-
Notifications
You must be signed in to change notification settings - Fork 3.3k
Pull requests: verl-project/verl
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[trainer,misc] fix: fix multiple bugs in fully async trainings
#5373
opened Feb 23, 2026 by
guillemgt
Loading…
7 of 8 tasks
[megatron] fix: patch support newer mcore version
#5372
opened Feb 23, 2026 by
HollowMan6
Loading…
8 tasks done
[env] fix: add .sh file supporting stable conda env building
#5371
opened Feb 22, 2026 by
HaCTang
Loading…
8 tasks done
[megatron] refactor: Refactor policy update failure handling from exception to warning in MegatronPPOActor to support FP16 training
#5369
opened Feb 22, 2026 by
mirrorboat
Loading…
8 tasks done
[algo] feat: Exception for agg_loss when
dp_size > 1 but global information is absent & fix: correct & consistent loss aggregation for "seq-mean-token-sum-norm"
#5366
opened Feb 22, 2026 by
tongyx361
Loading…
8 tasks done
[fsdp2] feat: add per-layer GPU optimizer step with async prefetch pipeline
#5364
opened Feb 21, 2026 by
aoshen524
Loading…
4 tasks
[fsdp] feat: checkpoint input CPU offload for gradient checkpointing
#5363
opened Feb 21, 2026 by
aoshen524
Loading…
3 tasks
[trainer] fix: Misaligned FSDP SFT loss mask
#5362
opened Feb 20, 2026 by
alek6kun
Loading…
8 tasks done
[tool] fix: add timeout for MCP tool calls
#5361
opened Feb 20, 2026 by
denismegerle
Loading…
7 of 8 tasks
[tool, rollout] fix: handle optional interaction kwargs
#5360
opened Feb 20, 2026 by
denismegerle
Loading…
7 of 8 tasks
[rollout, data] fix: honor train_max_samples/val_max_samples in fully async rollouter
#5359
opened Feb 20, 2026 by
denismegerle
Loading…
7 of 8 tasks
[trainer] feat: Support RL trainer with TorchtitanEngine
#5356
opened Feb 20, 2026 by
acisseJZhong
Loading…
6 tasks
[config] remove deprecated actor_rollout_ref.rollout.mode config field
#5352
opened Feb 19, 2026 by
dhruvnigam93
Loading…
4 of 7 tasks
build(deps): update transformers requirement from <5.0.0 to <6.0.0
dependencies
Pull requests that update a dependency file
python
Pull requests that update python code
#5347
opened Feb 18, 2026 by
dependabot
bot
Loading…
[fsdp] fix: make GDN work correctly when using ulysses and varlen sequence
#5346
opened Feb 18, 2026 by
nono-Sang
Loading…
8 tasks
[rollout] feat: enable R3 in fully async mode
#5344
opened Feb 18, 2026 by
guillemgt
Loading…
7 of 8 tasks
[vllm] fix: enable multi-node vllm replicas with expert parallelism
#5339
opened Feb 17, 2026 by
guillemgt
Loading…
6 of 8 tasks
[fsdp,algo] feat: add truncate_padding to reduce wasted compute on padded sequences
#5338
opened Feb 17, 2026 by
roeeh21
Loading…
2 tasks done
fix: preserve lora_alpha in model merger (#5325)
#5337
opened Feb 17, 2026 by
maximalmargin
Loading…
[algo] feat: add Focal-GRPO implementation
#5334
opened Feb 16, 2026 by
alekseymalakhov11
Loading…
8 tasks done
fix: handle labels=None in fused forward to avoid crash without MTP (#5273)
#5332
opened Feb 16, 2026 by
devchilll
Loading…
[tool] fix: make rate-limiter actor detached to prevent ActorDiedError
#5327
opened Feb 15, 2026 by
mirrorboat
Loading…
[ckpt, model] fix: preserve lora_alpha in model_merger via training meta
#5326
opened Feb 15, 2026 by
Yatogaii
Loading…
7 tasks done
Previous Next
ProTip!
no:milestone will show everything without a milestone.