Skip to content

Pull requests: NovaSky-AI/SkyRL

New pull request New
162 Open 1,286 Closed
162 Open 1,286 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

Make packed sequence alignment FP8-safe
#1828 opened Jun 24, 2026 by jinghanyao1-hub Collaborator Loading…
11
[fix] Memory-map dataset per SFT tokenize worker to avoid cache stampede
#1826 opened Jun 23, 2026 by dinhxuanvu Contributor Loading…
3
UI/UX: Poor Error Handling UX in Async Dispatcher
#1823 opened Jun 22, 2026 by tomaioo Loading…
2
fix(generators): compute batched rollout metrics from truncated responses
#1821 opened Jun 21, 2026 by EazyReal Loading…
fix(tracking): finalize MLflow runs on SIGTERM instead of leaving them RUNNING
#1819 opened Jun 20, 2026 by EazyReal Loading…
1
6
[megatron] Enable Nemotron-3-Ultra-550B GRPO RL + fix multi-rank (EP>16/PP>2) weight sync
#1816 opened Jun 19, 2026 by erictang000 Collaborator Loading…
4
[algorithm] add rollout KL loss + fix padding-microbatch field drops
#1811 opened Jun 19, 2026 by erictang000 Collaborator Loading…
3
[train] Async batch collation (double-buffering) for the SFT trainer
#1809 opened Jun 18, 2026 by dyurk-lila Loading…
2 tasks done
2
[train] Vectorize controller-side training-batch collation (SFT + RL)
#1808 opened Jun 18, 2026 by dyurk-lila Loading…
3 tasks done
4
[train] Skip building unused per-token loss_fn_outputs when the caller does not consume them
#1807 opened Jun 18, 2026 by dyurk-lila Loading…
[megatron] Stream ChunkedDistributedLogprob.backward into a preallocated buffer (lower peak memory)
#1806 opened Jun 18, 2026 by dyurk-lila Loading…
@erictang000
2
[megatron] Accept dtype-string optimizer_config_kwargs (coerce exp_avg_dtype etc. to torch.dtype)
#1805 opened Jun 18, 2026 by dyurk-lila Loading…
@erictang000
2
[chore]Upgrade vllm to 0.23.0 run_h100_gpu_ci Run H100 GPU CI run_train_gpu_ci run_train_megatron_gpu_ci run_train_old_inference_gpu_ci
#1800 opened Jun 17, 2026 by SumanthRH Member Loading…
5
[Fix] vLLM Metrics Scrapper throughput calculation
#1794 opened Jun 16, 2026 by zanderjiang Loading…
5
rename adv_estimator param to advantage_estimator in compute_advantages_and_returns
#1793 opened Jun 16, 2026 by KTanmay1 Loading…
1 task
1
[train] Save HF processor on checkpoint export for VLMs
#1785 opened Jun 14, 2026 by dinhxuanvu Contributor Loading…
1 of 2 tasks
10
[fix] Honor served_model_name and surface HTTP errors in RemoteInferenceEngine
#1783 opened Jun 13, 2026 by discobot Contributor Loading…
1
3
[fix] Use masked mean in advantage batch normalization
#1782 opened Jun 12, 2026 by discobot Contributor Loading…
1
3
Support top-K distillation (SDFT/OPSD): teacher top-K sampling + soft-target CE training
#1777 opened Jun 11, 2026 by atemaguer Contributor Loading…
4
Rollout Routing Replay (R3) for the fsdp backend, using vllm==0.22
#1772 opened Jun 10, 2026 by jamesbraza Contributor Loading…
5
Attach the failed actor's log tail when inference engines die during init
#1771 opened Jun 10, 2026 by jamesbraza Contributor Loading…
1
9
[wip] add router replay tests to CI
#1770 opened Jun 10, 2026 by erictang000 Collaborator • Draft
[megatron] Fused LM-head log-prob + entropy (avoid full [*, seq, vocab] logit materialization)
#1765 opened Jun 9, 2026 by dyurk-lila Loading…
4 tasks done
@erictang000
5
[WIP] RDT skyrl integration
#1753 opened Jun 5, 2026 by hao-aaron Collaborator • Draft
[train] VLM SFT support on Megatron backend (Qwen3-VL)
#1752 opened Jun 4, 2026 by s-chundi Loading…
10
ProTip! Exclude everything labeled bug with -label:bug.