Puyuan Liu

Results 24 comments of Puyuan Liu

The reward in step3 seems to be completely random without any noticeable increase.

My rewards seems even decrasing, despite the decrease in loss ![W B Chart 07_05_2023, 17_15_56](https://user-images.githubusercontent.com/121119211/236707391-668847fa-8d40-463f-b141-4949c146a31f.png) ![W B Chart 07_05_2023, 17_15_50](https://user-images.githubusercontent.com/121119211/236707393-75a54eec-ef23-433e-9ad9-8fee0c471620.png) ![W B Chart 07_05_2023, 17_15_13](https://user-images.githubusercontent.com/121119211/236707394-7f18f87b-536c-41e8-bc5f-e28d5515ec0a.png)

The reward in step3 seems to be completely random without any noticeable increase.

> @puyuanOT OK i got the solution. Try to disable the hybirdengine, this make the model always repeat 'a a a a a' not sure the reason. Thanks a lot!...

The reward in step3 seems to be completely random without any noticeable increase.

Perhaps it's related to this PR https://github.com/microsoft/DeepSpeedExamples/pull/470?

Resizing model embedding when loading the model

Looking forward to any replies. This resizing not only leads to the warning `You are resizing the embedding layer without providing a `pad_to_multiple_of` parameter. This means that the new embedding...