pjin/super-v3.5-dev-20260826 - #3941
Draft
pjin-nvidia wants to merge 26 commits into
Draft
Conversation
Signed-off-by: Jiaqi Zeng <jiaqiz@nvidia.com> Signed-off-by: Peter Jin <pjin@nvidia.com>
Signed-off-by: Peter Jin <pjin@nvidia.com>
Signed-off-by: Peter Jin <pjin@nvidia.com>
Signed-off-by: Peter Jin <pjin@nvidia.com>
Signed-off-by: Peter Jin <pjin@nvidia.com>
…ersion. Signed-off-by: Peter Jin <pjin@nvidia.com>
Signed-off-by: Peter Jin <pjin@nvidia.com>
Signed-off-by: Peter Jin <pjin@nvidia.com>
Signed-off-by: Peter Jin <pjin@nvidia.com>
…in/super-v3.5-dev-20260826 Signed-off-by: Peter Jin <pjin@nvidia.com>
Signed-off-by: Guyue Huang <guyueh@nvidia.com> Signed-off-by: Peter Jin <pjin@nvidia.com>
Signed-off-by: Peter Jin <pjin@nvidia.com>
Signed-off-by: Peter Jin <pjin@nvidia.com>
Signed-off-by: Peter Jin <pjin@nvidia.com>
Signed-off-by: Peter Jin <pjin@nvidia.com>
✅ Submodule Fast-Forward Check ResultsCheck based on commit: 0da8502 (PR #3941 from ✅ Submodules that are properly updated:Gym: ✅ PR branch is ahead of super-v3.5-posttraining branch (fast-forward) All submodule changes look good! ✨ |
vLLM 0.25 can raise a plain ValueError when a chat prompt exceeds the model context. The embedded endpoint surfaced that as HTTP 500, causing NeMo Gym to retry a deterministic request that cannot succeed. Translate only context-length-shaped ValueErrors to HTTP 400 while preserving all typed VLLMValidationError handling and re-raising unrelated ValueErrors. Keep local clamp errors recognizable by Gym and cover each exception path with unit tests. Signed-off-by: Yi-Fu Wu <yifu.wu@gmail.com> Signed-off-by: Peter Jin <pjin@nvidia.com>
✅ Submodule Fast-Forward Check ResultsCheck based on commit: a56840a (PR #3941 from ✅ Submodules that are properly updated:Gym: ✅ PR branch is ahead of super-v3.5-posttraining branch (fast-forward) All submodule changes look good! ✨ |
Signed-off-by: Peter Jin <pjin@nvidia.com>
✅ Submodule Fast-Forward Check ResultsCheck based on commit: 7fb0601 (PR #3941 from ✅ Submodules that are properly updated:Gym: ✅ PR branch is ahead of super-v3.5-posttraining branch (fast-forward) All submodule changes look good! ✨ |
…n; enable by setting `seq_logprob_error_force_on_policy: true`. This skips separate logprob computation, reducing the minimum total step time, but also subtly changes the sequence masking normalization. Signed-off-by: Peter Jin <pjin@nvidia.com>
Represent routed-expert data as logical Ray-backed slices across successive model calls, splice the next prefill route over the prior padded decode route, and preserve the segmented references while flattening training messages. Also pin the Gym correction-turn metadata fix so every call retains its router-replay identity. Signed-off-by: Yi-Fu Wu <yifu.wu@gmail.com> Signed-off-by: Peter Jin <pjin@nvidia.com>
Signed-off-by: Peter Jin <pjin@nvidia.com>
Signed-off-by: Peter Jin <pjin@nvidia.com>
…/?). Signed-off-by: Peter Jin <pjin@nvidia.com>
Signed-off-by: Peter Jin <pjin@nvidia.com>
github-actions
Bot
added
the
Documentation
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
4 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.
No description provided.