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.

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

Does AIT support BF16 inference now? #899

Open
sanbuphy opened this issue Aug 15, 2023 · 4 comments
Open

Does AIT support BF16 inference now? #899

sanbuphy opened this issue Aug 15, 2023 · 4 comments

Comments

@sanbuphy
Copy link

sanbuphy commented Aug 15, 2023

Hi, I would like to explore the model training in BF16 to high performance inference, the model only can run in bf16, if change it to fp16 it will out of normal precision. I try to search and found this issue: #110

@sanbuphy
Copy link
Author

@sanbuphy Yes, bfloat16 is now supported: see https://github.com/facebookincubator/AITemplate/blame/2f61d70c4b14d59f557c1fd9a98cbb5030307f9d/python/aitemplate/compiler/dtype.py#L28

Thanks , is that mean I can open a switch and it can make the all model run in bf16?

@sgrigory
Copy link

Yes, in theory that should work. I imagine some ops not having implementations for some of the data types though. Let us know if you notice any hiccups

@sanbuphy
Copy link
Author

Yes, in theory that should work. I imagine some ops not having implementations for some of the data types though. Let us know if you notice any hiccups

Thank's cool ! Let me have a try

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

No branches or pull requests

2 participants