Repository navigation
Conversation
TensorRT-LLM #19504 pins cosmos_guardrail 0.3.2, which copies the guardrail's NLTK data out of the Hugging Face cache as regular files on first use. The server-side NLTK_DATA setup documented for 0.3.0 is no longer needed. - Bump the documented guardrail pin from 0.3.0 to 0.3.2. - Remove the "Server-side NLTK data setup" section and every link to it from the Action and Transfer READMEs and TensorRT-LLM notebooks. - Update the validated TensorRT-LLM revision to cc63e66d (1.4.0rc0), which contains #19504 and #17476. Validated on 1x B200 with Cosmos3-Nano, guardrails enabled, NLTK_DATA unset, and a Hugging Face cache holding the symlinked guardrail snapshot: both Action notebooks, all five Transfer controls, and the Action and Transfer README quickstart snippets. Signed-off-by: Igor Shovkun <ishovkun@nvidia.com>
ConstBob
approved these changes
Oct 14, 2026
ConstBob
left a comment
ConstBob
left a comment
Collaborator
There was a problem hiding this comment.
LGTM. The NLTK workaround section is gone, and the Action/Transfer READMEs and TensorRT-LLM notebooks no longer link to it. The pin is cosmos_guardrail==0.3.2, and the validated TensorRT-LLM revision is cc63e66d. That matches the 0.3.0 symlink failure we hit with guardrails on. I did not re-run 0.3.2 here. Approving from the diff and your B200 check (guardrails on, NLTK_DATA unset).
lfengad
pushed a commit
to NVIDIA/cosmos-framework
that referenced
this pull request
Oct 29, 2026
The moved Cosmos3 cookbooks still pin `cosmos_guardrail==0.3.0` and require manual server-side NLTK setup. Pin `cosmos_guardrail==0.3.2`, which copies its NLTK data from the Hugging Face cache on first use, and remove the obsolete setup instructions from the Action and Transfer READMEs and notebooks. Update the TensorRT-LLM source pin to `cc63e66daf52644f1441e117bb7ba67d20566016`. Port of NVIDIA/cosmos#378 to the cookbooks' new location. Clarify that the empty video-content classifier warning does not disable the configured text checks or face blurring. Signed-off-by: Igor Shovkun <ishovkun@nvidia.com>
Collaborator
Collaborator
|
Thanks @ishovkun, and thanks @ConstBob for the review. Closing because this change has already landed in Cosmos Framework, where the cookbooks now live (they moved to
If anything from this PR is still missing there, please open a PR against NVIDIA/cosmos-framework. |
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
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.
Summary
Follow-up to #310. TensorRT-LLM #19504 pins
cosmos_guardrail0.3.2, which copies the guardrail's NLTK data out of the Hugging Face cache as regular files on first use. The server-sideNLTK_DATAworkaround documented for 0.3.0 is no longer needed.cc63e66d(1.4.0rc0), which contains #19504 and #17476.Overlaps with #343 by a few lines in
cookbooks/cosmos3/README.md(the guardrail install block); whichever lands second needs a trivial rebase.