-
Notifications
You must be signed in to change notification settings - Fork 4.1k
Pull requests: verl-project/verl
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[model] feat: add qwen3-122b long seq launch script for ascend
#6808
opened Jun 22, 2026 by
zjchenn
Contributor
Loading…
3 of 8 tasks
[model] feat: add qwen3-122b long seq launch script for ascend
#6807
opened Jun 22, 2026 by
zjchenn
Contributor
Loading…
3 of 8 tasks
[rollout, algo] fix: compute rollout_is_seq_fraction from raw weights
#6806
opened Jun 21, 2026 by
EazyReal
Loading…
3 tasks done
[algo] fix: remove dead code in GPG advantage estimator
#6803
opened Jun 21, 2026 by
EazyReal
Loading…
[rollout, algo] feat: add binary_kl (KPop) bidirectional KL rejection sampling
#6800
opened Jun 21, 2026 by
yan-sun-x
Loading…
5 of 8 tasks
[fully_async] fix: introduce accumulated_idle_time to record the actual rollouter idle time
#6798
opened Jun 19, 2026 by
mikequan0425
Contributor
Loading…
2 of 8 tasks
[fully_async, trainer] fix: align aggregated metrics logging with current step
#6796
opened Jun 18, 2026 by
huaiyizhao
Contributor
Loading…
8 tasks
[fully_async] fix: remove invalid rollout.single_turn_response_length override
#6795
opened Jun 18, 2026 by
Vivicai1005
Contributor
Loading…
4 of 8 tasks
[rollout][sglang] feat: delta weight sync (sparse trainer->rollout updates)
#6794
opened Jun 18, 2026 by
ChangyiYang
Contributor
•
Draft
[reward, data, rollout, worker] feat: add Open-R1 multimodal and TinyLLaVA-Video-R1 dataset support
#6793
opened Jun 18, 2026 by
lihanwen7
Loading…
3 of 4 tasks
[WIP][ci] chore: change two-node Ascend RayJob E2E workflow from A2 to A3
Ascend
#6787
opened Jun 17, 2026 by
wangdongleix
Contributor
Loading…
8 tasks
[rollout] feat: add Continuous Token for Agentic Rollout
#6779
opened Jun 16, 2026 by
gxlvera
Loading…
6 tasks
[hardware] refactor: per-model NPU patches with fault isolation
Ascend
#6777
opened Jun 16, 2026 by
tardis-key
Collaborator
Loading…
2 of 8 tasks
[fsdp, veomni] fix: wire fused top-k distillation outputs
#6737
opened Jun 15, 2026 by
zhangxin81
Loading…
4 of 8 tasks
[training_utils] fix: cap micro-batch tokens at max_token_len
#6735
opened Jun 15, 2026 by
Li-bf
Loading…
5 of 8 tasks
[vllm, rollout] fix: support vLLM pipeline parallel on NPU via engine_kwargs
Ascend
#6732
opened Jun 15, 2026 by
chengminhua
Contributor
Loading…
8 tasks
[algo] feat: add CPPO (position-weighted cumulative-prefix-divergence token mask)
#6731
opened Jun 15, 2026 by
chongqichuizi875
Loading…
5 of 8 tasks
fix(workers): prepare actor weights before rollout wakeup
Ascend
#6729
opened Jun 14, 2026 by
gaohongkui
Loading…
[Draft]Support Megatron LoRA adapter export for rollout
#6713
opened Jun 12, 2026 by
hbhflw2000
•
Draft
[rollout] feat: extract load balancing into pluggable router module
#6712
opened Jun 12, 2026 by
ZOULQ
Loading…
8 tasks
[ci] chore: add three baselines for npu nightly ci
Ascend
#6711
opened Jun 12, 2026 by
daikang6
Contributor
Loading…
8 tasks
[mcore] fix: OOB IndexError in preprocess_thd_engine when FP8 …
#6703
opened Jun 12, 2026 by
ewan0x79
Loading…
5 of 8 tasks
[fully_async, trainer] fix: sync optimizer total steps before trainer initialization
#6684
opened Jun 10, 2026 by
mikequan0425
Contributor
Loading…
2 of 8 tasks
Previous Next
ProTip!
Adding no:label will show everything without a label.