Skip to content

feat: Add torch._grouped_mm expert backend - #1228

Merged
hemildesai merged 8 commits into
mainfrom
hemil/torch-moe
Feb 13, 2026
Merged

hemildesai merged 8 commits into
mainfrom
hemil/torch-moe

Conversation

@hemildesai

Copy link
Copy Markdown
Contributor

Adds torch._grouped_mm for expert computation.

Convergence runs - https://wandb.ai/Nemo-automodel/automodel-qwen3-moe-torch-experts?nw=nwuserhemild

@copy-pr-bot

copy-pr-bot Bot commented Feb 10, 2026

Copy link
Copy Markdown

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.

@hemildesai

Copy link
Copy Markdown
Contributor Author

/ok to test 15ca668

@hemildesai

Copy link
Copy Markdown
Contributor Author

/ok to test bf03a55

@hemildesai

Copy link
Copy Markdown
Contributor Author

/ok to test 641a1a4

akoumpa
akoumpa previously approved these changes Feb 12, 2026
@hemildesai

Copy link
Copy Markdown
Contributor Author

/ok to test 7fa1dd7

@hemildesai

Copy link
Copy Markdown
Contributor Author

/ok to test 715d2d6

@hemildesai

Copy link
Copy Markdown
Contributor Author

/ok to test 895ecdf

Signed-off-by: Hemil Desai <hemild@nvidia.com>
Signed-off-by: Hemil Desai <hemild@nvidia.com>
Signed-off-by: Hemil Desai <hemild@nvidia.com>
Signed-off-by: Hemil Desai <hemild@nvidia.com>
Signed-off-by: Hemil Desai <hemild@nvidia.com>
Signed-off-by: Hemil Desai <hemild@nvidia.com>
Signed-off-by: Hemil Desai <hemild@nvidia.com>
Signed-off-by: Hemil Desai <hemild@nvidia.com>
@hemildesai

Copy link
Copy Markdown
Contributor Author

/ok to test c6566e0

This branch was previously deployed

2 inactive deployments
nemo-ci — c6566e0c Deployed Feb 13, 2026 by copy-pr-bot[bot] via L2_HF_Transformer_Finetune #3781
test — c6566e0c Deployed Feb 12, 2026 by copy-pr-bot[bot] via cicd-wait-in-queue #3781
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.

2 participants