v0.3.2: RL Baselines

What's Changed

• [misc] set dev version by @hiyouga in #372
• fix typo in tensorboard logger by @ning-mz in #373
• [utils] fix log probs by @hiyouga in #375
• Supports video input by @robinjoe93 in #376
• [data] fix video data by @hiyouga in #386
• [data] add filter worker by @hiyouga in #387
• [example] update qwen3_14b_dapo17k_dapo.sh by @Saigyouji-Yuyuko1000 in #389
• [misc] fix ci by @hiyouga in #391
• [ckpt] fix remove logic by @hiyouga in #393
• [example] update config by @hiyouga in #394
• [worker] fix typo in FSDPWorker by @jasper0314-huang in #396
• [misc] fix flops counter by @hiyouga in #401
• [readme] update wechat by @hiyouga in #404
• [readme] add our work to the readme by @JingchengYang4 in #408
• [example] change dapo verify by @Saigyouji-Yuyuko1000 in #407
• [reward] fix dapo verifier by @hiyouga in #410
• Update README.md to include Long-RL by @yukang2017 in #411
• [worker] add dynamic batching by @hiyouga in #417
• [readme] update wechat by @hiyouga in #418
• [worker] fix dp tokens by @hiyouga in #419
• [examples] fix config by @hiyouga in #420
• [misc] fix ops by @hiyouga in #421
• [worker] fix fsdp worker by @hiyouga in #422
• [worker] fix grad norm by @hiyouga in #423
• [data] better mm data collate by @hiyouga in #424
• [trainer] support auto resume by @hiyouga in #425
• [worker] add dynamic batching computational workload balance by @hiyouga in #426
• [readme] update usage of apptainer by @yzoaim in #434
• [readme] update wechat by @hiyouga in #442
• [protocol] non blocking false by default by @hiyouga in #445
• [readme] update wechat by @hiyouga in #447
• [feat] support ray.timeline by @yzoaim in #449
• [docker] upgrade vllm to 0.10 by @hiyouga in #453
• [worker] fix multi modal data oom by @hiyouga in #454
• [misc] fix data proto by @hiyouga in #458
• [readme] update wechat by @hiyouga in #461
• [trainer] fix checkpoint tracker by @hiyouga in #467
• [patch] fix fa utils by @hiyouga in #472
• [misc] fix fa patch by @hiyouga in #473
• [misc] fix model merger by @hiyouga in #479
• [misc] lint by @hiyouga in #480
• Fix valset loading for videos by @zhuohaoyu in #482
• [readme] update wechat by @hiyouga in #486
• [bugfix] fix position ids for latest transformers by @hiyouga in #494
• [readme] update wechat by @hiyouga in #495
• [misc] pin transformers to 4.56.1 by @hiyouga in #496
• [deps] upgrade transformers to 4.54 by @hiyouga in #501
• [release] v0.3.2 by @hiyouga in #502

New Contributors

• @ning-mz made their first contribution in #373
• @robinjoe93 made their first contribution in #376
• @jasper0314-huang made their first contribution in #396
• @JingchengYang4 made their first contribution in #408
• @yukang2017 made their first contribution in #411
• @yzoaim made their first contribution in #434
• @zhuohaoyu made their first contribution in #482

