提交 dcd0acba 编写于 作者: S shippingwang

add visualdl doc

上级 f3b2e8aa
...@@ -41,7 +41,8 @@ python -m paddle.distributed.launch \ ...@@ -41,7 +41,8 @@ python -m paddle.distributed.launch \
--selected_gpus="0,1,2,3" \ --selected_gpus="0,1,2,3" \
tools/train.py \ tools/train.py \
-c ./configs/ResNet/ResNet50_vd.yaml \ -c ./configs/ResNet/ResNet50_vd.yaml \
-o use_mix=1 -o use_mix=1 \
--vdl_dir=./scalar/
``` ```
...@@ -53,6 +54,13 @@ epoch:0 train step:522 loss:1.6330 lr:0.100000 elapse:0.210 ...@@ -53,6 +54,13 @@ epoch:0 train step:522 loss:1.6330 lr:0.100000 elapse:0.210
也可以直接修改模型对应的配置文件更新配置。具体配置参数参考[配置文档](config.md) 也可以直接修改模型对应的配置文件更新配置。具体配置参数参考[配置文档](config.md)
训练期间可以通过VisualDL实时观察loss变化,启动命令如下:
```bash
visualdl --logdir ./scalar --host <host_IP> --port <port_num>
```
### 2.2 模型微调 ### 2.2 模型微调
......
...@@ -82,6 +82,13 @@ def error(fmt, *args): ...@@ -82,6 +82,13 @@ def error(fmt, *args):
def scaler(name, value, step, writer): def scaler(name, value, step, writer):
"""
This function will draw a scalar curve generated by the visualdl.
Usage: Install visualdl: pip3 install visualdl==2.0.0b4
and then:
visualdl --logdir ./scalar --host 0.0.0.0 --port 8830
to preview loss corve in real time.
"""
writer.add_scalar(name, value, step) writer.add_scalar(name, value, step)
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册