-
Notifications
You must be signed in to change notification settings - Fork 603
Pull requests: areal-project/AReaL
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
fix(reward): validate scorer configuration before PRM execution
#1727
opened Sep 17, 2026 by
Le8r0nJames
Collaborator
Loading…
4 of 15 tasks
fix(recover): restore colocated inference before admission
#1726
opened Sep 17, 2026 by
Le8r0nJames
Collaborator
Loading…
4 of 15 tasks
fix(proxy): preserve configured chat template defaults
#1725
opened Sep 17, 2026 by
Le8r0nJames
Collaborator
Loading…
4 of 15 tasks
fix(examples): retain attributed Arena failures with zero reward
#1724
opened Sep 17, 2026 by
Le8r0nJames
Collaborator
Loading…
4 of 15 tasks
fix(engine): adapt colocated AWEX to SGLang scheduler APIs
#1723
opened Sep 17, 2026 by
Le8r0nJames
Collaborator
Loading…
4 of 15 tasks
feat(infra): add opt-in sample-level rollout refill
safe-to-test
Ready to run unit-tests in a PR.
#1722
opened Sep 17, 2026 by
dingzhiqiang
Collaborator
Loading…
6 of 9 tasks
feat(v2): support partial rollout groups
safe-to-test
Ready to run unit-tests in a PR.
#1721
opened Sep 16, 2026 by
sitabulaixizawaluduo
Collaborator
Loading…
3 of 15 tasks
build: upgrade Megatron Core to 0.19 and Bridge to 0.6
#1720
opened Sep 16, 2026 by
sitabulaixizawaluduo
Collaborator
•
Draft
feat(engine): support native MTP-only and packed Qwen SFT
safe-to-test
Ready to run unit-tests in a PR.
#1719
opened Sep 16, 2026 by
dingzhiqiang
Collaborator
•
Draft
fix(v2): propagate configured training RPC timeout to data proxy
#1712
opened Sep 15, 2026 by
sitabulaixizawaluduo
Collaborator
Loading…
4 of 9 tasks
fix(utils): zero single-element leave-one-out normalization at batch level
#1706
opened Sep 13, 2026 by
KunyangZhang
Loading…
7 of 9 tasks
feat(dpo): support SimPO reference-free preference optimization with target margin
#1699
opened Sep 10, 2026 by
hsusul
Contributor
Loading…
8 of 15 tasks
feat(reward): support composable reasoning rewards with format verification and length regulation
#1698
opened Sep 10, 2026 by
hsusul
Contributor
Loading…
6 of 15 tasks
feat: add training throughput, FLOPs and MoE balance metrics
safe-to-test
Ready to run unit-tests in a PR.
#1693
opened Sep 10, 2026 by
yulangz
Collaborator
Loading…
5 of 15 tasks
feat(openai): support zero-argument tools and XML-escaped parameter parsing in ToolCallParser
#1691
opened Sep 9, 2026 by
hsusul
Contributor
Loading…
6 of 15 tasks
feat(trainer): support score reward clipping and teacher weight normalization in MOPD loss
#1690
opened Sep 9, 2026 by
hsusul
Contributor
Loading…
7 of 9 tasks
fix(utils): treat missing packages as not matching version checks
#1689
opened Sep 9, 2026 by
hsusul
Contributor
Loading…
1 task done
feat(archon): add NPU support for varlen attention
#1686
opened Sep 8, 2026 by
262913
Loading…
5 of 15 tasks
refactor: unify AWEX converters and adapters across v1 and v2
safe-to-test
Ready to run unit-tests in a PR.
#1684
opened Sep 7, 2026 by
sitabulaixizawaluduo
Collaborator
Loading…
1 task done
feat(workflow): support turn-level rewards and raw reward preservation in MultiTurnWorkflow
#1681
opened Sep 5, 2026 by
hsusul
Contributor
Loading…
3 tasks done
feat(trainer): support configurable entropy regularization in PPO and GRPO actor
#1680
opened Sep 5, 2026 by
hsusul
Contributor
Loading…
3 tasks done
feat: support single-GPU GSM8K smoke runs
#1678
opened Sep 5, 2026 by
CharlesXu-HQ
Contributor
Loading…
5 of 15 tasks
perf(engine): encode each distinct image once per vision-tower forward
#1676
opened Sep 4, 2026 by
MrlixiangWE
Loading…
6 of 15 tasks
fix(openai): preserve multi-turn step rewards and raw baselines in group normalization
#1674
opened Sep 3, 2026 by
hsusul
Contributor
Loading…
3 tasks done
fix(trainer): prevent reward and advantage explosion on remainder group normalization
#1673
opened Sep 3, 2026 by
hsusul
Contributor
Loading…
3 tasks done
Previous Next
ProTip!
Updated in the last three days: updated:>2026-09-14.