Skip to content

Pull requests: areal-project/AReaL

New pull request New
99 Open 1,164 Closed
99 Open 1,164 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(openai): preserve multi-turn step rewards and raw baselines in group normalization
#1674 opened Sep 3, 2026 by hsusul Loading…
3 tasks done
fix(trainer): prevent reward and advantage explosion on remainder group normalization
#1673 opened Sep 3, 2026 by hsusul Loading…
3 tasks done
feat(engine): support VLM context parallelism
#1672 opened Sep 3, 2026 by sitabulaixizawaluduo Collaborator Loading…
perf: reduce duplicated VLM rollout processing and transfers safe-to-test Ready to run unit-tests in a PR.
#1671 opened Sep 3, 2026 by sitabulaixizawaluduo Collaborator Loading…
2 of 15 tasks
1
fix(megatron): preserve gradients across MTP depths
#1670 opened Sep 3, 2026 by kkli08 Loading…
5 of 15 tasks
2
feat(trainer): support dense step-level rewards and turn-aware advantage estimation in PPO
#1667 opened Sep 2, 2026 by hsusul Loading…
3 tasks done
1
feat(openai): tree-structured reward discounting and multi-turn turn id resolution
#1666 opened Sep 2, 2026 by hsusul Loading…
3 tasks done
Add Trustabl Agent Scanner to CI
#1665 opened Sep 2, 2026 by trustabl-kathrina Loading…
ci: resolve NPU test models across runner caches
#1664 opened Sep 2, 2026 by Adiactive Contributor Loading…
3 of 15 tasks
feat(openai): avoid re-rendering full chat history on multi-turn rollout (#1658)
#1663 opened Sep 2, 2026 by hsusul Loading…
3 tasks done
1
3
fix(v2): preserve per-interaction agent rewards
#1662 opened Sep 2, 2026 by hsusul Loading…
6 of 15 tasks
1
feat: add MTP training with speculative decoding rollout safe-to-test Ready to run unit-tests in a PR.
#1659 opened Sep 1, 2026 by HT-Yuan Collaborator Loading…
5 of 15 tasks
13
fix: make V2 recovery reconnect-safe safe-to-test Ready to run unit-tests in a PR.
#1656 opened Sep 1, 2026 by pyq623 Collaborator Loading…
1
feat(megatron): add MTP auxiliary training
#1653 opened Aug 31, 2026 by kkli08 • Draft
6 of 15 tasks
6
docs: add OrcaRouter as an OpenAI-compatible provider example
#1648 opened Aug 28, 2026 by putraperdana1207-pixel Loading…
feat(trainer): report PPO phase peak memory
#1630 opened Aug 22, 2026 by EazyReal Contributor • Draft
8 of 15 tasks
fix: harden workflow dispatcher task lifecycle
#1626 opened Aug 21, 2026 by pyq623 Collaborator Loading…
1
fix(recover): publish immutable checkpoints via latest pointer safe-to-test Ready to run unit-tests in a PR.
#1616 opened Aug 19, 2026 by Le8r0nJames Collaborator Loading…
4 of 15 tasks
1
feat(v2): support AWEX colocated training
#1614 opened Aug 18, 2026 by Le8r0nJames Collaborator Loading…
7 of 15 tasks
[WIP] feat(engine): add CPU-staged Megatron optimizers safe-to-test Ready to run unit-tests in a PR.
#1613 opened Aug 18, 2026 by yulangz Collaborator Loading…
5 of 15 tasks
10
feat: support RAO grouped agent rollouts stale
#1610 opened Aug 17, 2026 by MaxLEAF3824 Loading…
1
fix(v2): build complete MoE inference metadata for weight transfer stale
#1595 opened Aug 11, 2026 by Le8r0nJames Collaborator Loading…
3 of 15 tasks
6
feat(trainer): add multi-objective policy distillation safe-to-test Ready to run unit-tests in a PR.
#1592 opened Aug 11, 2026 by yulangz Collaborator Loading…
8 of 14 tasks
73
feat: add v2 DTE colocated weight transfer stale
#1587 opened Aug 7, 2026 by dingzhiqiang Collaborator Loading…
2
fix(infra): reap orphaned workers and survive partial teardown stale
#1585 opened Aug 6, 2026 by Le8r0nJames Collaborator Loading…
7 of 8 tasks
7
ProTip! Add no:assignee to see everything that’s not assigned.