v0.2.2

v0.2.2 is here! Thanks to everyone who contributed to this release.

Major Updates

In addition to multiple memory and performance improvements, v0.2.2 adds support for:

• Int4-QAT training
• Full R3 (Rollout Routing Replay) support with DeepEP and MTP
• Dependency upgrades: SGLang v0.5.7 and the Megatron dev branch

What's Changed

• add ckpt load save ci by @lilei199908 in #1104
• Add --rollout-all-samples-process-path for RLVE by @zhuzilin in #1107
• feat: support Qwen3 Moe BackEnd Kernel by @attack204 in #1071
• fix max response/context/prompt len by @lilei199908 in #1110
• fix max len by @lilei199908 in #1112
• [docker] remove amem and support deepep + r3 by @zhuzilin in #1115
• [Fix] Fix early return in init rollout engine by @yitianlian in #1118
• [Fix] Add sglang patch for weight version update by @yitianlian in #1119
• fix: improve tokenization by @nanjiangwill in #1113
• [Feature] Add CI test for weight version update by @yitianlian in #1120
• [docker] optimize r3 with base64 encode by @zhuzilin in #1124
• [docker] fix r3 gather buffer by @zhuzilin in #1129
• [docker] support mtp for r3 by @zhuzilin in #1131
• [Fix] Fix some bugs in retool example by @yitianlian in #1130
• Add finalize_model_grads_with_empty_cache by @zhuzilin in #1133
• Feat: add usage docs for fsdp by @lin0303-siyuan in #1092
• Reserve more ports for new sglang dp attn impl by @zhuzilin in #1142
• Blog: fix the path of the Blog's architecture image by @ShanningZhuang in #1125
• Support async save and add extra save at the end of the training by @zhuzilin in #1143
• fix: fix GemmeRMSNorm.forward() bug by @nanjiangwill in #1121
• [WIP][FSDP] Support FSDP for Qwen3Next by @rucnyz in #1116
• Megatron VLM Support (1/N) by @Zhuohao-Li in #1123
• Update deprecated huggingface-cli and fix broken links by @Lyken17 in #1147
• Added FSDP checkpoint handling to convert_torch_dist_to_hf.py by @cklxx in #1101
• minor fix for megatron compatibility by @zhuzilin in #1149
• Remove config_mapping to use megatron-bridge by @zhuzilin in #1166
• Avoids repeated work. by @qqwqqw689 in #1163
• Make tools/convert_torch_dist_to_hf.py not rely on megatron by @zhuzilin in #1167
• support converting dpsk mtp layer by @zhuzilin in #1169
• [FSDP] Add Masked importance sampling by @zijiexia in #1122
• [TIS/MIS] fix and add better metric by @ChangyiYang in #1174
• Fix optimizer schedule resume by @lr-tsinghua11 in #1152
• [docker] upgrade to megatron dev branch by @zhuzilin in #1153
• Minor fix by @lancerts in #1165
• Fix forward of Qwen3VLTextRotaryEmbedding in Megatron-Bridge by @zhuzilin in #1179
• Reuse the text llm config for qwen3 vl models by @zhuzilin in #1180
• Don't save AutoBridge in args by @zhuzilin in #1181
• [Fix] Fix port error in PD disaggregation setting by @yitianlian in #1175
• Fix prompt type bug in generate_with_search within examples/search-r1 by @jiahe7ay in #1182
• feat: support Qwen3 VL MoE by @nanjiangwill in #1171
• [Fix] Minor fix by @yitianlian in #1183
• Set parallel config for megatron bridge by @zhuzilin in #1184
• Fix tools/convert_hf_to_torch_dist.py by @zhuzilin in #1186
• Don't calculate entropy grad when coef is 0 by @zhuzilin in #1185
• Disable routing replay for critic by @zhuzilin in #1187
• Revert "Don't calculate entropy grad when coef is 0" by @zhuzilin in #1189
• Fix qwen3next for megatron dev branch by @zhuzilin in #1190
• fix: fix logging for rollout by @nanjiangwill in #1188
