Skip to content

Conversation

@ZhijunLStudio
Copy link

transformers version fix: Updated transformers version to resolve the AssertionError: Input and cos/sin must have the same dtype, got torch.float16 and torch.bfloat16 error.

deepspeed version fix: Updated deepspeed to version 0.15.4 to resolve the DeepSpeedZeRoOffload' object has no attribute '_register_hooks_recursively' error.

@0xvincii
Copy link
Contributor

Hi, thanks for your contribution. Can you describe when do you meet the bug?

@ZhijunLStudio
Copy link
Author

The issue encountered with Transformers:
AssertionError: Input and cos/sin must have the same dtype, got torch.float16 and torch.bfloat16.
For more details, please refer to this issue(#105).
The issue encountered with DeepSpeed: The latest version (0.16.4) causes the following error:
AttributeError: 'DeepSpeedZeRoOffload' object has no attribute '_register_hooks_recursively'.
To resolve this, DeepSpeed was downgraded to version 0.15.4, as this version does not have the issue.

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