AI4Math-ShanZhang comments

Repositories
Issues
Comments

Results 4 comments of


                                            AI4Math-ShanZhang

Got a nan loss and gradient norm when training swin-l on imagenet22k with O1

hi, did you solve the issue? i also encountered the satiation where the grad_norm.avg is nan, but other items are ok (loss and grad.val)

mismatch owing to pretrain model_final.pkl.

thanks for your reply, when i evaluate via tools/test_net.py, there is keyerror: devkit_directory. after checking, i found lossing the index of fsod dataset. so could you advise me the information...

There are 2 questions when I run your codes

> I have solved the second problem. > It is due to the version of cuda. I use cuda10.0, but pytorch0.4.1 uses cuda 9.0. Hi, i encountered the same question...

The training log and the query embedding

> Thank you for reply! I agree with you that the query embeddings encoded by nn.Embedding operation which would contain base classes information if there is no fine-tuning step and...