• sync internal features by @zhuzilin in #1192
• Fix check_weights api by @zhuzilin in #1194
• Add --custom-rollout-log-function-path and --custom-eval-rollout-log-function-path by @zhuzilin in #1196
• [Feature] Add more logging for health monitor by @yitianlian in #1195
• fix: SFT tools support by @maoquan-ms in #1198
• [Featuren] Change default value of rollout health check by @yitianlian in #1197
• Megatron VLM Support w/ SFT (2/N) by @Zhuohao-Li in #1150
• tiny fix for sft script after tokenizer improvement by @Zhuohao-Li in #1201
• tests: add test for multi turn loss mask by @maoquan-ms in #1204
• Always pass loss masks to model by @zhuzilin in #1205
• [on-policy distillation] update reward function to fix potential token mismatches by @ahxt in #1128
• Add ci for mtp by @zhuzilin in #1207
• Fix mla tflops by @lilei199908 in #1209
• update docs by @zhuzilin in #1211
• update docs by @zhuzilin in #1214
• [Feature] Support 0.3.0 sglang router for fault tolerance by @yitianlian in #1215
• sync internal features by @zhuzilin in #1216
• feat: add custom logic for processing list[list[Sample]] to training data by @nanjiangwill in #1218
• add int4_quant cuda kernel by @Hyaloid in #1220
• update doc by @zhuzilin in #1224
• Improve AMD tutorial with complete model/data setup workflow by @Vivicai1005 in #1212
• update megatron patch by @zhuzilin in #1228
• sync from internal by @zhuzilin in #1229
• fix model saving bug in megatron by @zhuzilin in #1230
• add new status by @nanjiangwill in #1219
• update customization docs by @nanjiangwill in #1233
• Revert data processing of VLM by @zhuzilin in #1232
• [VLM] optimize VLM processing by @nanjiangwill in #1234
• feat: add custom pg_loss reducer by @ChangyiYang in #1235
• fix: typo "sgalng" → "sglang" in ROCm Dockerfiles by @yurekami in #1282
• sync bugfix from internal by @zhuzilin in #1284
• sync internal bugfix by @zhuzilin in #1286
• add bshd support by @yueming-yuan in #1285
• [docker] fix bugs on pd disaggregation and add --disable-draft-cuda-graph by @zhuzilin in #1288
• Add longest_effective_sample_tokens_per_sec metric by @zhuzilin in #1291
• [fix] conditionally pass kwargs for megatron-bridge VLM by @yueming-yuan in #1290
• [VLM] Bugfix: image_patch_size for vision preprocessing by @coding-famer in #1227
• feat: add --custom-model-provider-path argument by @yurekami in #1239
• [Feature/Fix] Support IPv6 host resolution and robust URI formatting by @Chen-GX in #859
• Fix missing trust_remote_code in HfWeightIteratorBridge by @SwordFaith in #1287
• fix: remove invalid None default and fix misleading underscore variable naming by @lancerts in #1283
• fix: remove duplicate Megatron-LM installation in build_conda.sh by @yurekami in #1238
• fix dev megatron ckpt save bugs by @lilei199908 in #1294
• [Fix] fix image_patch_size in processing_utils by @coding-famer in #1295
• support hicache for pd disaggregation by @zhuzilin in #1296
• Optimize data.py for efficient data loading by @ppraneth in #696
• Auto Sync Code by @miles-code-angel in #1303
• [VLM] end2end geo3k multi-turn RL of VLM Recipe by @gxlvera in #1141
• [docker] Fix sglang ima on mtp + pd disaggregation by @zhuzilin in #1313
• fix: fix processing logic by @nanjiangwill in #1292
• [BugFix] Delete apply chat template for SFT by @PopSoda2002 in #1307
• Remove token retrieval test from main. by @qqwqqw689 in #1243
