Vincent Zhong

## Motivation add the minimum version of gguf as needed. ## Modifications the version SET is based on the link to `transformers` that has a bit more context to the...

## Motivation Motivated by the note in code, we should use `torch.clamp` cc @merrymercy if you could please look. https://pytorch.org/docs/stable/generated/torch.clamp.html ## Modifications use `torch.clamp` for val > The `torch.minimum` function...

Updated the use of `Type | None` to `Optional[Type]` for support `[email protected]`

1

## Motivation So Python 3.8 is supported. The versions here, they are not tested like this on the regular CI, so it can be missed but we should be careful....

Usage of persistent QK (rms) norm and performance vs. normal rmsnorm

5

# Question In persistent QK rms norm what is a real example use case where it beats regular path? https://github.com/flashinfer-ai/flashinfer/pull/1843 PR that introduce this feature (dispatch to either impl) ##...