Fix Mamba2 step() D handling when D_has_hdim=True#893
Open
ipnon wants to merge 1 commit intostate-spaces:mainfrom
Open
Fix Mamba2 step() D handling when D_has_hdim=True#893ipnon wants to merge 1 commit intostate-spaces:mainfrom
ipnon wants to merge 1 commit intostate-spaces:mainfrom
Conversation
When D_has_hdim=True, self.D has shape (nheads * headdim,) but step()
treated it as (nheads,) in both code paths, producing silent wrong
outputs during inference. forward() already handled this correctly
via rearrange("(h p) -> h p").
Add conditional reshape in both step() paths to match forward().
Add regression test comparing forward/step consistency.
Fixes state-spaces#887, fixes state-spaces#888.
This was referenced Apr 2, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
step()treatsself.Das shape(nheads,)in both code paths, but whenD_has_hdim=Trueit's(nheads * headdim,)— producing silent wrong outputs during inferenceforward()already handles this correctly viarearrange("(h p) -> h p", p=self.headdim)at line 191self.D_has_hdimconditional at bothstep()paths (lines 319 and 327) to matchforward()Test plan
test_mamba2_step_forward_consistencychecks forward/step output consistency for bothD_has_hdim=TrueandD_has_hdim=FalseFixes #887, fixes #888.