feat: Initial state support for Mamba SSM (1) #488
Conversation
This file contains 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
mzusman
changed the title
|
@mzusman I've noticed you made changes to files in the csrc directory, but I'm having trouble getting these changes to take effect in my environment. Could you please tell me the exact instructions to rebuild and install the mamba_ssm package so the changes are applied? It seems I always get the original package using |
|
@daphneOdera-618 Yeah, the default setup.py behaviour is to download the upstream's wheel upon "installing", What you would need to do to force build is to add |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Add chunked prefill / use initial state capability to Mamba ssm ( Mamba 1 ) , Done it by prepending the last forward pass state to the FWD pass kernel and read the data accordingly .
Latency is not affected. ( benchmark script shows similar latencies between this PR and main - 130ms )
Added tests that check correctness when running on chunks.
Limitations:
This PR enables efficient Speculative decoding, prefix caching and prefill chunking.
FIX #233 #473 #258 #101