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

Jump to bottom

[5.3.0] T5 model under FP16 is generating garbage #554

Open
lanking520 opened this issue Apr 12, 2023 · 2 comments
Open

[5.3.0] T5 model under FP16 is generating garbage #554

lanking520 opened this issue Apr 12, 2023 · 2 comments
Labels
bug Something isn't working

Comments

@lanking520
Copy link
Contributor

Branch/Tag/Commit

5.3.0

Docker Image Version

pytorch 22.09

GPU name

A10G

CUDA Driver

Any

Reproduced Steps

Similar to https://github.com/NVIDIA/FasterTransformer/issues/470. You can just change all weights into fp16 and inference using fp16.

Both Summarization and Translation task give very poor result.

Tested on:
- FlanT5 XXL, FlanT5-XL, FlanT5-UL2.

Tested using TP2 with 2 GPUs under fp16.
@lanking520 lanking520 added the bug Something isn't working label Apr 12, 2023
@lanking520
Copy link
Contributor Author

related to #470

@byshiue
Copy link
Collaborator

byshiue commented Apr 13, 2023

Thank you for the feedback. We find such behavior on large model and investigate the reason.

@byshiue byshiue mentioned this issue Apr 27, 2023
Closed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
bug Something isn't working
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@byshiue @lanking520