Inference: Do not let prompt tokens return from the engine, unless requested by sidsingh-nvidia · Pull Request #5918 · NVIDIA/Megatron-LM

sidsingh-nvidia · 2026-07-21T05:37:21Z

I, the PR author, have personally reviewed every line of this PR.

What does this PR do?

Transmitting prompt tokens, especially in agentic-RL at high ctx lengths, can be costly. This PR makes it such that prompt tokens are removed from the request object before serialization by default, unless those tokens are explicitly requested by the user. Benefits have been observed with cross-node DP

⚠️ 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 @NVIDIA/mcore-oncall.

Issue tracking

For PRs from open-source community contributors:

New features: a linked issue is required. Please open a feature request and reference it here before submitting the PR.
Small updates (bug fixes, minor improvements): a linked issue is recommended and will accelerate the PR review process.

Linked issue:

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 @NVIDIA/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"

When your PR is ready, click Ready for Review.
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.

Signed-off-by: Siddharth Singh <sidsingh@nvidia.com>

github-actions · 2026-07-21T05:37:30Z

This PR has been automatically converted to draft because all PRs must start as drafts.

When you are ready for review, click Ready for Review to begin the review process. This will:

Add the oncall reviewer (optional reviewer)
Add required review teams based on your changes

See the contribution guide for more details.

copy-pr-bot · 2026-07-21T05:37:31Z

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.

Cover the new DynamicInferenceRequest.serialize() drop/keep contract from PR NVIDIA#5918: prompt_length is always reported (needed for usage.prompt_tokens), the prompt_tokens tensor is dropped from the wire payload unless SamplingParams.return_prompt_tokens=True, self.prompt_tokens is restored on the local instance, the routing_indices shape check still uses the saved prompt_len when self.prompt_tokens is temporarily nulled, and the None-prompt edge case does not crash. Signed-off-by: shanmugamr1992 <shanmugamr@nvidia.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

svcnvidia-nemo-ci · 2026-07-22T23:51:55Z

🔄 Merge queue validation started!

You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/29967398748

svcnvidia-nemo-ci · 2026-07-23T02:14:12Z

🔄 Merge queue validation started!

You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/29973870891

megatron-core now drops `prompt_tokens` from a finished inference request before serializing it unless `SamplingParams.return_prompt_tokens` is set (NVIDIA/Megatron-LM#5918), so the reply the inference client hands back carries `prompt_tokens=None` and packing the result raised `AttributeError: 'NoneType' object has no attribute 'tolist'`. Take the prompt from the batch that was submitted instead of from the engine's reply. NeMo-RL already holds those tokens, so this avoids shipping the prompt back over ZMQ purely to echo it, and it works against both the old and new megatron-core behaviour without a version branch. Three L1 Megatron generation tests failed outright on this; two async ones exited 0 because the rollout loop catches per-sample generation errors, so they were silently generating nothing. Signed-off-by: shanmugamr1992 <shanmugamr1992@gmail.com>

…quested (NVIDIA#5918) Signed-off-by: Siddharth Singh <sidsingh@nvidia.com> Signed-off-by: shanmugamr1992 <shanmugamr@nvidia.com> Co-authored-by: Shanmugam Ramasamy <111910568+shanmugamr1992@users.noreply.github.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…quested (NVIDIA#5918) Signed-off-by: Siddharth Singh <sidsingh@nvidia.com> Signed-off-by: shanmugamr1992 <shanmugamr@nvidia.com> Co-authored-by: Shanmugam Ramasamy <111910568+shanmugamr1992@users.noreply.github.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Signed-off-by: Dmytro Pykhtar <dpykhtar@nvidia.com>

do not let prompt tokens travel from engine -> coordinator

12124a3

Signed-off-by: Siddharth Singh <sidsingh@nvidia.com>

sidsingh-nvidia requested review from a team as code owners July 21, 2026 05:37

svcnvidia-nemo-ci marked this pull request as draft July 21, 2026 05:37

copy-pr-bot Bot temporarily deployed to public July 21, 2026 05:37 Inactive

copy-pr-bot Bot temporarily deployed to public July 21, 2026 05:41 Inactive

copy-pr-bot Bot temporarily deployed to public July 21, 2026 05:50 Inactive

kvareddy approved these changes Jul 21, 2026

View reviewed changes

shanmugamr1992 mentioned this pull request Jul 21, 2026

test: add unit tests for return_prompt_tokens serialization sidsingh-nvidia/Megatron-LM#5

