8000 Use `torch.distributed.checkpoint.state_dict.set_model_state_dict` in `load_checkpoint_in_model` by ringohoffman · Pull Request #3432 · huggingface/accelerate · GitHub

More Web Proxy on the site http://driver.im/

Use `torch.distributed.checkpoint.state_dict.set_model_state_dict` in `load_checkpoint_in_model`#3432

Merged

SunMarc merged 15 commits intohuggingface:mainfrom

ringohoffman:load_checkpoint_in_model-use-set_model_state_dict

Apr 11, 2025

Commits on Mar 10, 2025

Use torch.distributed.checkpoint.state_dict.set_model_state_dict in load_checkpoint_in_model
Matthew Hoffman
committed

Commits on Mar 11, 2025

Guard torch.distributed.checkpoint.state_dict with is_torch_version('>=', '2.2.0')
Matthew Hoffman
committed
Add test for non-distributed, TP, and DDP for load_checkpoint_and_dispatch(device_map=None) using set_model_state_dict
Matthew Hoffman
committed
Merge branch 'main' into load_checkpoint_in_model-use-set_model_state_dict
Matthew Hoffman
committed

Commits on Mar 12, 2025

Verify minimum version for broadcast_from_rank0
Matthew Hoffman
committed
Mark transformers as required for broadcast_from_rank0 tests, mark min version of torch to test as 2.4.0
Matthew Hoffman
committed

Commits on Mar 13, 2025

Add model_devices guard to set_model_state_dict
Matthew Hoffman
committed
Move decorators to top of test class
Matthew Hoffman
committed
Unindent functions
Matthew Hoffman
committed
Add condition for w/ explanatory links for set_model_state_dict model device restrictions
Matthew Hoffman
committed
Fix distribution of 2.2.0 condition
Matthew Hoffman
committed

Commits on Apr 1, 2025

Remove tensor parallel test
Matthew Hoffman
committed

Commits on Apr 2, 2025

Fix model materialization example
Matthew Hoffman
committed
Fix materialization example
Matthew Hoffman
committed
Remove old tensor parallel test
Matthew Hoffman
committed

0