Skip to content

Do not let chunked prefill generate decode logprobs#3777

Merged
tdene merged 1 commit intoNVIDIA:mainfrom
tdene:tde/fix_chunked_prefill_bug
Mar 12, 2026
Merged

Do not let chunked prefill generate decode logprobs#3777
tdene merged 1 commit intoNVIDIA:mainfrom
tdene:tde/fix_chunked_prefill_bug

Conversation

@tdene
Copy link
Contributor

@tdene tdene commented Mar 10, 2026

What does this PR do ?

⚠️ For major changes (either in lines of code or in its impact), please make sure to first share a design doc with the team. If you're unsure what's the best way to do so, contact the @mcore-oncall.

Contribution process

Pre-checks

  • I have added relevant unit tests
  • I have added relevant functional tests
  • I have added proper typing to my code Typing guidelines
  • I have added relevant documentation
  • I have run the autoformatter.sh on my PR

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"

  1. When your PR is ready, click Ready for Review.
  2. An oncall reviewer is auto-assigned and expert reviewers are notified based on your changes.
    • Some PRs may jump straight to step 2. This is determined by .github/CODEOWNERS.

⚠️ Only mark as ready once merge-conflicts are resolved and the CI is passing.
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, the Final Review label 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 Approved label 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.com or zijiey@nvidia.com.

@copy-pr-bot
Copy link

copy-pr-bot bot commented Mar 10, 2026

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

@tdene tdene marked this pull request as ready for review March 10, 2026 17:25
@tdene tdene requested review from a team as code owners March 10, 2026 17:25
@svcnvidia-nemo-ci svcnvidia-nemo-ci requested a review from a team March 10, 2026 17:26
@svcnvidia-nemo-ci svcnvidia-nemo-ci added this to the Core 0.16 milestone Mar 10, 2026
Copy link
Contributor

@santhnm2 santhnm2 left a comment

Choose a reason for hiding this comment

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

Can we add a unit test for this?

@tdene tdene added the Expert Review [deprecated] Apply this label to indicate that your PR is ready for expert review. label Mar 10, 2026
request.prompt_log_probs = []
request.generated_log_probs = []
is_chunked_prefill = request_id == self.context.chunked_prefill_request_id
is_prefill = len(request_log_probs) > 1
Copy link
Contributor

Choose a reason for hiding this comment

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

I believe we have scheduling code to prevent prefill requests from having a single token, but would be good to double check that there's no edge case here

Copy link
Contributor Author

Choose a reason for hiding this comment

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

We guarantee prefill requests have more than one token.

We do not, in fact, guarantee that chunked prefill requests have more than one token. But that does not affect our logic here.

I left a comment for future code editors to understand.

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Comment has been invalidated by a new merge with main.

Everything is good and makes sense after merging with main.

@svcnvidia-nemo-ci svcnvidia-nemo-ci added the Final Review PR is in the "final review" stage label Mar 10, 2026
@Phlip79 Phlip79 removed Expert Review [deprecated] Apply this label to indicate that your PR is ready for expert review. labels Mar 10, 2026
@Phlip79
Copy link
Member

Phlip79 commented Mar 10, 2026

@tdene FYI we don't use the "Expert Review" label anymore

@tdene
Copy link
Contributor Author

tdene commented Mar 12, 2026

/claude review

Copy link
Contributor

@claude claude bot left a comment

Choose a reason for hiding this comment

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

LGTM

@tdene tdene force-pushed the tde/fix_chunked_prefill_bug branch from d779897 to a404777 Compare March 12, 2026 14:03
@tdene tdene changed the title Do not let chunked prefill generate decode tokens Do not let chunked prefill generate decode logprobs Mar 12, 2026
@tdene tdene force-pushed the tde/fix_chunked_prefill_bug branch from a404777 to 9db621d Compare March 12, 2026 14:12
@tdene tdene enabled auto-merge March 12, 2026 14:23
@tdene tdene added this pull request to the merge queue Mar 12, 2026
@svcnvidia-nemo-ci
Copy link

🔄 Merge queue validation started!

You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/23009338011

Merged via the queue into NVIDIA:main with commit 46227e0 Mar 12, 2026
51 of 53 checks passed
@tdene tdene deleted the tde/fix_chunked_prefill_bug branch March 12, 2026 16:16
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

complexity: low Final Review PR is in the "final review" stage

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants