add reduction mode none in soft_target_cross_entropy_loss

facebookresearch / ClassyVision

An end-to-end PyTorch framework for image and video classification

https://classyvision.ai

MIT License

1.59k stars 278 forks source link

Closed stephenyan1231 closed 3 years ago

stephenyan1231 commented 3 years ago

Summary: To support customized per-sampling weighting during loss computation, allow reduction mode to be none in soft_target_cross_entropy_loss.

Differential Revision: D27035879

facebook-github-bot commented 3 years ago

This pull request was exported from Phabricator. Differential Revision: D27035879