Skip to content

Fix deprecated torch.cuda.amp.GradScaler usage in trainer#3682

Open
hsleonis wants to merge 2 commits intoflairNLP:masterfrom
hsleonis:master
Open

Fix deprecated torch.cuda.amp.GradScaler usage in trainer#3682
hsleonis wants to merge 2 commits intoflairNLP:masterfrom
hsleonis:master

Conversation

@hsleonis
Copy link

@hsleonis hsleonis commented Aug 6, 2025

Replace deprecated torch.cuda.amp.GradScaler with the new torch.amp.GradScaler API to resolve FutureWarning in PyTorch 2.4+.

Changes:
Updated trainer.py line 545 to use torch.amp.GradScaler('cuda', ...) instead of torch.cuda.amp.GradScaler(...)

Issue:
Fixes deprecation warning:

FutureWarning: `torch.cuda.amp.GradScaler(args...)` is deprecated. 
Please use `torch.amp.GradScaler('cuda', args...)` instead.

Compatibility:

  • Maintains backward compatibility (PyTorch 1.10+ supports both APIs)
  • No functional changes - purely API migration
  • Tested with existing training workflows

Fix deprecation warning in Flair's trainer code. The issue is that: Flair is using the old PyTorch AMP (Automatic Mixed Precision) API.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants