feat(vlm): add Nemotron-Omni RADIO post-load patches#2311
Merged
Conversation
yuekaizhang
requested review from
HuiyingLi,
ZhiyuLi-Nvidia,
adil-a,
akoumpa,
athitten,
hemildesai,
pthombre and
zyzhou5
as code owners
HuiyingLi
previously approved these changes
May 25, 2026
Contributor
HuiyingLi
left a comment
HuiyingLi
left a comment
There was a problem hiding this comment.
LGTM thank you @yuekaizhang !
Contributor
|
/ok to test 29bae71 |
Contributor
|
Hi @yuekaizhang could you please fix the ci errors, thank you~ |
- enable_radio_vit_fused_attn(): route RADIO timm ViT attention through F.scaled_dot_product_attention so the (B, H, seq, seq) attention tensor (~5 GiB per block at RADIO-v2-H + dynamic-resolution patch counts) is not materialized. Mirrors the Megatron-Bridge path's vision_config.use_flash_attn=True. No-op on non-RADIO models; invoked unconditionally from apply_model_infrastructure(). - apply_parameter_freezing(): new freeze_video_embedder knob (default False). patch_generator.video_embedder is only exercised on video inputs; on image-only training it sits in the optimizer without state (no grad → no lazy init), so dcp.load on resume raises a missing-key error. Independent of freeze_vision_tower so the image encoder can stay trainable while the video branch is frozen out. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Signed-off-by: root <zhangyuekai@foxmail.com>
Contributor
Author
Done. Thanks! |
Contributor
|
Thank you @yuekaizhang , I think the codecov is still failing. Would you mind fixing that? Appreciate it! |
- enable_radio_vit_fused_attn: flips fused_attn on all blocks, resolves vision_model via both top-level and nested model.model paths, no-ops when RADIO is absent, and tolerates blocks that lack an attn attribute. - apply_parameter_freezing(freeze_video_embedder=...): True freezes only patch_generator.video_embedder.* and leaves the rest of patch_generator trainable; False (default) keeps it trainable. Pushes codecov/patch on these additions over the 80% threshold. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Signed-off-by: root <zhangyuekai@foxmail.com>
Contributor
Author
|
/ok to test 9bd5d2c |
Contributor
Author
@HuiyingLi Fixed it now. |
yuekaizhang
added a commit
to yuekaizhang/RL
that referenced
this pull request
May 27, 2026
Pulls in "feat(vlm): add Nemotron-Omni RADIO post-load patches" (NVIDIA-NeMo/Automodel#2311). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Signed-off-by: Yuekai Zhang <zhangyuekai@foxmail.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This PR improves nemotron-3-omni: