-
Notifications
You must be signed in to change notification settings - Fork 403
Pull requests: NovaSky-AI/SkyRL
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[Tinker] Execute GSPO in one native forward-backward pass
#2043
opened Aug 14, 2026 by
yapdianang
Contributor
•
Draft
[deps][megatron] bump megatron-bridge to 0.7.0 and megatron-core to 0.20.0
#2042
opened Aug 14, 2026 by
erictang000
Collaborator
Loading…
[dependencies] Upgrade to cuda 13 by default
#2040
opened Aug 14, 2026 by
erictang000
Collaborator
Loading…
1 task
[docs] 10/n towards Kimi K2.6: Kimi K2.7-Code INT4 QAT + LoRA end-to-end example
#2032
opened Aug 13, 2026 by
casper-hansen
Contributor
•
Draft
[tinker] 9/n towards Kimi K2.6: base-model sampling and colocated engine wake/offload
#2031
opened Aug 13, 2026 by
casper-hansen
Contributor
Loading…
[tinker] 8/n towards Kimi K2.6: protobuf forward_backward wire format (SDK >= 0.25)
#2030
opened Aug 13, 2026 by
casper-hansen
Contributor
Loading…
[tinker] 7/n towards Kimi K2.6: apply client LoRA config before config validation
#2029
opened Aug 13, 2026 by
casper-hansen
Contributor
Loading…
[deps] 6/n towards Kimi K2.6: CUDA-13 deploy stack for Blackwell-Ultra (sm103 / B300)
#2028
opened Aug 13, 2026 by
casper-hansen
Contributor
Loading…
[megatron] 5/n towards Kimi K2.6: Kimi K2.5-family (KimiK25ForConditionalGeneration) support with INT4 QAT + LoRA
#2027
opened Aug 13, 2026 by
casper-hansen
Contributor
•
Draft
[megatron] 4/n towards Kimi K2.6: make merge_lora=false adapter sync viable for very large MoE
#2026
opened Aug 13, 2026 by
casper-hansen
Contributor
Loading…
[megatron] 3/n towards Kimi K2.6: skip MLA THD value pad on sm100+ to keep fused attention trainable
#2025
opened Aug 13, 2026 by
casper-hansen
Contributor
Loading…
[feat][checkpoint] Allow selective state restore on resume
#2010
opened Aug 10, 2026 by
bvolpato
Contributor
Loading…
[fix][fsdp] Resume bitsandbytes checkpoints strictly
#2007
opened Aug 10, 2026 by
bvolpato
Contributor
Loading…
[fix][data] Apply chat-template kwargs during filtering
#2006
opened Aug 10, 2026 by
bvolpato
Contributor
Loading…
[fix][train] Defer W&B authentication to SDK
#2005
opened Aug 10, 2026 by
bvolpato
Contributor
Loading…
perf(sample-support): reuse replay scores for entropy
#2004
opened Aug 7, 2026 by
dyurk-lila
Contributor
Loading…
Initialize base inference before training model creation
#2003
opened Aug 7, 2026 by
j316chuck
Contributor
Loading…
[tinker] Keep the multi-tenant LoRA runtime warm on last unload
#2001
opened Aug 7, 2026 by
avigyabb
Collaborator
Loading…
[fix][megatron] Make multi-LoRA adapter swaps safe against colocation offload
#2000
opened Aug 7, 2026 by
erictang000
Collaborator
Loading…
Previous Next
ProTip!
Type g i on any issue or pull request to go back to the issue listing page.