Full Changelog: v0.3.1...v0.3.2

v0.3.1: Multi-modal DAPO

What's Changed

• [example] fix runtime env by @hiyouga in #224
• Update Awesome Work using EasyR1 by @RainBowLuoCS in #240
• Update Awesome Work using EasyR1 by @xyliugo in #239
• [trainer] support async reward by @hiyouga in #252
• [readme] add baselines by @hiyouga in #253
• [script] fix merge script by @hiyouga in #254
• [misc] update baselines & docker image by @hiyouga in #256
• [readme] update baseline by @hiyouga in #258
• [data] support custom chat template by @hiyouga in #270
• [reward] support batch reward by @hiyouga in #271
• [example] change env vars by @hiyouga in #272
• [Readme] Add awesome work using EasyR1 by @Wangbiao2 in #273
• [model] add qwen3 support by @hiyouga in #276
• [example] update script by @hiyouga in #277
• [readme] update wechat by @hiyouga in #280
• [misc] fix logger by @hiyouga in #288
• [readme] update wechat by @hiyouga in #292
• add get model from modelscope by @Saigyouji-Yuyuko1000 in #297
• [readme] update wechat by @hiyouga in #301
• Add a new work based on EasyR1 by @LiuRicky in #303
• add new work based on EasyR1 by @waltonfuture in #313
• [logger] fix tensorboard by @hiyouga in #316
• Add a new work based on EasyR1 by @Gabesarch in #325
• [misc] fix console hanging by @hiyouga in #293
• [misc] several update by @hiyouga in #329
• Update README.md by @CSfufu in #330
• [perf] pass raw image data between workers by @tongxiao2002 in #318
• [readme] add our work using EasyR1 by @kxfan2002 in #331
• add our work using EasyR1 by @YutingLi0606 in #337
• [data] fix position ids for qwen2vl mrope & add test by @hiyouga in #339
• [worker] colocate actor and ref model by @hiyouga in #342
• [trainer] save best checkpoint by @hiyouga in #343
• [trainer] fix bug by @hiyouga in #344
• [utils] update data protocol by @hiyouga in #345
• [trainer] repeat rollout and prepare filter by @hiyouga in #346
• [worker] expose rollout manager by @hiyouga in #347
• [worker] fix vllm sharding manager by @hiyouga in #348
• fix: bug by @gdw439 in #350
• [trainer] fix progress bar by @hiyouga in #355
• [readme] update docker image by @hiyouga in #357
• [trainer] add online filtering by @Saigyouji-Yuyuko1000 in #358
• [worker] update reward manager by @hiyouga in #360
• Fix/vllm processor cache for text only model by @cyc00518 in #359
• [breaking] support text-image mixed data by @hiyouga in #361
• [model] fix qwen2vl bug by @hiyouga in #363
• [tracking] add tensorboard exp name by @hiyouga in #365
• [worker] do not load ref if kl is disabled by @hiyouga in #366
• [worker] fix skip ref model by @hiyouga in #367
• [examples] add qwen3_14b_dapo17k_dapo by @Saigyouji-Yuyuko1000 in #369
• [release] 0.3.1 by @hiyouga in #370

New Contributors

• @RainBowLuoCS made their first contribution in #240
• @xyliugo made their first contribution in #239
• @Saigyouji-Yuyuko1000 made their first contribution in #297
• @waltonfuture made their first contribution in #313
• @Gabesarch made their first contribution in #325
• @CSfufu made their first contribution in #330
• @tongxiao2002 made their first contribution in #318
• @kxfan2002 made their first contribution in #331
• @YutingLi0606 made their first contribution in #337
• @gdw439 made their first contribution in #350
• @cyc00518 made their first contribution in #359

