update flops compute for linear module

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: Major changes

In tranformer-type model, we often use linear layer to perform matrix-matrix multiplication A * B where A has shape (B, L, C1) and B is the weight matrix with shape (C1, C2). The current profiler will ignore the size L. Fix it.
A minor logging fix to dataloader_limit_wrapper.py.

Differential Revision: D28172235

facebook-github-bot commented 3 years ago

This pull request was exported from Phabricator. Differential Revision: D28172235

facebook-github-bot commented 3 years ago

This pull request was exported from Phabricator. Differential Revision: D28172235

facebook-github-bot commented 3 years ago

This pull request has been merged in facebookresearch/ClassyVision@93b78ea45502c58f23882d3b764a65be3d9c209f.