Merged

2 of 3 tasks

shanmugamr1992 and others added 2 commits July 21, 2026 14:54

add unit test and lint

3a5130f

sidsingh-nvidia marked this pull request as ready for review July 21, 2026 22:17

Merge branch 'main' into siddharth/trim-prompt-tokens-from-engine

998f20d

svcnvidia-nemo-ci added the Final Review PR is in the "final review" stage label Jul 21, 2026

copy-pr-bot Bot temporarily deployed to public July 21, 2026 22:18 Inactive

svcnvidia-nemo-ci added the complexity: low label Jul 21, 2026

copy-pr-bot Bot had a problem deploying to test July 21, 2026 22:19 Error

copy-pr-bot Bot temporarily deployed to public July 21, 2026 22:21 Inactive

copy-pr-bot Bot temporarily deployed to public July 21, 2026 22:22 Inactive

copy-pr-bot Bot temporarily deployed to public July 21, 2026 22:31 Inactive

copy-pr-bot Bot temporarily deployed to public July 22, 2026 04:53 Inactive

copy-pr-bot Bot temporarily deployed to public July 22, 2026 04:54 Inactive

copy-pr-bot Bot temporarily deployed to public July 22, 2026 05:02 Inactive

Merge branch 'main' into siddharth/trim-prompt-tokens-from-engine

89c0c1e

copy-pr-bot Bot temporarily deployed to public July 22, 2026 17:56 Inactive

copy-pr-bot Bot temporarily deployed to test July 22, 2026 17:57 Inactive

copy-pr-bot Bot temporarily deployed to public July 22, 2026 17:59 Inactive

copy-pr-bot Bot temporarily deployed to public July 22, 2026 18:00 Inactive

copy-pr-bot Bot temporarily deployed to public July 22, 2026 18:10 Inactive

sidsingh-nvidia added this pull request to the merge queue Jul 22, 2026

Merged via the queue into NVIDIA:main with commit 602fad0 Jul 23, 2026
88 of 89 checks passed

sidsingh-nvidia deleted the siddharth/trim-prompt-tokens-from-engine branch July 23, 2026 05:57

This was referenced Jul 27, 2026

fix: stop relying on megatron-core echoing prompt tokens back NVIDIA-NeMo/RL#3363

Closed

Feature/enable inference optimized qwen moe #5700

Merged

Navigation Menu

Repository navigation

Inference: Do not let prompt tokens return from the engine, unless requested - #5918

Inference: Do not let prompt tokens return from the engine, unless requested#5918
sidsingh-nvidia merged 6 commits into
NVIDIA:mainfrom
sidsingh-nvidia:siddharth/trim-prompt-tokens-from-engine

sidsingh-nvidia commented Jul 21, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Jul 21, 2026

Uh oh!

copy-pr-bot Bot commented Jul 21, 2026

Uh oh!

svcnvidia-nemo-ci commented Jul 22, 2026

Uh oh!

svcnvidia-nemo-ci commented Jul 23, 2026

Uh oh!

Uh oh!

This branch was previously deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

sidsingh-nvidia commented Jul 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What does this PR do?

Issue tracking

Contribution process

Pre-checks

Code review

Step 1: Mark PR as "Ready for Review"

Step 2: Final Review

Step 3: Approved

Merge

Uh oh!

github-actions Bot commented Jul 21, 2026

Uh oh!

copy-pr-bot Bot commented Jul 21, 2026

Uh oh!

svcnvidia-nemo-ci commented Jul 22, 2026

Uh oh!

svcnvidia-nemo-ci commented Jul 23, 2026

Uh oh!

Uh oh!

This branch was previously deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

sidsingh-nvidia commented Jul 21, 2026 •

edited

Loading