• update quick start doc by @zijiexia in #1193
• fix: replace blocking sleep with async sleep and fix file handle leak by @lancerts in #1200
• Set base_gpu_id for sglang from placement groups by @vpj in #1306
• [FSDP] Move gptoss scripts by @PopSoda2002 in #1317
• [refactor] Make sglang_rollout.py shorter and add prefix cached info by @zhuzilin in #1318
• set spec args for mtp ci by @zhuzilin in #1322
• Add non_generation_time stat in sample by @zhuzilin in #1323
• ad slime test images ci by @lilei199908 in #1325
• update sglang to lmsysorg/sglang:nightly-dev-20260103-24c91001 by @lilei199908 in #1324
• update sglang to lmsysorg/sglang:nightly-dev-20260103-24c91001 by @lilei199908 in #1331
• code sync by @miles-code-angel in #1329
• perf: replace quadratic list flattening with linear chaining in rollout manager by @ppraneth in #1319
• Implement local GPU ID remapping based on CUDA_VISIBLE_DEVICES for SGLang Engine by @zijiexia in #1327
• Fix: Remove --apply-chat-template from Qwen3-235B SFT script by @kaysonyu in #1315
• [Megatron Bridge] Support save hf format model by @coding-famer in #1289
• update default paths and disable offloading for AMD qwen3-4B training by @Vivicai1005 in #1225
• [internal sync] reset optimizer state and dynamic global batch size by @zhuzilin in #1330
• [refactor] minor code refactor for save_hf by @zhuzilin in #1334
• optimize long prompt filter by @zhuzilin in #1335
• remove deprecated interface by @zhuzilin in #1336
• Revert "remove deprecated interface" by @zhuzilin in #1337
• code cleanup by @zhuzilin in #1338
• save sglang v0.5.7 patch by @zhuzilin in #1339
• [Feature] Add CI for fault tolerance by @yitianlian in #1222
• Patch validate_non_overlapping_shards_metadata to speed up ckpt loading by @zhuzilin in #1342
• [Feature] Reorganize CI by @yitianlian in #1343
• [Feature] Option not to save optimizer states to save disk space by @yzlnew in #1333
• Add clear_num_new_engines and some code cleanup by @zhuzilin in #1349
• fix get_response_lengths by @zhuzilin in #1350
• bugfix by @zhuzilin in #1351
• default setting --tool-keys to tools by @UbeCc in #1352
• code sync by @miles-code-angel in #1356
• Revert "code sync" by @zhaochenyang20 in #1357
• update code by @miles-code-angel in #1358
• [sync] sync internal bugfixes by @zhuzilin in #1371
• feat: add int4 reinforcement learning training support (Part1) by @GeLee-Q in #1362
• [docker] Fix mtp r3 and add tilelang by @zhuzilin in #1380
• [docker] Comment out 'quant weights to fp8 ue8m0' by @zhuzilin in #1381
• geo3k VLM multi-turn megatron update by @gxlvera in #1378
• [Doc] Add docs for R2/R3 by @Hecate0821 in #1382
• [ci] borrow bot-slash-lint.yaml from miles by @zhuzilin in #1384
• feat: add int4 reinforcement learning training support (Part2) by @fy1214 in #1172
• feat: add int4 reinforcement learning training support (Part3) by @Gao016 in #1368
• fix lint by @zhuzilin in #1385
• VLM Multi-turn, add Megatron in README by @gxlvera in #1387
• Fix grammar and formatting in README.md by @zhaochenyang20 in #1388
• [refactor] refactor int4 qat code by @zhuzilin in #1390
• [1/X] Refactor: unify training backends by general utils, tested Megatron & FSDP alignment by @yueming-yuan in #1373
• bugfix by @zhuzilin in #1394
• bugfix by @zhuzilin in #1395
• fix ppo utils & mis by @yueming-yuan in #1396
• feat(examples): add strands-sglang integration for agentic RL with TITO support by @Lawhy in #1359
