Add Cosmos3 action modality#4102
Conversation
Signed-off-by: Maciej Bala <mbala@nvidia.com>
Signed-off-by: Maciej Bala <mbala@nvidia.com>
Signed-off-by: Bartosz Stefaniak <bstefaniak@nvidia.com>
Signed-off-by: Bartosz Stefaniak <bstefaniak@nvidia.com>
Signed-off-by: Bartosz Stefaniak <bstefaniak@nvidia.com>
…lags Signed-off-by: Bartosz Stefaniak <bstefaniak@nvidia.com>
Signed-off-by: lishunyang12 <lishunyang12@163.com>
…nd tokenizer Signed-off-by: Bartosz Stefaniak <bstefaniak@nvidia.com>
Signed-off-by: Bartosz Stefaniak <bstefaniak@nvidia.com>
Signed-off-by: Bartosz Stefaniak <bstefaniak@nvidia.com>
Signed-off-by: Bartosz Stefaniak <bstefaniak@nvidia.com>
Signed-off-by: Bartosz Stefaniak <bstefaniak@nvidia.com>
Signed-off-by: Bartosz Stefaniak <bstefaniak@nvidia.com>
Signed-off-by: Bartosz Stefaniak <bstefaniak@nvidia.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 1b410e6601
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
1b410e6 to
aab40ee
Compare
Signed-off-by: Bartosz Stefaniak <bstefaniak@nvidia.com>
|
Let's test it and update the recipe for other three models in the series. |
Signed-off-by: lishunyang12 <lishunyang12@163.com>
Recipe docs for the action modality are included in this PR ( |
Signed-off-by: lishunyang12 <lishunyang12@163.com>
Signed-off-by: lishunyang12 <lishunyang12@163.com>
Signed-off-by: lishunyang12 <lishunyang12@163.com>
The recipes are being relocated from recipes/nvidia to recipes/cosmos3 in vllm-project/vllm-omni#4102; update the links to the new path. Signed-off-by: lishunyang12 <lishunyang12@163.com>
The "Generator with vLLM-Omni" section has some stale references now that the upstream PRs have landed. **Compatibility status paragraph** - #3454 (text-to-image / text-to-video / image-to-video) is merged on vllm-omni `main`, not "being upstreamed." - video-with-sound (vllm-project/vllm-omni#4073) is also merged — it's listed as a pending follow-up. - Action is still in review (vllm-project/vllm-omni#4102); video-to-video is still planned. - Linked the maintained recipes (Cosmos3-Nano / Cosmos3-Super) for per-modality usage. **Install snippet** - Since #3454/#4073 are merged, install from `main` instead of the `@refs/pull/3454/head` PR ref. **Dead link** - `examples/online_serving/cosmos3` 404s (it doesn't exist on `main`, and isn't added by #4102). Repointed both references to the maintained `recipes/nvidia`. Scoped to the vLLM-Omni section; Docker quick-start and parallelism options left as-is. --------- Signed-off-by: lishunyang12 <lishunyang12@163.com> Co-authored-by: lishunyang12 <lishunyang12@163.com>
Signed-off-by: Maciej Bala <mbala@nvidia.com> Signed-off-by: Bartosz Stefaniak <bstefaniak@nvidia.com> Signed-off-by: lishunyang12 <lishunyang12@163.com> Co-authored-by: Maciej Bala <mbala@nvidia.com> Co-authored-by: lishunyang12 <lishunyang12@163.com> Co-authored-by: SYLAR <125541396+lishunyang12@users.noreply.github.com>
|
@bastefaniak Hi, just want to ask if you are planning to support backward dynamics soon to complete the full support. |
|
@lishunyang12 Yes, that will be part of video2video PR, as we need to pass a full video to the server, which is not currently supported. |
Signed-off-by: Maciej Bala <mbala@nvidia.com> Signed-off-by: Bartosz Stefaniak <bstefaniak@nvidia.com> Signed-off-by: lishunyang12 <lishunyang12@163.com> Co-authored-by: Maciej Bala <mbala@nvidia.com> Co-authored-by: lishunyang12 <lishunyang12@163.com> Co-authored-by: SYLAR <125541396+lishunyang12@users.noreply.github.com> Signed-off-by: akshatvishu <akshatnayak197@gmail.com>
Signed-off-by: Maciej Bala <mbala@nvidia.com> Signed-off-by: Bartosz Stefaniak <bstefaniak@nvidia.com> Signed-off-by: lishunyang12 <lishunyang12@163.com> Co-authored-by: Maciej Bala <mbala@nvidia.com> Co-authored-by: lishunyang12 <lishunyang12@163.com> Co-authored-by: SYLAR <125541396+lishunyang12@users.noreply.github.com>
Purpose
This PR is a follow-up to #3454 and #4073, it adds action policy/forward dynamics/inverse dynamics generation.
Cosmos3 model is available under https://huggingface.co/nvidia/Cosmos3-Nano (and more variants)
Test Plan
Unit tests
cd tests; python -m pytest -v -m "core_model and cpu"Added 4 new tests, updated 9 existing ones to also support action modality
Serving tests
Host server with:
vllm serve nvidia/Cosmos3-Nano --omniForward dynamics:
Policy:
Online inference of inverse dynamics will be added in follow-up MR
Test Result
3185 passed, 15 skipped, 1533 deselected, 55 warnings in 227.68s (0:03:47)Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model. Please runmkdocs serveto sync the documentation editions to./docs.BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)