Draft
Conversation
Signed-off-by: Keshav Santhanam <ksanthanam@nvidia.com>
sidsingh-nvidia
requested changes
Mar 10, 2026
| if remainder > 0: | ||
| self.request_query_lengths[N_total - 1] += remainder | ||
|
|
||
| self.request_kv_length_offsets[0:N_total].fill_(0) |
Contributor
There was a problem hiding this comment.
Can we use = here? We have seen weird issues with fill_ at times, wherein it creates a duplicate tensor.
| prefill_tokens = T - N_decode | ||
| tokens_per_prefill = prefill_tokens // N_prefill | ||
| remainder = prefill_tokens % N_prefill | ||
| self.request_query_lengths[N_decode:N_total].fill_(tokens_per_prefill) |
Contributor
There was a problem hiding this comment.
same here, would advise using = instead of fill_
| token_offset = N_decode | ||
| for i in range(N_prefill): | ||
| qlen = tokens_per_prefill + (remainder if i == N_prefill - 1 else 0) | ||
| self.token_to_request_idx[token_offset : token_offset + qlen] = N_decode + i |
Contributor
There was a problem hiding this comment.
can we compute all of the values to write on the CPU first, and then do a single write to GPU?
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What does this PR do ?
EP dummy ranks for Mamba hybrid models could have
real_prefill_count == 0butpadded_prefill_count > 0after the EP all-reduce MAX synchronization inmatch_graph_config. This causedcu_seqlensinMambaMetadata.updateto be all-zeros (since there were no real prefill tokens to compute cumulative lengths from), triggering an illegal memory access in the Mamba SSM prefill kernel. This PR defers the metadata construction in dummy ranks until after the EP synchronization to avoid this issue.Contribution process
Pre-checks
Code review
Feel free to message or comment the @mcore-oncall to help accelerate your merge into main. The less complex your PR is, the faster it will be approved and merged!
All PRs start as draft. If you open a non-draft PR, it will be automatically converted to draft.
Step 1: Mark PR as "Ready for Review"
.github/CODEOWNERS.Final Review might get declined if these requirements are not fulfilled.
Step 2: Final Review
For PRs that change
megatron/core, once all expert reviewers have approved, theFinal Reviewlabel is applied automatically and final reviewers are assigned.For PRs outside
megatron/core, this step is skipped.Step 3: Approved
Once all required reviewers have approved, the
Approvedlabel is applied automatically.Merge
Any member of mcore-engineers will be able to merge your PR.
For MRs into `dev` branch
The proposed review process for `dev` branch is under active discussion.MRs are mergable after one approval by either
eharper@nvidia.comorzijiey@nvidia.com.