| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| README.md | 2026-08-25 | 8.9 kB | |
| v2.1.0 source code.tar.gz | 2026-08-25 | 12.7 MB | |
| v2.1.0 source code.zip | 2026-08-25 | 13.4 MB | |
| Totals: 3 Items | 26.1 MB | 0 | |
What's Changed
- chore(swe): remove unused SWE SFT dataset loader by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1467
- fix(docker): rename zeroclaw installer flag to --skip-quickstart by @sitabulaixizawaluduo in https://github.com/areal-project/AReaL/pull/1469
- fix(megatron): use dp_reshardable optimizer sharding and load optimizer state correctly by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1468
- fix(hermes): preserve singleton online reward signal by @jszzr in https://github.com/areal-project/AReaL/pull/1474
- fix(mcore): TP-shard GroupRMSNorm gate-norm weight for DCP checkpointing by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1470
- fix(rollout): stop controller-managed workers from dp-scaling staleness capacity by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1471
- fix(v2): export local workflow statistics by @jszzr in https://github.com/areal-project/AReaL/pull/1478
- docs: update NPU installation guide for ascend-v1.0.4 by @Adiactive in https://github.com/areal-project/AReaL/pull/1502
- Docs: Update README and cli docs by @sitabulaixizawaluduo in https://github.com/areal-project/AReaL/pull/1521
- fix(engine): route text-only VLM microbatches through the padded forward by @Adiactive in https://github.com/areal-project/AReaL/pull/1529
- fix: align CP metrics and TP grad norm metadata by @yulangz in https://github.com/areal-project/AReaL/pull/1497
- feat(infra): add HTTP-based Ray Scheduler by @HwVanICI in https://github.com/areal-project/AReaL/pull/1441
- fix(data_proxy): require admin key for /register_model to prevent SSRF by @sebastionoss in https://github.com/areal-project/AReaL/pull/1550
- test(inference-service): send admin key in data proxy register_model tests by @sitabulaixizawaluduo in https://github.com/areal-project/AReaL/pull/1556
- fix(inference): reject incomplete sampling evidence by @EazyReal in https://github.com/areal-project/AReaL/pull/1554
- fix(proxy): drop retry-orphan completions to prevent trajectory split by @yulangz in https://github.com/areal-project/AReaL/pull/1498
- fix(tests): assert all_gather_param passthrough by storage identity by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1562
- feat(rollout): add grouped reward normalization controls by @dingzhiqiang in https://github.com/areal-project/AReaL/pull/1516
- feat(vlm): add Qwen3.6 LoRA GRPO training support for 27B and 35B-A3B by @Lei00764 in https://github.com/areal-project/AReaL/pull/1444
- fix(engine): gate per-PP weight sync on SGLang backend by @koladefaj in https://github.com/areal-project/AReaL/pull/1564
- docs: allow ray scheduler in agent workflow compatibility notes by @wuallen57730 in https://github.com/areal-project/AReaL/pull/1567
- fix(rollout): attribute output tokens to the serving version by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1569
- fix(engine): engage Megatron deterministic mode before model build by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1544
- fix: emit PEFT-standard disk LoRA adapter keys so vLLM can load them (incl. [#1577]) by @sitabulaixizawaluduo in https://github.com/areal-project/AReaL/pull/1579
- feat(examples): add SGLang patch enabling Bailing hybrid radix cache by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1573
- fix(openai): align proxy tool schemas with sglang chat-completions rendering by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1499
- feat(mcore): apply fp32 lm head forward when enabled by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1574
- feat(ppo): report rejection-aware token and log-prob statistics by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1572
- fix(infra): fail fast when a local inference server dies during launch by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1545
- perf: reduce Megatron training memory peaks by @yulangz in https://github.com/areal-project/AReaL/pull/1555
- feat(colocate): support AWEX colocated actor-rollout training by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1500
- fix(scheduler): treat every terminal Slurm state as a dead worker by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1583
- feat: update awex to 0.8.0 by @sitabulaixizawaluduo in https://github.com/areal-project/AReaL/pull/1591
