-
Notifications
You must be signed in to change notification settings - Fork 2.9k
Pull requests: volcengine/verl
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[training_utils] fix: RM extra scaling in KL/PG losses
#4711
opened Dec 29, 2025 by
JacobHelwig
Loading…
[tool] fix: correct torch profiler logic for distributed trace collection
#4707
opened Dec 28, 2025 by
AkiRusProd
Loading…
fix: use model_dump() for proper Pydantic serialization in token2text
#4706
opened Dec 28, 2025 by
yurekami
Loading…
3 tasks done
[trainer] fix: handle empty tensors in extract_multi_modal_inputs
#4705
opened Dec 28, 2025 by
yurekami
Loading…
1 of 3 tasks
fix(model_merger): add LoRA weight merging support
#4704
opened Dec 28, 2025 by
yurekami
Loading…
1 of 3 tasks
[megatron] fix: Bump Megatron-Bridge commit for PEFT recompute
#4702
opened Dec 28, 2025 by
HollowMan6
Loading…
7 tasks done
[merger] fix: support vision_model keys in Megatron merger for VL models
#4701
opened Dec 28, 2025 by
yurekami
Loading…
1 task
feat(trainer): add logging of extra reward metrics
#4698
opened Dec 27, 2025 by
yurekami
Loading…
2 tasks
[trainer] fix: use dp_size instead of world_size in _balance_batch
#4697
opened Dec 27, 2025 by
yurekami
Loading…
2 tasks
fix(fsdp): handle param offloading in get_actor_weights_info
#4696
opened Dec 27, 2025 by
yurekami
Loading…
2 tasks
fix(vllm): correct max_tokens calculation in async server
#4694
opened Dec 27, 2025 by
yurekami
Loading…
2 tasks
fix: correct infer_tp size calculation in rollout device mesh
#4693
opened Dec 27, 2025 by
yurekami
Loading…
1 of 2 tasks
[docs] chore: improve docstring consistency in device.py
#4691
opened Dec 27, 2025 by
yurekami
Loading…
2 tasks done
Fix Profiler class using wrong config attribute names
#4689
opened Dec 27, 2025 by
yurekami
Loading…
3 tasks
feat: Add SHM cache for multi-turn multi-modal workloads
#4688
opened Dec 27, 2025 by
aoshen524
Loading…
7 tasks
[ckpt] feat: Added an option to save only the top checkpoints in FSDP SFT
#4684
opened Dec 26, 2025 by
AJain9199
Loading…
4 of 7 tasks
[fsdp, fully-async] feat: add validate process on trainer node when use_trainer_do_valida…
#4683
opened Dec 26, 2025 by
chenjiaoAngel
Loading…
7 tasks
[algo, cfg, worker, doc] feat: add SOAP optimizer support
#4681
opened Dec 26, 2025 by
CedricHwong
Loading…
7 tasks done
[rollout, hardware] feat: support MXFP8 rollout on Ascend NPU
#4679
opened Dec 26, 2025 by
SlightwindSec
•
Draft
2 of 9 tasks
perf(ppo): vectorize advantage estimators and optimize tensor ops
#4677
opened Dec 26, 2025 by
herniqeu
Loading…
2 of 7 tasks
[megatron] feat: Share actor and ref in LoRA
#4673
opened Dec 25, 2025 by
HollowMan6
•
Draft
7 tasks done
[ray,rollout,trtllm] feat: Adding tensorrt_llm as new rollout engine
#4665
opened Dec 25, 2025 by
joyang-nv
Loading…
5 tasks
[recipe, megatron, fsdp] Support fully async on NPU
#4658
opened Dec 24, 2025 by
Shangwei-Li
Loading…
6 of 7 tasks
[recipe, ckpt] feat: bypass the CPU in the checkpoint engine for fully_async mode.
#4654
opened Dec 23, 2025 by
ZSL98
Loading…
3 of 7 tasks
Previous Next
ProTip!
Adding no:label will show everything without a label.