Skip to content

Pull requests: NovaSky-AI/SkyRL

New pull request New
272 Open 1,552 Closed
272 Open 1,552 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

[WIP] Add DAPO example for the Tinker codepath
#2248 opened Sep 20, 2026 by xinyuangui2 Contributor • Draft
feat(megatron): add LoRA-only IQuest LoopCoder support
#2245 opened Sep 19, 2026 by j316chuck Contributor Loading…
9
[algorithm] Score centering for off-policy correction (arXiv 2609.20807)
#2244 opened Sep 19, 2026 by erictang000 Collaborator • Draft
[tinker] Require tinker SDK >= 0.25.0 and drop the JSON wire path for fwd_bwd/sample
#2243 opened Sep 18, 2026 by erictang000 Collaborator Loading…
4
feat(sample-support): report the policy entropy over the recorded support x-lila
#2241 opened Sep 18, 2026 by erictang000 Collaborator • 11/11 Loading…
3
feat(fsdp): score policy logprobs over the recorded support under Ulysses SP x-lila
#2240 opened Sep 18, 2026 by erictang000 Collaborator • 10/11 Loading…
6
feat(sample-support): score policy logprobs over the recorded support end to end x-lila
#2239 opened Sep 18, 2026 by erictang000 Collaborator • 9/11 Loading…
4
perf(sample-support): carry the captured support to the trainer as one packed field x-lila
#2238 opened Sep 18, 2026 by erictang000 Collaborator • 8/11 Loading…
@SumanthRH
6
feat(sample-support): capture the sampler's bounded support and trace it across turns x-lila
#2237 opened Sep 18, 2026 by erictang000 Collaborator • 7/11 Loading…
@SumanthRH
9
perf(batch): ship large per-token side channels to the trainer zero-copy x-lila
#2236 opened Sep 18, 2026 by erictang000 Collaborator • 6/11 Loading…
@SumanthRH
4
perf(r3): route rollout experts to the trainer packed with cu_seqlens x-lila
#2235 opened Sep 18, 2026 by erictang000 Collaborator • 5/11 Loading…
@SumanthRH
12
refactor(wire): generic packed-ndarray codec and an N-field response-body splice x-lila
#2234 opened Sep 18, 2026 by erictang000 Collaborator • 4/11 Loading…
@avigyabb
5
perf(r3): collate routes through a container-aware pool, and refuse replay under VPP x-lila
#2233 opened Sep 18, 2026 by erictang000 Collaborator • 3/11 Loading…
@erictang000
4
support Megatron GRPO with vLLM on AMD
#2230 opened Sep 18, 2026 by amd-pratmish Loading…
3 tasks
39
[eval][2/3] Async Evals
#2227 opened Sep 17, 2026 by kyuds Collaborator • 4/4 Loading…
13
[megatron] GLM-5.3-Flash + in-memory LoRA sync in one tree
#2226 opened Sep 17, 2026 by j316chuck Contributor Loading…
2 tasks done
11
[docs] Document Megatron LoRA offload/backload semantics under colocate_all
#2224 opened Sep 17, 2026 by avigyabb Collaborator Loading…
3
feat: add IsoExec full-distribution comparison
#2223 opened Sep 17, 2026 by Future-Outlier • Draft
[weight_sync] In-memory LoRA adapter sync for Megatron (lora.sync_mode=memory)
#2221 opened Sep 16, 2026 by avigyabb Collaborator • Draft
[Native LoRA 3/3] Integrate persistent GPU adapter transport
#2220 opened Sep 16, 2026 by hershg Contributor • Draft
[Native LoRA 2/3] Add atomic named-generation lifecycle
#2219 opened Sep 16, 2026 by hershg Contributor • Draft
[Native LoRA 1/3] Add immutable layout and ownership planning
#2218 opened Sep 16, 2026 by hershg Contributor • Draft
[skyrl-train] Add reward-variance top-p/top-k filtering
#2217 opened Sep 16, 2026 by ZihanWang314 Loading…
@erictang000
16
[megatron] Qwen3.8-Flash-Next (qwen4_exp)
#2216 opened Sep 16, 2026 by casper-hansen Contributor • Draft
fix(megatron): restore saved scheduler progress without accumulation
#2214 opened Sep 16, 2026 by hershg Contributor • Draft
ProTip! Follow long discussions with comments:>50.