- fix: isolate AWEX control barriers from payload transfer by @pyq623 in https://github.com/areal-project/AReaL/pull/1596
- fix(infra): stabilize inference worker routing and caches by @pyq623 in https://github.com/areal-project/AReaL/pull/1589
- feat(scheduler): honor reservation, exclusive and user env overrides on Slurm by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1584
- feat(engine): support fixed warmup steps by @daihaowz in https://github.com/areal-project/AReaL/pull/1597
- chore: change codeowners by @sitabulaixizawaluduo in https://github.com/areal-project/AReaL/pull/1599
- docs(algorithms): describe metadata-driven group normalization by @EazyReal in https://github.com/areal-project/AReaL/pull/1602
- fix(models): support FP32 operands with chunked LM head by @yulangz in https://github.com/areal-project/AReaL/pull/1594
- feat(scheduler): support grouped colocation in the Ray scheduler by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1575
- fix(models): configure bridge provider for deterministic builds by @sitabulaixizawaluduo in https://github.com/areal-project/AReaL/pull/1603
- fix(infra): preserve LD_PRELOAD in local launchers by @YAO-001 in https://github.com/areal-project/AReaL/pull/1578
- docs(api): correct obsolete no_save_optim guidance by @koladefaj in https://github.com/areal-project/AReaL/pull/1609
- feat(engine): enable model-owned THD for Qwen3-VL by @Adiactive in https://github.com/areal-project/AReaL/pull/1608
- feat(engine): support Qwen3-VL with native AWEX colocate by @sitabulaixizawaluduo in https://github.com/areal-project/AReaL/pull/1605
- docs: update NPU installation guide for ascend-v1.0.5 by @Adiactive in https://github.com/areal-project/AReaL/pull/1617
- fix(infra): pause proxy workers during weight updates by @Adiactive in https://github.com/areal-project/AReaL/pull/1618
- feat(trainer): add flexible GAE lambda strategies by @RanranranQAQ in https://github.com/areal-project/AReaL/pull/1600
- feat: support qwen35 awex colocate weight update by @sitabulaixizawaluduo in https://github.com/areal-project/AReaL/pull/1620
- chore(deps): bump awex to 0.8.1 by @sitabulaixizawaluduo in https://github.com/areal-project/AReaL/pull/1621
- fix(lora): regression in megatron+lora support due to missing vllm lora aliases by @gursimar in https://github.com/areal-project/AReaL/pull/1601
- fix(megatron): release actor weights after async save by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1615
- fix: make rollout sampling deterministic by @pyq623 in https://github.com/areal-project/AReaL/pull/1625
- feat(awex): add separation AdamW delta weight transfer by @pyq623 in https://github.com/areal-project/AReaL/pull/1623
- docs: clarify terminology in the M2PO guides by @Wrestle7 in https://github.com/areal-project/AReaL/pull/1635
- fix(v2): forward retry-orphan cleanup by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1627
- fix(v2): balance offline session routing by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1624
- feat(v2): support SWE proxy compatibility by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1628
- feat(v2): support grouped reward normalization by @Le8r0nJames in https://github.com/areal-project/AReaL/pull/1619
- fix(examples): retune GSM8K GRPO learning rate for FP32 master weights by @harooos in https://github.com/areal-project/AReaL/pull/1634
- perf(megatron): avoid synthetic BSHD padding rows by @EazyReal in https://github.com/areal-project/AReaL/pull/1632
- fix(api): return float zero after async reward timeout by @harooos in https://github.com/areal-project/AReaL/pull/1541
- chore: bump v2.1.0 by @sitabulaixizawaluduo in https://github.com/areal-project/AReaL/pull/1639
New Contributors
- @jszzr made their first contribution in https://github.com/areal-project/AReaL/pull/1474
- @Lei00764 made their first contribution in https://github.com/areal-project/AReaL/pull/1444
- @wuallen57730 made their first contribution in https://github.com/areal-project/AReaL/pull/1567
- @YAO-001 made their first contribution in https://github.com/areal-project/AReaL/pull/1578
- @RanranranQAQ made their first contribution in https://github.com/areal-project/AReaL/pull/1600
- @Wrestle7 made their first contribution in https://github.com/areal-project/AReaL/pull/1635
- @harooos made their first contribution in https://github.com/areal-project/AReaL/pull/1634
Full Changelog: https://github.com/areal-project/AReaL/compare/v2.0.0...v2.1.0