Skip to content

Jwilber/evo2 generate patch #949

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.

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 19 commits into
base: main
Choose a base branch
from
Open

Conversation

jwilber
Copy link
Collaborator

@jwilber jwilber commented Jun 18, 2025

Fix the incompatibility between the Megatron/NeMi-based implementation of hyena operators in BioNeMo and the original Vortex kernels used for generation. Namely, add in Vortex's stateful convolution logic used during inference, based on a previous migration from @antonvnv.

Type of changes

  • Bug fix (non-breaking change which fixes an issue)
  • New feature (non-breaking change which adds functionality)
  • Refactor
  • Documentation update
  • Other (please describe):

Usage

TODO: Add code snippet

Pre-submit Checklist

  • I have tested these changes locally
  • I have updated the documentation accordingly
  • I have added/updated tests as needed
  • All existing tests pass successfully

antonvnv and others added 16 commits June 12, 2025 15:35
Signed-off-by: John St John <[email protected]>
- NeMo: f17dbf74d (Anton's hyena changes + seq_len_interpolation_factor)
- Megatron-LM: 5fc4f681 (Anton's inference context fixes)
- Based on bionemo-compatible versions (6a78ab8 and fb7c3f8)
Copy link

copy-pr-bot bot commented Jun 18, 2025

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

Jared Wilber added 2 commits June 18, 2025 16:40
Signed-off-by: Jared Wilber <[email protected]>
Signed-off-by: Jared Wilber <[email protected]>
@jwilber jwilber mentioned this pull request Jun 19, 2025
Copy link
Collaborator

@farhadrgh farhadrgh left a comment

Choose a reason for hiding this comment

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

Is the plan to merge this PR into main, or will it remain a feature branch without merging? CC: @trvachov
I’ve noticed that this PR is undoing some of the improvements we previously applied to the NeMo implementation in hyena_utils and hyena_mixer, particularly regarding memory overhead (e.g., unnecessary variable creation and excessive tensor rearrangements).

We should not merge this PR into main until these issues are addressed. Merging as-is would complicate future NeMo upgrades and prevent us from leveraging cuHyena efficiently.

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.

4 participants