-
Notifications
You must be signed in to change notification settings - Fork 431
Pull requests: NVIDIA-NeMo/RL
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
ci: Bump Megatron-Bridge to 81d6fd6
CI:L1
Run doctests, unit tests, and functional tests
#2880
opened Jun 22, 2026 by
svcnvidia-nemo-ci
Contributor
Loading…
fix(megatron): only build the MTP loss mask when MTP is enabled
#2876
opened Jun 19, 2026 by
yfw
Contributor
Loading…
4 tasks
feat: add Mistral Medium 3.5 (128B) text-only DAPO support
CI:L1
Run doctests, unit tests, and functional tests
#2875
opened Jun 19, 2026 by
sharonyu-115
Contributor
Loading…
3 of 4 tasks
fix: fix several tests by pin triton moe backend
CI:Lfast
Runs a fast test suite and re-use nightly `main` container (but sync dependencies to PRs version)
#2873
opened Jun 19, 2026 by
yuki-97
Contributor
Loading…
fix(megatron): honor policy.logprob_chunk_size in the training loss path
community-request
waiting-on-maintainers
Waiting on maintainers to respond
#2872
opened Jun 18, 2026 by
kaloyan-inherent
Contributor
Loading…
DRAT: fix: run Nemotron Nano v2 workplace assistant recipe
#2868
opened Jun 18, 2026 by
snowmanwwg
Contributor
Loading…
4 tasks
DRAFT fix: prefer real NeMo-Gym package in actor
#2867
opened Jun 18, 2026 by
snowmanwwg
Contributor
Loading…
4 tasks
feat(megatron): add large-scale MoE tuning knobs and longer PG timeout
community-request
waiting-on-maintainers
Waiting on maintainers to respond
#2866
opened Jun 18, 2026 by
dafu-wu
Loading…
1 of 4 tasks
fix: tokenize system-led conversations for templates requiring a user turn
community-request
waiting-on-maintainers
Waiting on maintainers to respond
#2864
opened Jun 17, 2026 by
bzantium
Loading…
3 of 4 tasks
feat: add NCCL timeout config, stale ZMQ socket cleanup, OmegaConfig resolvers
community-request
waiting-on-maintainers
Waiting on maintainers to respond
#2862
opened Jun 17, 2026 by
puneeshkhanna
Loading…
1 of 4 tasks
fix(logger): summarize list-valued metrics to avoid MLflow key explosion
community-request
waiting-on-maintainers
Waiting on maintainers to respond
#2861
opened Jun 17, 2026 by
mrm-196
Contributor
Loading…
3 of 4 tasks
fix(data): stabilize multi-turn chat chunking and tokenization
CI:Lfast
Runs a fast test suite and re-use nightly `main` container (but sync dependencies to PRs version)
#2856
opened Jun 17, 2026 by
jinglinglingling
Contributor
Loading…
docs(xtoken): X-Token distillation guide and README updates
Documentation
Improvements or additions to documentation
#2854
opened Jun 16, 2026 by
avenkateshha
Contributor
Loading…
test: add vLLM HTTP logprobs contract test for NeMo-Gym capture
CI:Lfast
Runs a fast test suite and re-use nightly `main` container (but sync dependencies to PRs version)
#2845
opened Jun 16, 2026 by
ananthsub
Contributor
Loading…
feat: add vLLM prefix cache and preemption metrics
community-request
waiting-on-maintainers
Waiting on maintainers to respond
#2843
opened Jun 16, 2026 by
puneeshkhanna
Loading…
1 of 4 tasks
feat(ppo): Megatron value-model sequence packing + context parallelism
CI:L1
Run doctests, unit tests, and functional tests
#2839
opened Jun 16, 2026 by
bg51717
Contributor
Loading…
3 of 4 tasks
test(data_plane): session-scope mooncake fixtures
CI:Lfast
Runs a fast test suite and re-use nightly `main` container (but sync dependencies to PRs version)
#2838
opened Jun 16, 2026 by
ZhiyuLi-Nvidia
Contributor
Loading…
feat: Support for dtensor ppo
CI:Lfast
Runs a fast test suite and re-use nightly `main` container (but sync dependencies to PRs version)
#2837
opened Jun 16, 2026 by
fujial-code
Loading…
feat: Support Chunked Linear Fusion for GRPO
community-request
Documentation
Improvements or additions to documentation
waiting-on-maintainers
Waiting on maintainers to respond
#2833
opened Jun 16, 2026 by
pengdurice
Contributor
Loading…
4 tasks done
feat: super-v3 recipe and docs
CI:L0
Run doctests and unit tests
Documentation
Improvements or additions to documentation
super-v3
#2829
opened Jun 15, 2026 by
macandro96
Contributor
Loading…
4 tasks
Previous Next
ProTip!
Follow long discussions with comments:>50.