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

Example of bitblas/ladder with dtypes like int3, int5, int6, int7 #244

Open
yaoyaoding opened this issue Nov 11, 2024 · 4 comments
Open
Labels
enhancement New feature or request good first issue Good for newcomers

Comments

@yaoyaoding
Copy link

Thanks for the great work!

I am curious whether bitblas/ladder has support for the quantized data types like int3, int5, int6 (or uint3, uint5, uint6, float6_e3m2, etc)?

@LeiWang1999
Copy link
Contributor

hi @yaoyaoding , thanks for your attention! currently we do not have implementations of these bits which is not a pow of 2 (which may lead to some empty slot when compress data, for example, only 2 int3 can be stored at a int8 storage value which lead to only 6/8 utilization).

But the implementation is simple, if you really request these datatypes, I'll be pleasure to add them :)

@yaoyaoding
Copy link
Author

hi @yaoyaoding , thanks for your attention! currently we do not have implementations of these bits which is not a pow of 2 (which may lead to some empty slot when compress data, for example, only 2 int3 can be stored at a int8 storage value which lead to only 6/8 utilization).

But the implementation is simple, if you really request these datatypes, I'll be pleasure to add them :)

Thanks for your timely reply, @LeiWang1999 !

Will the slot appear in global memory, or only appear in the shared memory and register?

@LeiWang1999
Copy link
Contributor

@yaoyaoding It depends, bitblas has about three different templates for dequantization (in global, in shared, in register), we can select one to deploy this computation :)

@yaoyaoding
Copy link
Author

@LeiWang1999 Got it, thanks for the explanation!

@LeiWang1999 LeiWang1999 added enhancement New feature or request good first issue Good for newcomers labels Dec 16, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
enhancement New feature or request good first issue Good for newcomers
Projects
None yet
Development

No branches or pull requests

2 participants