Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

HF Checkpoint OOM fix with earlier dtype conversion #1380

Closed
wants to merge 1 commit into from
Closed

HF Checkpoint OOM fix with earlier dtype conversion #1380

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 8 additions & 10 deletions llmfoundry/callbacks/hf_checkpointer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -442,19 +442,22 @@ def dtensor_to_tensor_hook(
prefix: str,
*args: Any,
) -> Dict[str, Any]:
dtensor_fqns = []
for fqn in state_dict.keys():
tensor = state_dict[fqn]
if isinstance(tensor, DTensor):
dtensor_fqns.append(fqn)
tensor = tensor.full_tensor() # type: ignore
if dist.get_global_rank() == 0:
if cpu_offload:
tensor = tensor.cpu()
state_dict[fqn] = tensor
else:
state_dict[fqn] = None
# Convert the state dict to the requested precision
if isinstance(tensor, torch.Tensor):
state_dict[fqn] = tensor.to(dtype=self.dtype)
del tensor
if dist.get_global_rank() != 0:
for fqn in dtensor_fqns:
del state_dict[fqn]
state_dict = {}
return state_dict

hooks = []
Expand All @@ -474,11 +477,6 @@ def dtensor_to_tensor_hook(
for hook in hooks:
hook.remove()

# Convert the state dict to the requested precision
for k, v in state_dict.items():
if isinstance(v, torch.Tensor):
state_dict[k] = v.to(dtype=self.dtype)

new_model_instance = None # Need this for pyright because variable could be unbound

if dist.get_global_rank() == 0:
Expand Down Expand Up @@ -537,7 +535,7 @@ def dtensor_to_tensor_hook(
original_tokenizer.save_pretrained(temp_save_dir)

# Only need to edit files for MPT because it has custom code
if original_model.config.model_type == 'mpt':
if new_model_instance.config.model_type == 'mpt':
log.debug('Editing MPT files for HuggingFace compatibility')
edit_files_for_hf_compatibility(
temp_save_dir,
Expand Down
Loading