-
Notifications
You must be signed in to change notification settings - Fork 47
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
Matmul quantized test metadata #481
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Really nice detailed explanation! 🙂
I have some minor comments, mostly over form more than content.
Co-authored-by: Guillaume Lagrange <[email protected]>
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM to me apart from superficial comments
we introduce different algorithms provided by CubeCL. | ||
This is a best effort list and we focus first on non-trivial algorithms | ||
deserving more explanations than what is reasonable to put in the API documentation. | ||
This section is also a bit more technical compared to the others as it serve two purposes. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
serves
For example, if \\(\sigma = 0.3\\), then \\(f = -1\\) as \\(2^{-1} = 0.5 > 0.3 \\) | ||
and \\(2^{-2} = 0.25 < 0.3\\). |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This sentence is confusing
} | ||
``` | ||
Of course, | ||
in CubeCL, we stride to provide the fastest implementation for GPU devices. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
we stride? 😄
I write a new section for the quantized matmul in the book. This will be used as reference for the implementation. I also added support for matmul metadata to the testing suite.