Skip to content

Pull requests: NVIDIA-NeMo/RL

New pull request New
552 Open 2,374 Closed
552 Open 2,374 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

fix(async): do not let the vLLM prefix cache go stale across weight updates Documentation Improvements or additions to documentation
#4018 opened Sep 5, 2026 by linj-glitch Loading…
4 tasks done
1
ci: Bump Megatron-Bridge to da6680c CI:L1 Run doctests, unit tests, and functional tests
#4017 opened Sep 5, 2026 by svcnvidia-nemo-ci Contributor Loading…
3
test: [3/3] add L3 Nano 4B Gym training E2E
#4015 opened Sep 5, 2026 by yaoyu-33 Contributor • Draft
7
test: [2/3] add L2 Gym-to-RL rollout acceptance CI:L0 Run doctests and unit tests
#4014 opened Sep 5, 2026 by yaoyu-33 Contributor • Draft
7
fix(grpo): preserve prompt groups across rollout paths
#4013 opened Sep 5, 2026 by ananthsub Contributor Loading…
1
feat: add opd/tropd metrics for super v3.5 Documentation Improvements or additions to documentation
#4011 opened Sep 5, 2026 by kbhardwaj-nvidia • Draft
1
feat(mllm-v2): Support single-controller for Megatron-Inference. CI:L1 Run doctests, unit tests, and functional tests
#4009 opened Sep 5, 2026 by cspades Contributor Loading…
2 of 4 tasks
@cspades
24
feat: shared-prefix GRPO training community-request
#4008 opened Sep 4, 2026 by zaifengp-maker • Draft
2 of 4 tasks
1
feat: length-aware rollout scheduling for cross-DP generation CI Relating to CI community-request Documentation Improvements or additions to documentation
#4007 opened Sep 4, 2026 by zaifengp-maker • Draft
2 of 4 tasks
1
feat(vlm): mixed image+video Gym manifests and video media alignment verification community-request
#4005 opened Sep 4, 2026 by pulkitkumar95 Loading…
3 of 4 tasks
2
fix(ray.sub): revalidate log dir before sentinel existence checks community-request
#4004 opened Sep 4, 2026 by pulkitkumar95 Loading…
2 of 4 tasks
2
fix(megatron): make mtp_num_layers=0 strip MTP for repeated-layer checkpoints community-request
#4003 opened Sep 4, 2026 by pulkitkumar95 Loading…
2 of 4 tasks
2
refactor(docker): read the actor venv list from a Python leaf module CI:L1 Run doctests, unit tests, and functional tests Documentation Improvements or additions to documentation
#4002 opened Sep 4, 2026 by terrykong Collaborator Loading…
12
test: disable classifier dropout in the tiny Qwen2 test checkpoint community-request
#4000 opened Sep 4, 2026 by tigist-far Loading…
4 tasks done
1
feat(fp8): pass custom quantizer factory to MCore CI:Lfast Runs a fast test suite and re-use nightly `main` container (but sync dependencies to PRs version)
#3998 opened Sep 4, 2026 by jepio Contributor Loading…
3 of 4 tasks
3
fix(dtensor): keep v1 model loading on Transformers classes CI:Lfast Runs a fast test suite and re-use nightly `main` container (but sync dependencies to PRs version)
#3997 opened Sep 4, 2026 by jepio Contributor Loading…
3 of 4 tasks
7
feat(gdn): Pass GDN kernel backend to Mcore CI:Lfast Runs a fast test suite and re-use nightly `main` container (but sync dependencies to PRs version)
#3995 opened Sep 4, 2026 by vasunvidia Loading…
4 tasks
5
docs: add a sampling-params reference for the generation backends community-request Documentation Improvements or additions to documentation
#3994 opened Sep 4, 2026 by yupengtang Loading…
2 of 4 tasks
1
1
fix(generation): report no OpenAI server as an empty list community-request
#3993 opened Sep 4, 2026 by yupengtang Loading…
3 of 4 tasks
1
1
feat: add Qwen3.8 Flash Next support Documentation Improvements or additions to documentation
#3992 opened Sep 4, 2026 by zpqiu Contributor • Draft
4 of 8 tasks
1
2
fix(dynamo): emit each generation metric as a single series community-request
#3988 opened Sep 4, 2026 by yupengtang Loading…
3 of 4 tasks
1
fix(grpo): reject epoch bounds that train zero steps community-request
#3986 opened Sep 3, 2026 by bzantium Loading…
2 of 4 tasks
1
fix(refit): release trainer memory before collective export CI:L1 Run doctests, unit tests, and functional tests Documentation Improvements or additions to documentation
#3984 opened Sep 3, 2026 by seonjinn Contributor Loading…
3 of 4 tasks
4
feat: skip validation and reuse pre-converted arrow cache CI:Lfast Runs a fast test suite and re-use nightly `main` container (but sync dependencies to PRs version)
#3983 opened Sep 3, 2026 by macandro96 Contributor Loading…
4 tasks
2
fix(runtime): honor system Python for all actor environments CI:Lfast Runs a fast test suite and re-use nightly `main` container (but sync dependencies to PRs version)
#3982 opened Sep 3, 2026 by jepio Contributor Loading…
4 tasks done
6
ProTip! Adding no:label will show everything without a label.