip3 install torchvision-0.10.0a0+e04d001.dtk2210-cp37-cp37m-manylinux2014_x86_64.whl -i https://pypi.tuna.tsinghua.edu.cn/simple
pip3 install mmcv_full-1.3.16-cp36-cp36m-linux_x86_64.whl -i https://pypi.tuna.tsinghua.edu.cn/simple pip3 install mmcv_full-1.6.1+gitdebbc80.dtk2210-cp37-cp37m-manylinux2014_x86_64.whl -i https://pypi.tuna.tsinghua.edu.cn/simple
mmpose安装 mmseg 安装
cd mmsegmentation cd mmsegmentation-0.29.1
pip3 install -e . pip3 install -e .
``` ```
注:测试不同版本的dtk,需安装对应版本的库whl包,dtk22.04.1使用python3.7,dtk21.10.1使用python3.6,如果测试优化后的版本,需要设置export HIP_UPSAMPLE_OPTIMIZE=1 注:测试不同版本的dtk,需安装对应版本的库whl包,如果测试优化后的版本,需要设置export HIP_UPSAMPLE_OPTIMIZE=1
## PSPNet R50测试 ## PSPNet R50测试
...@@ -41,101 +41,38 @@ pip3 install -e . ...@@ -41,101 +41,38 @@ pip3 install -e .
./sing_test.sh configs/pspnet/pspnet_r50-d8_512x1024_40k_cityscapes.py ./sing_test.sh configs/pspnet/pspnet_r50-d8_512x1024_40k_cityscapes.py
``` ```
若使用dtk22.04.1测试,需设置 `export MIOPEN_FIND_MODE=1,export MIOPEN_USE_APPROXIMATE_PERFORMANCE=0`以下测试均相同.
#### 参数说明 #### 参数说明
configs/_base_/datasets/cityscapes.py中batch_size=samples_per_gpu*卡数,性能计算方法:batch_size/time configs/_base_/datasets/cityscapes.py中batch_size=samples_per_gpu*卡数,性能计算方法:batch_size/time
![1659172384601](image/train/1659172384601.png)
#### 性能关注:time #### 性能关注:time
### 多卡测试(单精度) ### 多卡测试(单精度)
```python #### 单机多卡训练
./multi_test.sh configs/pspnet/pspnet_r50-d8_512x1024_40k_cityscapes.py
```
## DeepLabV3 R50测试
### 单卡测试(单精度) 1.pytorch单机多卡训练
```python ```python
./sing_test.sh configs/deeplabv3/deeplabv3_r50-d8_512x1024_40k_cityscapes.py ./multi_test.sh configs/pspnet/pspnet_r50-d8_512x1024_40k_cityscapes.py
```
#### 参数说明
configs/_base_/datasets/cityscapes.py中batch_size=samples_per_gpu*卡数,性能计算方法:batch_size/time
![1659172468356](image/train/1659172468356.png)
#### 性能关注:time
### 多卡测试(单精度)
```python
./multi_test.sh configs/deeplabv3/deeplabv3_r50-d8_512x1024_40k_cityscapes.py
```
## FCN R50测试
### 单卡测试(单精度)
```python
./sing_test.sh configs/fcn/fcn_r50-d8_512x1024_40k_cityscapes.py
```
#### 参数说明
configs/_base_/datasets/cityscapes.py中batch_size=samples_per_gpu*卡数,性能计算方法:batch_size/time
![1659172530175](image/train/1659172530175.png)
#### 性能关注:time
### 多卡测试(单精度)
```python
./multi_test.sh configs/fcn/fcn_r50-d8_512x1024_40k_cityscapes.py
```
## UperNet R50测试
### 单卡测试(单精度)
```python
./sing_test.sh configs/upernet/upernet_r50_512x1024_40k_cityscapes.py
```
#### 参数说明
configs/_base_/datasets/cityscapes.py中batch_size=samples_per_gpu*卡数,性能计算方法:batch_size/time
![1659172631957](image/train/1659172631957.png)
#### 性能关注:time
### 多卡测试(单精度)
```python
./multi_test.sh configs/upernet/upernet_r50_512x1024_40k_cityscapes.py
``` ```
## DeepLabV3plus_R50测试 #### 多机多卡训练
### 单卡测试(单精度) 1.pytorch多机多卡训练 在第一台机器上: NODES=2 NODE_RANK=0 PORT=12345 MASTER_ADDR=10.1.3.56 sh tools/dist_train.sh configs/pspnet/pspnet_r50-d8_512x1024_40k_cityscapes.py 4 在第二台机器上: NODES=2 NODE_RANK=1 PORT=12345 MASTER_ADDR=10.1.3.56 sh tools/dist_train.sh configs/pspnet/pspnet_r50-d8_512x1024_40k_cityscapes.py 4
```python ### 半精度测试
./sing_test.sh configs/deeplabv3plus/deeplabv3plus_r50-d8_512x1024_40k_cityscapes.py
```
#### 参数说明 修改configs文件,添加fp16 = dict(loss_scale=512.),单机多卡和多机多卡测试与单精度测试方法相同。
configs/_base_/datasets/cityscapes.py中batch_size=samples_per_gpu*卡数,性能计算方法:batch_size/time ### 其他模型测试
![1659172696867](image/train/1659172696867.png)
#### 性能关注:time 其他模型的测试步骤和pspnet_r50相同,只需修改对应的config文件即可,下面列出相关模型对应的config文件列表:
### 多卡测试(单精度) | 模型 | configs |
| ----------------- | ------------------------------------------------------------ |
| DeepLabV3 R50 | configs/deeplabv3/deeplabv3_r50-d8_512x1024_40k_cityscapes.py |
| FCN R50 | configs/fcn/fcn_r50-d8_512x1024_40k_cityscapes.py |
| UperNet R50 | configs/upernet/upernet_r50_512x1024_40k_cityscapes.py |
| DeepLabV3plus_R50 | configs/deeplabv3plus/deeplabv3plus_r50-d8_512x1024_40k_cityscapes.py |
```python
./multi_test.sh configs/deeplabv3plus/deeplabv3plus_r50-d8_512x1024_40k_cityscapes.py
```
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment