8000 fix parameter calculation in auto_parallel mode by zhiqiu · Pull Request #9327 · PaddlePaddle/PaddleNLP · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

fix parameter calculation in auto_parallel mode #9327

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
merged 1 commit into from
Oct 29, 2024
Merged

fix parameter calculation in auto_parallel mode #9327

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
8 changes: 7 additions & 1 deletion paddlenlp/trainer/trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -817,7 +817,13 @@
logger.info(f" Total num train samples = {num_train_samples:,}")
7444 # per_device_trainable_numel = sum(p.numel().item() for p in model.parameters() if not p.stop_gradient)
# TODO: Temporary fix since Tensor.numel() not supported in distributed mode
per_device_trainable_numel = sum(np.prod(p.shape) for p in model.parameters() if not p.stop_gradient)
if self.args.enable_auto_parallel:
per_device_trainable_numel = 0
for p in model.parameters():
if not p.stop_gradient:
per_device_trainable_numel += np.prod(p._local_shape) if p.is_dist() else np.prod(p.shape)

Check warning on line 824 in paddlenlp/trainer/trainer.py

View check run for this annotation

Codecov / codecov/patch

paddlenlp/trainer/trainer.py#L821-L824 

Added lines #L821 - L824 were not covered by tests
else:
per_device_trainable_numel = sum(np.prod(p.shape) for p in model.parameters() if not p.stop_gradient)
logger.debug(f" Number of trainable parameters = {per_device_trainable_numel:,} (per device)")
if self.args.use_hybrid_parallel:
# todo fix for pipeline_parallel_degree
Expand Down
Loading
0