Skip to content

Pull requests: NovaSky-AI/SkyRL

New pull request New
231 Open 1,453 Closed
231 Open 1,453 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(tinker): avoid SQLite pool exhaustion under rollout load
#2097 opened Aug 27, 2026 by hershg • Draft
[fix][megatron] Restore token-batched output order
#2096 opened Aug 27, 2026 by hershg Loading…
3
[fix][megatron] Include Fast Hadamard Transform dependency
#2095 opened Aug 27, 2026 by hershg Loading…
6
[tinker] Make forwarding inference timeout configurable
#2094 opened Aug 27, 2026 by j316chuck Contributor Loading…
2
Target GatedDeltaNet projections with Megatron LoRA
#2093 opened Aug 26, 2026 by kalectory Contributor Loading…
@avigyabb
2
[fix][sft] Locate the assistant response boundary instead of subtracting prompt length
#2092 opened Aug 24, 2026 by bharatmekala Loading…
3 tasks done
5
fix(r3): balance replay padding across expert ranks
#2091 opened Aug 23, 2026 by EazyReal Contributor Loading…
2
[fsdp] Trim dense padding per final microbatch
#2089 opened Aug 22, 2026 by EazyReal Contributor Loading…
6
[train] FSDP2: wrap only input embeddings, not every nn.Embedding (fixes glm4v)
#2088 opened Aug 22, 2026 by dzorlu Loading…
1
2
feat(sample-support): report the policy entropy over the recorded support
#2086 opened Aug 22, 2026 by dyurk-lila Contributor Loading…
2
feat(fsdp): score policy logprobs over the recorded support under Ulysses SP
#2085 opened Aug 22, 2026 by dyurk-lila Contributor Loading…
4
feat(sample-support): score policy logprobs over the recorded support end to end
#2084 opened Aug 22, 2026 by dyurk-lila Contributor Loading…
6
perf(sample-support): carry the captured support to the trainer as one packed field
#2083 opened Aug 22, 2026 by dyurk-lila Contributor Loading…
6
feat(sample-support): capture the sampler's bounded support and trace it across turns
#2082 opened Aug 22, 2026 by dyurk-lila Contributor Loading…
7
perf(batch): ship large per-token side channels to the trainer zero-copy
#2081 opened Aug 22, 2026 by dyurk-lila Contributor Loading…
8
perf(r3): route rollout experts to the trainer packed with cu_seqlens
#2080 opened Aug 22, 2026 by dyurk-lila Contributor Loading…
10
refactor(wire): generic packed-ndarray codec and an N-field response-body splice
#2079 opened Aug 22, 2026 by dyurk-lila Contributor Loading…
10
perf(r3): collate routes through a container-aware pool, and refuse replay under VPP
#2078 opened Aug 22, 2026 by dyurk-lila Contributor Loading…
6
[fix][generators] VLM generator: honor generator.chat_template in renders + image-token/feature integrity guard
#2077 opened Aug 21, 2026 by dzorlu • Draft
1
2
FP8 RL MXFP8 Weight Sync
#2072 opened Aug 20, 2026 by YJHMITWEB Contributor Loading…
@jinghanyao1-hub
15
Add PiSSA initialization for Megatron LoRA adapters
#2069 opened Aug 19, 2026 by kalectory Contributor Loading…
@SumanthRH
12
[tinker] 17/n towards Kimi K2.6: run Tinker API forward requests through the training loss path
#2068 opened Aug 19, 2026 by casper-hansen Contributor Loading…
@avigyabb
3
[tinker] 15/n towards Kimi K2.6: continuous sampling for the Tinker API engine under colocate_all
#2066 opened Aug 19, 2026 by casper-hansen Contributor Loading…
6
[tinker] 14/n towards Kimi K2.6: sleep colocated engines around Tinker API checkpoint ops
#2065 opened Aug 19, 2026 by casper-hansen Contributor Loading…
2
[tinker] 13/n towards Kimi K2.6: adapter-only colocated weight sync for Tinker API LoRA (merge_lora=false)
#2064 opened Aug 19, 2026 by casper-hansen Contributor Loading…
7
ProTip! Find all pull requests that aren't related to any open issues with -linked:issue.