Skip to content

Bugfix for Mamba with Chunked-Prefill #2293

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.

Sign up for GitHub

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

Open
wants to merge 6 commits into
base: main
Choose a base branch
from
+287 −8
Open

Bugfix for Mamba with Chunked-Prefill #2293

Changes from 3 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
bc66c18
bugfix
sidsingh-nvidia Nov 18, 2025
2033904
Merge branch 'main' into siddharth/mamba-chunked-prefill-bugfix
sidsingh-nvidia Nov 18, 2025
fd49a23
Merge branch 'main' into siddharth/mamba-chunked-prefill-bugfix
sidsingh-nvidia Nov 20, 2025
875c681
add functional test
sidsingh-nvidia Nov 21, 2025
024af51
Merge branch 'main' into siddharth/mamba-chunked-prefill-bugfix
sidsingh-nvidia Nov 21, 2025
2b7207d
add functional test for mamba
sidsingh-nvidia Nov 21, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion megatron/core/ssm/mamba_mixer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -482,7 +482,8 @@ def dynamic_inference(self, hidden_states: torch.Tensor, context: DynamicInferen
zxBCdt_chunked_prefill = zxBCdt[
active_token_count - chunked_prefill_request_token_count : active_token_count
]
batch_index_chunked_prefill = batch_indices[context.chunked_prefill_request_id]
pos = torch.where(context.request_ids == context.chunked_prefill_request_id)[0][0]
Copy link
Contributor

@kvareddy kvareddy Nov 19, 2025

Choose a reason for hiding this comment

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

Can we abstract away this extraction in context?

batch_index_chunked_prefill = batch_indices[pos]

y_prefill_chunked = self.ssm_prefill(
zxBCdt_chunked_prefill,
Expand Down
Loading