Skip to content

Pull requests: NovaSky-AI/SkyRL

New pull request New
67 Open 659 Closed
67 Open 659 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

[skyrl-train] Add SFT support via forward_backward(loss_fn="cross_entropy")
#961 opened Jan 26, 2026 by tyler-griggs Loading…
3
[WIP] [tx] Make it easy to run on a multi-node Ray cluster
#955 opened Jan 25, 2026 by Future-Outlier Loading…
6
Separate infrastructure logs from training progress
#953 opened Jan 25, 2026 by tyler-griggs Loading…
2 of 3 tasks
4
Add SKYRL_LOG_LEVEL env var and suppress verbose Ray logs by default
#947 opened Jan 25, 2026 by tyler-griggs Loading…
1 of 2 tasks
2
Add Pydantic Configuration System with CLI Override Support
#946 opened Jan 25, 2026 by tyler-griggs Loading…
10
Stage 5: Add TinkerTrainingAdapter for forward_backward/optim_step
#938 opened Jan 24, 2026 by tyler-griggs Loading…
7
[skyrl-train][inference] HTTP Inference Integration (Feature-Flagged) 4/N
#931 opened Jan 24, 2026 by kouroshHakha • Draft
[skyrl-train] Fix loss reduction by moving normalization to the advantage computation
#925 opened Jan 23, 2026 by justinvyu • Draft
1
Add loss_fn parameterization to forward_backward
#924 opened Jan 23, 2026 by tyler-griggs • Draft
3
[train][Harbor][1/N] Upstream Harbor integration
#923 opened Jan 22, 2026 by CharlieFRuan • Draft
DSPy integration
#921 opened Jan 22, 2026 by Harryllh Loading…
[tx] add per-layer gradient checkpointing with scan for memory-efficient training tx
#906 opened Jan 21, 2026 by raulchen Loading…
4 tasks done
3
[skyrl-train][refactor] 2/N Inference Server Refactor -- RemoteInferenceClient
#904 opened Jan 20, 2026 by kouroshHakha Loading…
@CharlieFRuan
6
Unify Megatron and FSDP training interfaces with forward_backward + optim_step
#901 opened Jan 20, 2026 by tyler-griggs Loading…
1 of 2 tasks
4
[train] Enable RayPrometheusStatLogger for async vLLM engine
#900 opened Jan 20, 2026 by kanwang Loading…
@CharlieFRuan
6
[skyrl-train] Add GDPO Support to PPO Utils
#897 opened Jan 19, 2026 by devpatelio Loading…
2 tasks
3
[tx] Implement cutlass kernel for ragged_dot with group_offset tx
#896 opened Jan 19, 2026 by pcmoritz Loading…
4
[tx] DeepseekV3 implementation tx
#889 opened Jan 16, 2026 by tanmaysachan Loading…
5 tasks done
13
[skyrl-train] Update Megatron + LoRA to use PEFT only bridge
#885 opened Jan 15, 2026 by erictang000 • Draft
1 of 3 tasks
[algorithm] Sampling mask from Inference Engine Applies to Trainer
#883 opened Jan 15, 2026 by devpatelio Loading…
9
[skyrl-train] Add example for on-policy distillation with GOLD
#881 opened Jan 15, 2026 by pbokc • Draft
2
[tx] Cuda tile for expert parallelism tx
#880 opened Jan 15, 2026 by agolajko • Draft
4 tasks
skyrl-agent osworld integration
#875 opened Jan 14, 2026 by shuo-yuan Loading…
2
[tx] Add experimental SkyRL-train backend that supports SFT tx
#871 opened Jan 13, 2026 by pcmoritz Loading…
9
[SkyRL-tx] Integrate ragged_dot pallas kernel with group_offset support tx
#867 opened Jan 13, 2026 by tanmaysachan Loading…
2 of 4 tasks
5
ProTip! Add no:assignee to see everything that’s not assigned.