Skip to content

[None][feat] Support Helix CP with Eagle3 - #14831

Closed
brb-nv wants to merge 4 commits into
NVIDIA:mainfrom
brb-nv:user/brb/helix-cp-eagle-support
Closed

[None][feat] Support Helix CP with Eagle3#14831
brb-nv wants to merge 4 commits into
NVIDIA:mainfrom
brb-nv:user/brb/helix-cp-eagle-support

Conversation

@brb-nv

@brb-nv brb-nv commented Jun 1, 2026

Copy link
Copy Markdown
Collaborator

@coderabbitai summary

Description

  • Known limitation (acceptance rate, not correctness):the Eagle3 draft submodule is built with the TP-repurposed mapping, so its MLA attention doesn't yet do the Helix all-to-all. Final output is correct (the target verify uses correct Helix attention), but draft acceptance may be low until the draft attention is given the CP mapping (mapping_with_cp). This is a follow-up; flagged inhelix.mdand the code.
  • Highest-risk areas to watch in e2e:the KV-slot bookkeeping when all drafts are accepted (off-by-one edge), the overlap-scheduler verify sub-branch (Helix+overlap is "Untested" upstream), and CUDA-graph capture withdraft_len > 0under Helix. The verify path and ownership math are the parts I'm most confident in; the draft-loopkv_lenshandling is the part most likely to need iteration.

Test Coverage

PR Checklist

Please review the following before submitting your PR:

  • PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.

  • PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.

  • Test cases are provided for new code paths (see test instructions)

  • If PR introduces API changes, an appropriate PR label is added - either api-compatible or api-breaking. For api-breaking, include BREAKING in the PR title.

  • Any new dependencies have been scanned for license and vulnerabilities

  • CODEOWNERS updated if ownership changes

  • Documentation updated as needed

  • Update tava architecture diagram if there is a significant design change in PR.

  • The reviewers assigned automatically/manually are appropriate for the PR.

  • Please check this after reviewing the above items as appropriate for this PR.

GitHub Bot Help

To see a list of available CI bot commands, please comment /bot help.

Signed-off-by: Balaram Buddharaju <169953907+brb-nv@users.noreply.github.com>
@brb-nv
brb-nv force-pushed the user/brb/helix-cp-eagle-support branch from 685ae73 to 04c8822 Compare June 1, 2026 20:40
brb-nv added 2 commits June 1, 2026 22:20
Signed-off-by: Balaram Buddharaju <169953907+brb-nv@users.noreply.github.com>
$ pytest tests/integration/defs/accuracy/test_disaggregated_serving.py::TestKimiK25::test_nvfp4_eagle3 -s -v

Signed-off-by: Balaram Buddharaju <169953907+brb-nv@users.noreply.github.com>
@brb-nv
brb-nv force-pushed the user/brb/helix-cp-eagle-support branch from df40ea6 to 31b598e Compare June 3, 2026 02:09
Signed-off-by: Balaram Buddharaju <169953907+brb-nv@users.noreply.github.com>
@brb-nv

brb-nv commented Jul 6, 2026

Copy link
Copy Markdown
Collaborator Author

Closing in favor of #16003.

@brb-nv brb-nv closed this Jul 6, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant