Skip to content

Pull requests: NovaSky-AI/SkyRL

New pull request New
94 Open 940 Closed
94 Open 940 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

[megatron] enable bucketed weight sync for non-colocated nccl weight sync in megatron
#1324 opened Mar 14, 2026 by erictang000 Loading…
4
[train][fix] Fix concurrency limitations in the new inference codepath
#1320 opened Mar 13, 2026 by SumanthRH • Draft
1
WIP Qwen 3.5 Dense FSDP
#1313 opened Mar 11, 2026 by CharlieFRuan • Draft
[fix] Fix placement group bundle ordering for inference engines
#1308 opened Mar 10, 2026 by SumanthRH Loading…
1
10
[bug][train] Fix max_seq_len calculation
#1303 opened Mar 10, 2026 by tamoghnokandar Loading…
1
7
MoE/FSDP2 training fixes, Anthropic Messages API, and robustness improvements
#1298 opened Mar 9, 2026 by ashutoshuiuc Loading…
4 tasks done
2
25
[wip] loss reduction
#1296 opened Mar 9, 2026 by justinvyu • Draft
2
[train][FullyAsync] Implement customizable weight sync frequency
#1293 opened Mar 8, 2026 by tamoghnokandar Loading…
4
feat(train): prefix-aware merge for step-wise trajectories (#1277)
#1289 opened Mar 6, 2026 by deepsheth3 Loading…
@CharlieFRuan
11
[tinker][SkyRL] Add sample and renderer API to new inference client
#1287 opened Mar 6, 2026 by nithinvc Loading…
39
[train] Make TrainingInputBatch to PAD only to left, hence response tensors be right-aligned
#1285 opened Mar 5, 2026 by CharlieFRuan Loading…
3 tasks done
@erictang000
9
WIP VLM SFT Training via SkyRL with SkyRL train
#1282 opened Mar 5, 2026 by nithinvc • Draft
13
[train] Add validation for step-wise GeneratorOutput
#1281 opened Mar 5, 2026 by CharlieFRuan • Draft
3 tasks done
WIP: return_dict=False fixes + H200 validation scripts
#1280 opened Mar 5, 2026 by tyler-griggs • Draft
[train] Add sampler-as-reference KL option
#1263 opened Mar 3, 2026 by tyler-griggs • Draft
2 tasks
[train] Add KL-in-advantages mode
#1262 opened Mar 3, 2026 by tyler-griggs • Draft
2 tasks
[train] Add importance weight diagnostics and fix IS loss overflow
#1261 opened Mar 3, 2026 by tyler-griggs • Draft
2 tasks
[train] Add "sum" loss reduction mode
#1260 opened Mar 3, 2026 by tyler-griggs • Draft
1 task
[train] Add DRO (Direct Reward Optimization) policy loss
#1259 opened Mar 3, 2026 by tyler-griggs • Draft
2 tasks
[WIP][Harbor] Add step-wise training for Harbor
#1257 opened Mar 3, 2026 by CharlieFRuan • Draft
Bump transformers to >=5.0.0 for GLM-4.7-Flash
#1241 opened Mar 2, 2026 by tyler-griggs Loading…
2
Add llm_as_a_judge_local example with frozen vLLM reward model
#1208 opened Feb 25, 2026 by ghShu Loading…
22
[tx] Add Qwen3 VL model tx
#1196 opened Feb 23, 2026 by tamoghnokandar Loading…
23
Use the last LoRA path in the vLLM inference engine instead of "dummy_lora_path"
#1188 opened Feb 20, 2026 by ebronstein Loading…
@SumanthRH
6
Restrict mini-swe-agent version to <2.0
#1180 opened Feb 19, 2026 by bluerose73 Loading…
5
ProTip! Type g p on any issue or pull request to go back to the pull request listing page.