search

microsoft / unilm

Large-scale Self-supervised Pre-training Across Tasks, Languages, and Modalities
https://aka.ms/GeneralAI
MIT License
19.61k stars 2.5k forks source link

[textdiffuser-2] where to set the loss type during training? #1563

Closed shuyueW1991 closed 4 months ago

shuyueW1991 commented 4 months ago

hello, I am trying to use textdiffuser-2. When I am in the stage of M1's layout planning training, I found that there is no flag for setting the type of loss, whereas in the paper the author said such training requires loss type being cross-entropy. How come to set this configuration? and by the way, how can I monitor the training loss curve like the one in readme.md?

JingyeChen commented 4 months ago

Thanks for your attention to TextDiffuser-2. Please dive into the Transformer Trainer code. The default loss is the cross entropy.

shuyueW1991 commented 4 months ago

Thanks for your attention to TextDiffuser-2. Please dive into the Transformer Trainer code. The default loss is the cross entropy.

thanks for you tips. i thought i missed some cricial argumental code haha.