Full Changelog: v0.3.0...v0.3.1

v0.3.0: Initial release

What's Changed

• update readme by @hiyouga in #4
• [readme] update readme by @hiyouga in #5
• [worker] fix small models by @hiyouga in #14
• feat: swanlab examples by @Zeyi-Lin in #13
• [example] add ReMax support by @Shenzhi-Wang in #20
• fix:vllm length by @AL-377 in #18
• fix: math reward fn by @yueyang130 in #26
• [readme] update readme by @hiyouga in #29
• Fix template issue by @wzq016 in #31
• [example] fix length by @hiyouga in #32
• [readme] update hardware requirement by @hiyouga in #33
• [worker] fix model attn init by @hiyouga in #37
• Witness the Aha Moment on Counting Task by @BUAADreamer in #38
• [example] fix clevr example by @hiyouga in #47
• Fix: save processor for VLMs by @wzq016 in #48
• [perf] support padding-free training for VLMs by @hiyouga in #61
• [readme] update readme by @hiyouga in #62
• [readme] add fig explain by @hiyouga in #64
• [readme] update fig by @hiyouga in #65
• [trainer] support resume ckpt by @hiyouga in #66
• [config] update default config by @hiyouga in #68
• [readme] update wechat by @hiyouga in #71
• [env] fix memory leak & enable vLLM v1 by @hiyouga in #73
• [readme] update readme by @hiyouga in #75
• [readme] update readme by @hiyouga in #80
• Add new baseline GeoQA8k from R1V by @chenllliang in #86
• [feat] support freeze vision tower by @hiyouga in #99
• [config] increase prompt length by @hiyouga in #100
• update readme - add ## Awesome Work using EasyR1 by @LengSicong in #101
• Add the work Vision-R1 that uses EasyR1 by @Osilly in #102
• fix:OOM by @dirtyDan0 in #111
• [trainer] verify arg by @hiyouga in #112
• [misc] sync feat from upstream by @hiyouga in #113
• [misc] clean some code by @hiyouga in #114
• [example] add examples by @hiyouga in #118
• [checkpoint] fix load checkpoint by @hiyouga in #119
• [trainer] gather metrics by @hiyouga in #120
• [misc] add doc string by @hiyouga in #121
• Add seg zero to README by @LiuRicky in #122
• Update README.md by @PzySeere in #124
• fix readme by @hiyouga in #127
• [core] remove entropy loss by @hiyouga in #132
• [trainer] support val sampling by @hiyouga in #133
• misc: save at the last step by @dirtyDan0 in #138
• feat: swanlab add easyr1 and verl config by @Zeyi-Lin in #140
• [version] upgrade vllm to 0.8 by @hiyouga in #143
• [readme] update docker file by @hiyouga in #146
• [readme] update wechat by @hiyouga in #147
• [readme] update dockerfile by @hiyouga in #148
• Update requirements.txt for multinode by @chenllliang in #154
• [trainer] support channel-wise reward by @hiyouga in #155
• Update README.md by @PzySeere in #157
• [trainer] support save limit & fix oom issue by @hiyouga in #158
• [misc] update docker files by @hiyouga in #162
• [trainer] support 32b by @hiyouga in #164
• [data] use hf-native template by @hiyouga in #165
• [misc] fix dataset by @hiyouga in #166
• [readme] update tutorial by @hiyouga in #167
• [tracking] add tensorboard by @hiyouga in #170
• [misc] support adamw bf16 by @hiyouga in #171
• [misc] fix config by @hiyouga in #172
• [misc] fix metrics by @hiyouga in #173
• [misc] refactor val gen log by @hiyouga in #174
• update Awesome Work using EasyR1 by @appletea233 in #179
• [misc] fix masked mean by @hiyouga in #181
• [misc] algo improvement by @hiyouga in #184
• [misc] minor update by @hiyouga in #188
• [fix] arg check by @hiyouga in #189
• [bugfix] fix vllm 0.8.3 rollout by @hiyouga in #197
• [deps] upgrade to vllm 0.8.3 by @hiyouga in #202
• [core] separate score fn & vllm logit bias by @hiyouga in #204
• Supports loading format prompt from a file by @Wangbiao2 in #208
• [data] update data configs by @hiyouga in #214
• fix: enable user to filter overlong examples in RLHFDataset by @0x404 in #210
• [data] fix rl dataset by @hiyouga in #215
• [misc] lint by @hiyouga in #216
• [data] add multi image dataset by @hiyouga in #217
• [readme] add multi node script by @hiyouga in #218
• [torch] fix saving bf16 optimizer by @hiyouga in #221
• [version] release 0.3.0 by @hiyouga in #222

New Contributors

• @Zeyi-Lin made their first contribution in #13
• @AL-377 made their first contribution in #18
• @yueyang130 made their first contribution in #26
• @wzq016 made their first contribution in #31
• @BUAADreamer made their first contribution in #38
• @chenllliang made their first contribution in #86
• @LengSicong made their first contribution in #101
• @Osilly made their first contribution in #102
• @dirtyDan0 made their first contribution in #111
• @LiuRicky made their first contribution in #122
• @PzySeere made their first contribution in #124
• @appletea233 made their first contribution in #179
• @Wangbiao2 made their first contribution in #208
• @0x404 made their first contribution in #210

Full Changelog: https://github.com/hiyouga/EasyR1/commits/v0.3.0