Skip to content

Conversation

thoangtrvn
Copy link
Contributor

@thoangtrvn thoangtrvn commented Mar 21, 2025

This is related to the prior PR: #47

This adds the second Triton kernel (decode stage) to be used in mamba-based model for the inference purpose.

@facebook-github-bot facebook-github-bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Mar 21, 2025
cache_seqlens: Optional[torch.Tensor] = None,
conv_state_indices: Optional[torch.Tensor] = None,
pad_slot_id: int = PAD_SLOT_ID,
):
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

you are returning o but it is not listed here in your function signature?

for example: cache_indices = [pad_slot_id, 1 ,20 ,pad_slot_id]
in this case, the kernel will not process entries at
indices 0 and 3
out: (batch, dim) or (batch, dim, seqlen)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

also inconsistent - out? (vs o)

conv_state_indices=conv_state_indices,
pad_slot_id=pad_slot_id,
)
return o
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit but not a fan of using o by itself... out or output etc. makes it more clear imo.

@lessw2020
Copy link
Contributor

Hi @thoangtrvn - thanks for the update and sorry for the delay!
I had a question re: the lack of return signature as well as inconsistent naming (out, then o) which are minor but would be nice to clarify it.
Otherwise looks good!

@thoangtrvn
Copy link
Contributor Author

Thanks @lessw2020 , I'll update base on your feedback.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants