-
Notifications
You must be signed in to change notification settings - Fork 4k
Pull requests: verl-project/verl
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[ci] migrate pip→uv in all Ascend CI workflows
Ascend
#6535
opened May 29, 2026 by
KadenZhang3321
Loading…
3 tasks
[rollout] feat: load-balance away from sticky sessions under imbalance
#6533
opened May 29, 2026 by
aoshen02
Contributor
Loading…
2 tasks done
[misc] fix: harden chat template prompt inference
#6529
opened May 29, 2026 by
anzhsoft
Contributor
Loading…
4 of 8 tasks
[megatron] feat: align optimizer states and DDP grad bucket with model precision
#6526
opened May 28, 2026 by
kolehma8
Loading…
2 of 7 tasks
[fsdp, model] feat: support glm_moe_dsa FSDP training with DSA attention
#6525
opened May 28, 2026 by
Kite0011
Contributor
Loading…
8 tasks
[vllm] fix: reset all caches after weight updates
#6522
opened May 28, 2026 by
s-isaev
Loading…
7 of 8 tasks
[ci] chore: npu ci use cann9.0.0
Ascend
#6520
opened May 28, 2026 by
daikang6
Contributor
Loading…
8 tasks
[fsdp, algo] no grad for entropy and kl if the loss coef is 0
#6519
opened May 28, 2026 by
huaiyizhao
Contributor
Loading…
8 tasks
[megatron] fix: zero out mtp_num_layers and trim csa_compress_ratios on vanilla_mbridge=True path
#6515
opened May 28, 2026 by
Meirtz
Loading…
[fsdp, model] feat: per-unit LoRA summon, FSDP1/2 compatibility, and strip-modules support
#6512
opened May 27, 2026 by
qinganrice
Loading…
fix: restore rollout state after checkpoint update failures
#6510
opened May 27, 2026 by
athreesh
Contributor
Loading…
[reward, worker] fix: guard reward_tensor index -1 when valid_response_length == 0
#6508
opened May 27, 2026 by
donghyq
Loading…
7 of 8 tasks
[veomni, cfg] feat: add missing config fields to veomni.yaml
#6505
opened May 27, 2026 by
mikequan0425
Contributor
Loading…
2 of 8 tasks
[algo] fix: normalize GDPO advantages over responses
#6497
opened May 27, 2026 by
lucky9-cyou
Loading…
4 of 8 tasks
[worker, model] fix: respect attn_implementation override in load_valuehead_model
#6495
opened May 27, 2026 by
harryge00
Loading…
[reward_manager] fix: guard against empty responses and None overlong_buffer_cfg
#6484
opened May 26, 2026 by
imitater-dou
Loading…
4 tasks done
[agent_loop, tool] fix: support hermes-format tool calls on gpt-oss tokenizer models
#6481
opened May 26, 2026 by
dafu-wu
Contributor
Loading…
2 of 6 tasks
Zmj/add qwen3.5 npu longcontext
Ascend
#6474
opened May 26, 2026 by
Mengyuyang
Contributor
Loading…
8 tasks
[megatron] feat: support DeepSeek V4 GRPO
#6473
opened May 26, 2026 by
HollowMan6
Collaborator
Loading…
8 tasks done
[megatron] fix: clamp num_tokens=0 in MTP loss & add normalized scale for MTP per token loss
#6464
opened May 25, 2026 by
arvyanh
Contributor
Loading…
8 tasks done
[rollout, vllm] fix: avoid SIGSEGV on ROCm TP=1 by conditionally omitting distributed_executor_backend
#6459
opened May 25, 2026 by
HeShiLie
Loading…
4 of 5 tasks
[algo] fix rollout rejected rows in group advantages
#6452
opened May 23, 2026 by
haoyang9804
Contributor
Loading…
Previous Next
ProTip!
Follow long discussions with comments:>50.