未验证 提交 1d6a6f3e 编写于 作者: X Xiaoyao Xi 提交者: GitHub

Update README.md

上级 9e8490b4
...@@ -135,7 +135,17 @@ if __name__ == '__main__': ...@@ -135,7 +135,17 @@ if __name__ == '__main__':
controller.train() controller.train()
``` ```
默认情况下每5个训练step打印一次训练日志,如下(该日志在8卡P40机器上运行得到),可以看到loss值随着训练收敛。在训练结束后,`Controller`自动为mrqa任务保存预测模型。
```
Global step: 5. Task: mrqa, step 5/13 (epoch 0), loss: 4.976, speed: 0.11 steps/s
Global step: 10. Task: mrqa, step 10/13 (epoch 0), loss: 2.938, speed: 0.48 steps/s
Global step: 15. Task: mrqa, step 2/13 (epoch 1), loss: 2.422, speed: 0.47 steps/s
Global step: 20. Task: mrqa, step 7/13 (epoch 1), loss: 2.809, speed: 0.53 steps/s
Global step: 25. Task: mrqa, step 12/13 (epoch 1), loss: 1.744, speed: 0.50 steps/s
mrqa: train finished!
mrqa: inference model saved at output_model/firstrun/infer_model
```
## DEMO2:多任务训练与推理 ## DEMO2:多任务训练与推理
......
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