Skip to content

Add NIM inference path for the Cosmos 3 Reasoner#175

Merged
Dinghow merged 1 commit into
NVIDIA:mainfrom
KrzysztofTomala:docs/reasoner-nim-inference
Jun 3, 2026
Merged

Add NIM inference path for the Cosmos 3 Reasoner#175
Dinghow merged 1 commit into
NVIDIA:mainfrom
KrzysztofTomala:docs/reasoner-nim-inference

Conversation

@KrzysztofTomala
Copy link
Copy Markdown
Contributor

Add NIM inference path for the Cosmos 3 Reasoner

@KrzysztofTomala KrzysztofTomala changed the title Add NIM inference path for the Cosmos 3 Reasoner Draft: Add NIM inference path for the Cosmos 3 Reasoner Jun 1, 2026
@KrzysztofTomala KrzysztofTomala force-pushed the docs/reasoner-nim-inference branch 2 times, most recently from 7d63992 to ea5be36 Compare June 1, 2026 15:19
@KrzysztofTomala KrzysztofTomala marked this pull request as draft June 1, 2026 15:26
Comment thread cookbooks/cosmos3/reasoner/README.md Outdated

Video inputs use the `video_url` content type and accept `media_io_kwargs`
frame-sampling controls through `extra_body`
(`{"media_io_kwargs": {"video": {"fps": 3.0}}}`). See the
Copy link
Copy Markdown

Choose a reason for hiding this comment

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

default fps is 4, maybe change it to 4 for consistency.

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

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

done

Comment thread cookbooks/cosmos3/reasoner/README.md Outdated
Comment on lines +180 to +181
assets are sent as base64 data URIs and video frame sampling is controlled with
NIM's `media_io_kwargs` instead of the vLLM-cosmos3 `mm_processor_kwargs`. Each
Copy link
Copy Markdown

Choose a reason for hiding this comment

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

This sentence says NIM uses media_io_kwargs instead of mm_processor_kwargs. These are actually different controls:

media_io_kwargs is for frame sampling (fps, num_frames), video decoder level
mm_processor_kwargs is for pixel budget (shortest/longest edge), model processor level
Both work in NIM. You can remove it or say "video frame sampling is controlled with media_io_kwargs, while spatial resolution is controlled with mm_processor_kwargs."

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

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

done

@KrzysztofTomala KrzysztofTomala force-pushed the docs/reasoner-nim-inference branch from ea5be36 to cfcd7ae Compare June 1, 2026 15:50
@KrzysztofTomala KrzysztofTomala marked this pull request as ready for review June 1, 2026 15:55
@KrzysztofTomala KrzysztofTomala changed the title Draft: Add NIM inference path for the Cosmos 3 Reasoner Add NIM inference path for the Cosmos 3 Reasoner Jun 1, 2026
@lfengad
Copy link
Copy Markdown
Collaborator

lfengad commented Jun 2, 2026

@KrzysztofTomala some conflicts might be resolved. THX!

@KrzysztofTomala KrzysztofTomala force-pushed the docs/reasoner-nim-inference branch from cfcd7ae to dd87dc3 Compare June 2, 2026 08:58
@KrzysztofTomala
Copy link
Copy Markdown
Contributor Author

KrzysztofTomala commented Jun 2, 2026

@lfengad I have resolved the conflicts

Copy link
Copy Markdown
Collaborator

@Dinghow Dinghow left a comment

Choose a reason for hiding this comment

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

LGTM

@Dinghow Dinghow merged commit 335392c into NVIDIA:main Jun 3, 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.

5 participants