Morph MORPH
SIGN IN SIGN UP
verl-project / verl UNCLAIMED

verl: Volcano Engine Reinforcement Learning for LLMs

0 0 220 Python

COMMITS

main
April 20, 2026
P
[rollout] chore: single turn agent loop also enable rollout trace as tool loop (#6048)
Peng Wu committed
ca69921
L
[veomni] feat: support Qwen3.5 SP and add GRPO trainer demo using VeOmniEngine (#6061)
Lu Di committed
a6b352f
April 19, 2026
G
[misc] fix: project name refactor - volcengine -> verl (#6053)
Gao Ziyuan committed
58a361a
April 17, 2026
V
[fully_async] fix: preserve per-iteration routed_experts on partial rollout resume (#6046)
Vadim Vorobev committed
bcb6386
J
[rollout] fix: RM sleep/wake teacher replicas (#6041)
Jacob Helwig committed
f512e11
J
[trainer, rollout, algo] refactor: Remove OPD colocate mode (#6039)
Jacob Helwig committed
0c0c412
C
[5/n][trainer] feat: flowgrpo trainer (#5951)
Cheung Ka Wai committed
4b5a9a9
G
[trainer] fix: add missing rollout dump and corrected validation logging in main_ppo_sync (#6024)
guillemgt committed
dea887c
H
[trainer,algo] feat: Support On-Policy Distillation in `main_ppo_sync` (#5997)
Han Zhenyu 韩振宇 committed
0588bb6
鐘
[veomni] fix: use local paths for VeOmni model loading (#6034)
鐘天楽 committed
267cce6
V
[fully_async] fix: replace routed_experts on partial rollout resume i… (#6029)
Vadim Vorobev committed
d079d72
April 16, 2026
Z
[data, trainer] fix: batch padding for multi-trajectory (#5969)
ZhentaoFan committed
170a593
J
[misc] chore: remove deprecated requirements.txt (#6032)
Joel committed
5f1e86d
R
[ci] chore: qwen3.5 add docker file add x86 CANN8.5.2 (#6031)
ruanhao566 committed
8504a87
鐘
[veomni] feat: add DeepSeek-V3 to MOE_PARAM_HANDERS (#5996)
鐘天楽 committed
d4b7462
W
[fully_async] fix: add fully async grpo qwen3-235b npu script in main branch (#6012)
wangshuyang31 committed
369c544
Y
[ci] fix: update docker-build-ascend-a3-qwen3_5 (#6022)
yyyy2000 committed
d8dc429
R
[fsdp] feat: qwen3.5 modify npu docker file based on CANN 8.5.2 (#6017)
ruanhao566 committed
e06f2fa
S
[tool, rollout, cfg] feat: per-sample tool environment routing for ToolAgentLoop (#5978)
Sngwon committed
37fe672
B
[data] fix: pad data in preprocess_packed_seqs if shorter than align_size (#6001)
Beirong Zhou committed
45c0f58
X
[ci] chore: add sglang new version docker for NPU (#6021)
xiazhahe committed
910ba34
J
[megatron, trainer] fix: respect calculate_entropy config in megatron actor update (#6016)
Jiarui Yao committed
45bfd3c
April 15, 2026
E
[megatron] fix: add missing FP8 padding for router replay (#5989)
eternally-z committed
9bda8b9
L
[veomni] feat: bump veomni to v0.1.8 (#5900)
Lu Di committed
59f53cc
ℍ
[megatron] fix: update patch for MLA flashattn forward (#6005)
ℍ𝕠𝕝𝕝𝕠𝕨 𝕄𝕒𝕟 committed
b9d71f9
S
[misc, fully_async] feat: add Qwen3-VL-8B fully async GRPO training script on geo3k (#6006)
Silas committed
014fd56
April 14, 2026
X
[doc] feat: move msprobe to ascend_tutorial (#6004)
Xiaobo Hu committed
809f2d8
T
[perf] feat: simplify precision_debugger config behavior and docs (#5986)
Tjh-UKN committed
f8b0dd2
R
[fsdp] feat: qwen3.5 add npu docker file (#5991)
ruanhao566 committed
9b54564
S
[rollout, vllm] fix: auto-convert disable_mm_preprocessor_cache to mm_processor_cache_gb for vllm >= 0.13.0 (#5961)
Silas committed
1581220