Skip to content

Fix OOMs happening in case of accelerate >= 0.16.0 #310

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

Merged
borzunov merged 1 commit into from
Apr 25, 2023
Merged

Fix OOMs happening in case of accelerate >= 0.16.0 #310

Show file tree
Hide file tree
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
2 changes: 1 addition & 1 deletion setup.cfg
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,7 +33,7 @@ python_requires = >=3.7
install_requires =
torch>=1.12
bitsandbytes==0.38.0.post2
accelerate>=0.15.0,<1.0.0
accelerate>=0.16.0,<1.0.0
Copy link
Collaborator Author

@borzunov borzunov Apr 25, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

set_module_tensor_to_device's dtype arg didn't exist before 0.16.0

huggingface-hub>=0.11.1,<1.0.0
transformers>=4.25.1,<5.0.0
speedtest-cli==2.1.3
Expand Down
2 changes: 1 addition & 1 deletion src/petals/bloom/from_pretrained.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -68,7 +68,7 @@ def load_pretrained_block(
param = state_dict[param_name]
if torch_dtype != "auto" and not str(param.dtype).startswith(("torch.uint", "torch.int", "torch.bool")):
param = param.to(torch_dtype)
set_module_tensor_to_device(block, param_name, "cpu", value=param)
set_module_tensor_to_device(block, param_name, "cpu", value=param, dtype=param.dtype)

logger.info(f"Loaded {converted_model_name_or_path} block {block_index}, {report}")
return block
Expand Down