-
Notifications
You must be signed in to change notification settings - Fork 74
Pull requests: nvidia-cosmos/cosmos-rl
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
fix(rollout): keep the heartbeat alive until unregister, bound the unregister retry
#759
opened Sep 19, 2026 by
maximilianigl
Contributor
Loading…
feat: normalize sample- and episode-weighted optimizer objectives
#758
opened Sep 19, 2026 by
shuangwu
Collaborator
Loading…
Fail fast on checkpoint resume metadata disagreement
#756
opened Sep 18, 2026 by
shuangwu
Collaborator
Loading…
fix(vla): make evaluation, resume, and shutdown reliable
#755
opened Sep 18, 2026 by
shuangwu
Collaborator
Loading…
feat(payload): bound NCCL receive memory with explicit consumer leases
#754
opened Sep 18, 2026 by
shuangwu
Collaborator
Loading…
feat: coordinate expanded training schedules with recoverable empty contributions
#753
opened Sep 18, 2026 by
shuangwu
Collaborator
Loading…
feat: add controller-owned identified completion admission
#752
opened Sep 18, 2026 by
shuangwu
Collaborator
Loading…
fix: own payload transport attachment and bounded teardown
#751
opened Sep 18, 2026 by
shuangwu
Collaborator
Loading…
feat(controller): support graceful application-requested stop
#750
opened Sep 18, 2026 by
shuangwu
Collaborator
Loading…
Support consistent application checkpoint recovery and safe sampling replay
#749
opened Sep 18, 2026 by
shuangwu
Collaborator
Loading…
feat(logging): configure W&B identity and synchronize run ownership
#748
opened Sep 18, 2026 by
shuangwu
Collaborator
Loading…
fix: enforce transport prefetch deadlines and contain explicit fatal exits
#747
opened Sep 17, 2026 by
shuangwu
Collaborator
Loading…
fix(policy): condition shutdown ordering on fast reap
#726
opened Aug 17, 2026 by
maximilianigl
Contributor
•
Draft
4 tasks done
feat(rollout): give the rollout backend a pause for weight sync
#725
opened Aug 5, 2026 by
karkuspeter
Loading…
fix(rollout): hold payload egress for a whole weight sync
#724
opened Aug 5, 2026 by
karkuspeter
Loading…
perf(rollout): group the synchronous trainable-subset R2R broadcast
#723
opened Aug 5, 2026 by
karkuspeter
Loading…
fix(vllm_rollout): Filter prompts exceeding max_model_len instead of …
#690
opened May 21, 2026 by
Pritiks23
Loading…
feat: mimic pi05 config and episode control
#584
opened Feb 13, 2026 by
littlespray
Contributor
Loading…
Bugfixes - Custom hooks; optimizer warmup as epochs; dataset cache as a separate configuration; validation at end of epoch fix
#505
opened Jan 20, 2026 by
ramanathan831
Loading…
[DINO] [1/n] Copy of model modules, data loader, and kernels
#386
opened Nov 14, 2025 by
heslami
Contributor
Loading…
ProTip!
What’s not been updated in a month: updated:<2026-08-18.