COMMITS
/ train.py June 8, 2026
Z
Only upload per sample stats to wandb (#2027)
Zilin Zhu committed
May 6, 2026
L
fix ppo value offload bugs (#1882)
LiLei committed
April 24, 2026
L
refactor/ppo (#1856)
LiLei committed
March 26, 2026
Z
Fix uploading sglang metrics to wandb (#1768)
Zilin Zhu committed
March 9, 2026
Z
[docker] cherry pick qwen3.5 bugfix (#1691)
Zilin Zhu committed
March 8, 2026
S
Z
support offloading non-updatable server (#1668)
Zilin Zhu committed
February 26, 2026
C
Fix #1595: pass rollout_id explicitly to offload_train (#1631)
Chengxing Xie committed
January 16, 2026
Z
[cleanup] clean up utils folder (#1437)
Zilin Zhu committed
January 6, 2026
Z
Add clear_num_new_engines and some code cleanup (#1349)
Zilin Zhu committed
January 5, 2026
Z
[refactor] minor code refactor for save_hf (#1334)
Zilin Zhu committed
C
[Megatron Bridge] Support save hf format model (#1289)
Chenhe committed
December 23, 2025
Z
sync internal features (#1192)
Zilin Zhu committed
December 18, 2025
Z
Support async save and add extra save at the end of the training (#1143)
Zilin Zhu committed
December 5, 2025
S
November 30, 2025
F
Support zero host or device memory waste for weight update (#973)
fzyzcjy committed
November 26, 2025
F
Support checking SGLang weight update correctness (#927)
fzyzcjy committed
November 17, 2025
F
Tiny unify wandb and tensorboard code (#801)
fzyzcjy committed
F
Super tiny configure logger (#797)
fzyzcjy committed
November 16, 2025
F
Support saving memory by disabling tensor backuper (#776)
fzyzcjy committed
C
[Feature] Tiny fix for wandb run id (#730)
Chengxing Xie committed
November 15, 2025
F
Tiny refactor train script (#765)
fzyzcjy committed
October 24, 2025
F
Tiny call clear cache when offloading rollout but not train (#574)
fzyzcjy committed
F
Split args.offload to train and rollout (#569)
fzyzcjy committed
F
Try to fix train.py wrong logic about saving checkpoints (#564)
fzyzcjy committed
F
Support evaluation-only runs (#527)
fzyzcjy committed
October 23, 2025
October 21, 2025
R
[Feat] Support offload cuda graph (#354)
ryang committed
October 16, 2025
Z
fix rollout dataset loading (#504)
Zilin Zhu committed
October 14, 2025
Z
use SingletonMeta for _TensorboardAdapter (#494)
Zilin Zhu committed
October 13, 2025
Z
[bugfix] initialize rollout manager first to calculate num_rollout (#473)
Zilin Zhu committed
October 12, 2025
N
Fix typo and optimize TensorBoard logging (#464)
none0663 committed
October 10, 2025
F
Allow checking accuracy correctness programmatically (#453)
fzyzcjy committed
October 5, 2025
N
[test] add tensorboard (#420)
none0663 committed
September 30, 2025
Z
[feat] support fault tolerant for rollout engines (#405)
Zilin Zhu committed
September 28, 2025
Z
[refactor] remove Registry and change the order of init (#398)
Zilin Zhu committed
September 19, 2025
Z
[refactor] Add actor registry (#359)
Zilin Zhu committed
Z
[feat] add --critic-lr and --num-critic-only-steps (#350)
Zilin Zhu committed
September 18, 2025
L
feature: ppo (#347)
LiLei committed
September 17, 2025
Z
[feat] init support for PPO (#342)
Zilin Zhu committed
September 16, 2025
H
[Refactor] Merge rollout controller into rollout manager (#304)
Huapeng Zhou committed
September 5, 2025
Z
[refactor] Add isort back and move global gloo to global util (#273)
Zilin Zhu committed
August 14, 2025
August 11, 2025
Z
[refactor] separate generate and eval in Buffer to make the code cleaner (#161)
Zilin Zhu committed
Z
[refactor] move log_eval_data to Buffer (#159)
Zilin Zhu committed
August 9, 2025
S
Support Multi-Stage Awake (#149)
Stefan He committed
Z
Revert "[update weight] resume sglang in multi-stage (#150)" (#151)
Zilin Zhu committed
Z
[update weight] resume sglang in multi-stage (#150)
Zilin Zhu committed
July 31, 2025
C
Fix some tiny bugs and add assert (#122)
Chengxing Xie committed