Chien-Chin Huang comments

Results 119 comments of


                                            Chien-Chin Huang

[FSDP][optim_state_dict][10/N] Make optim_state_dict and optim_state_dict_to_load public

@pytorchbot merge

[FSDP][optim_state_dict][10/N] Make optim_state_dict and optim_state_dict_to_load public

@pytorchbot merge -f "The failing test is not related."

[FSDP][optim_state_dict][10/N] Make optim_state_dict and optim_state_dict_to_load public

@pytorchbot merge

Fix an import loop

@Skylion007 It's still there according to https://docs.python.org/3.13/library/__future__.html. And the doc show the original plan is to make it mandatory in 3.11 but the decision was postponed in definitely. But no...

Fix an import loop

@Skylion007 Thanks for the information.

[RFC][inductor] Add collective function fusion: AllReduce + Split to ReduceScatter

cc., @yifuwang

Fix default world_size when running on 1 or 0 GPU

@pytorchbot merge

Use stateful dataloader to checkpoint data iteration order and token buffer

@gokulavasan, @tianyu-l https://github.com/pytorch/pytorch/pull/125335 and https://github.com/pytorch/pytorch/pull/125334 should unblock this PR.

[DCP] Provides default AsyncStager

The PR looks good to me. The BC linter may be flaky due to the inheritance change but it is good to double check. There is also a linter issue.

[DSD] Fix to remove non_persistent buffer in distributed state dict

@pytorchbot merge -f "The failing tests are not related."