Tongzhou Wang comments

Repositories
Issues
Comments

Results 15 comments of


                                            Tongzhou Wang

Needs updating for PyTorch HEAD (no_grad)

We probably should start a branch with updated scripts as people are starting to compile from source by themselves.

Filling zero after `optimizer.step()`

FYI, it is fixed today. But it is only fixed on master.

dist2plane scaled missing lambda

@ferrine That makes sense. However, should doc be updated to reflect this?

Cannot re-initialize CUDA in forked subprocess

Yeah, you should just use spawn @gautamkmr . Closing since this looks like user error.

Cannot re-initialize CUDA in forked subprocess

@gautamkmr `set_start_method('spawn')` should be inside `if __name__ == '__main__'`. It really is not relevant to `launch.py` but just how you should use multiprocessing data loader with CUDA tensors. Alternatively, you...