Skip to content

Pull requests: NVIDIA-NeMo/RL

New pull request New
301 Open 1,804 Closed
301 Open 1,804 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

cp: fix: cap dapo-deepseek-v3-64n8g perf test NUM_MINUTES at 240 (cluster limit) (3052) into r0.7.0 cherry-pick Run CICD
#3082 opened Jul 6, 2026 by svcnvidia-nemo-ci Contributor Loading…
1
ci: demonstrate the pyrefly baseline gate CI:Lfast Runs a fast test suite and re-use nightly `main` container (but sync dependencies to PRs version)
#3081 opened Jul 6, 2026 by terrykong Collaborator • Draft
2
ci: upgrade pyrefly to 1.1.1 and gate on a baseline file CI Relating to CI
#3080 opened Jul 6, 2026 by terrykong Collaborator • Draft
1
cp: ci: reduce number of steps for glm5.1 test case (3077) into r0.7.0 cherry-pick Run CICD
#3079 opened Jul 6, 2026 by svcnvidia-nemo-ci Contributor Loading…
2
cp: fix: fix sft-llama3.1-70b-8n8g-tp4pp2-long-megatron (3073) into r0.7.0 cherry-pick Run CICD
#3078 opened Jul 6, 2026 by svcnvidia-nemo-ci Contributor Loading…
2
chore: bump Gym submodule to v0.4.0 CI:L1 Run doctests, unit tests, and functional tests r0.7.0
#3076 opened Jul 6, 2026 by kajalj22 Contributor Loading…
7
cp: ci: fix super environment tests (3046) into r0.7.0 cherry-pick Run CICD
#3075 opened Jul 6, 2026 by svcnvidia-nemo-ci Contributor Loading…
2
fix: fix grpo-nanov3-30BA3B-3n8g-megatron_generation-async-gym CI:Lfast Runs a fast test suite and re-use nightly `main` container (but sync dependencies to PRs version)
#3074 opened Jul 6, 2026 by yuki-97 Contributor • Draft
2
ci: Bump Megatron-Bridge to a136b90 CI:L1 Run doctests, unit tests, and functional tests
#3072 opened Jul 6, 2026 by svcnvidia-nemo-ci Contributor Loading…
3
fix(checkpoint): break metric ties by recency in get_best_checkpoint_path CI:Lfast Runs a fast test suite and re-use nightly `main` container (but sync dependencies to PRs version) community-request
#3071 opened Jul 6, 2026 by tianyi-zhang-02 Loading…
3 of 4 tasks
1
5
feat(RL): NeMo-RL × Megatron ModelExpress v2 weight-refit (trainer path) community-request Documentation Improvements or additions to documentation
#3068 opened Jul 6, 2026 by KavinKrishnan • Draft
1
fix(async-grpo): add startup pipeline barrier to prevent checkpoint-resume deadlock
#3066 opened Jul 5, 2026 by snowmanwwg Contributor Loading…
1
feat(nemo-gym): opt-in truncation of non-contiguous rollout episodes
#3063 opened Jul 4, 2026 by Kh4L • Draft
2
fix(docker): pre-build megatron-core datasets helpers_cpp extension community-request waiting-on-maintainers Waiting on maintainers to respond
#3062 opened Jul 3, 2026 by dafu-wu Loading…
2 of 4 tasks
4
fix: keep math-verify reward when answer extraction fails
#3058 opened Jul 3, 2026 by zpqiu Contributor Loading…
4 tasks
1
fix: synchronize packed broadcast streams before return
#3055 opened Jul 3, 2026 by haitian-nvidia Loading…
3 tasks done
2
fix: capture GRPO reference policy from base weights, not resumed checkpoint community-request waiting-on-maintainers Waiting on maintainers to respond
#3048 opened Jul 2, 2026 by CanadaApollo6 Loading…
3 of 4 tasks
1
1
feat: per-step inline HF export for Megatron checkpoints community-request waiting-on-maintainers Waiting on maintainers to respond
#3045 opened Jul 2, 2026 by waleko Loading…
2 tasks done
1
refactor(eval): migrate eval datasets to response registry CI:L1 Run doctests, unit tests, and functional tests Documentation Improvements or additions to documentation
#3039 opened Jul 2, 2026 by NolenLiang Contributor Loading…
5
fix(grpo): restore NeMo Gym response logging guard in async training community-request waiting-on-maintainers Waiting on maintainers to respond
#3038 opened Jul 2, 2026 by MarkXCloud Loading…
3 of 4 tasks
2
fix: add validation for make_sequence_length_divisible_by CI:L1 Run doctests, unit tests, and functional tests
#3028 opened Jul 1, 2026 by NolenLiang Contributor Loading…
4
ci: Add nightly GB200 super test CI:Lfast Runs a fast test suite and re-use nightly `main` container (but sync dependencies to PRs version) r0.7.0
#3016 opened Jun 30, 2026 by ashors1 Contributor Loading…
4 tasks
4
feat: Llama true-on-policy mode
#3015 opened Jun 30, 2026 by guyueh1 Contributor • Draft
4 tasks
1
fix: reduce NUM_MINUTES to 240 in dapo-deepseek-v3-64n8g.v2 test
#3014 opened Jun 30, 2026 by kajalj22 Contributor Loading…
1 task
1
fix: stabilize GB200 16n4g issue-2579 repro CI:Lfast Runs a fast test suite and re-use nightly `main` container (but sync dependencies to PRs version)
#3013 opened Jun 30, 2026 by jinglinglingling Contributor Loading…
1
2
ProTip! Filter pull requests by the default branch with base:main.