• remove swe-agent example by @zhuzilin in #1397
• fix to suppot dpsk-v3.2 bf16 weight convert to fp8 by @Gao016 in #1392
• Only rank0 should call post_process_weights by @zhuzilin in #1398
• bugfix by @zhuzilin in #1400
• [docker] support mtp in dpsk v3.2 by @zhuzilin in #1401
• bug gix by @lilei199908 in #1403
• [style] minor: remove subclass by @yueming-yuan in #1402
• [doc] suggest pip install -e . --no-deps by @zhuzilin in #1405
• [docs] add note for cudnn by @zhuzilin in #1406
• [minor] Delete unused util file by @yueming-yuan in #1408
• [docker] ignore slime in sglang SafeUnpickler by @zhuzilin in #1409
• add r3 ci by @lilei199908 in #1407
• Fix rollout-all-samples by @fzyzcjy in #1410
• [docs] fix ai generate response by @zijiexia in #1412
• bugfix on UpdateWeightFromDistributed by @zhuzilin in #1420
• [docker] add tunable indexer is_neox_style by @zhuzilin in #1421
• [docker] remove rm /root/.tmux.conf by @zhuzilin in #1422
• [sync] sync internal feature and bugfix by @zhuzilin in #1423
• [docker] update stable patches by @zhuzilin in #1424
• skip logits.div when temp is 1.0 by @zhuzilin in #1428
• [docker] support offload NSATokenToKVPool by @zhuzilin in #1429
• [doc] cleanup redundant example and scripts by @zhuzilin in #1431
• [docs] move low precision example into main doc by @zhuzilin in #1432
• [docs] move reporducibility to main doc by @zhuzilin in #1433
• [docs] a bit addition info for pd disaggregation by @zhuzilin in #1434
• [docs] add debug suggestion for ima by @zhuzilin in #1435
• Fix retool example incorrectly handling max_tool_calls by @fzyzcjy in #1427
• Docs: Add qqr to "Projects Built upon slime" section by @bcol23 in #1425
• [docs] fix doc by @zhuzilin in #1436
• Fix Hf model to Mcore checkpoint conversion on AMD gpus by @gramesh-amd in #279
• [cleanup] clean up utils folder by @zhuzilin in #1437
• [FSDP][Fix] Fix redundant import by @Hecate0821 in #1354
• [Fix] Update deprecated sglang ep args in docs and scripts by @coding-famer in #1344
• Add Qwen3-Coder-30B-A3B-Instruct model script by @maoquan-ms in #1213
• Revert "[style] minor: remove subclass" by @zhuzilin in #1441
• [revert] revert the parallel state change by @zhuzilin in #1442
• [FSDP] remove cp in fsdp by @zhuzilin in #1443
• [fsdp] remove tis by @zhuzilin in #1444
• Megatron VLM Support (Qwen2.5-VL series) (3/N) by @Zhuohao-Li in #1210
• fix the loss mask for mask_offpolicy_in_partial_rollout by @zhuzilin in #1445
• [Fix] Return origin_samples instead of False in filter_long_prompt by @kaysonyu in #1438
• [docker] fix sglang streaming output bug by @zhuzilin in #1446
• [docker] change base image from lmsysorg to slimerl/sglang by @zhuzilin in #1447
• Fix: Apply loss mask to KL in REINFORCE++ returns calculation by @kaysonyu in #1372
• [docs] add docs for ppo by @zhuzilin in #1448
• [release] bump to v0.2.2 by @zhuzilin in #1345

New Contributors

• @attack204 made their first contribution in #1071
• @lin0303-siyuan made their first contribution in #1092
• @ShanningZhuang made their first contribution in #1125
• @rucnyz made their first contribution in #1116
• @Lyken17 made their first contribution in #1147
• @cklxx made their first contribution in #1101
• @qqwqqw689 made their first contribution in #1163
• @zijiexia made their first contribution in #1122
• @lr-tsinghua11 made their first contribution in #1152
• @jiahe7ay made their first contribution in #1182
• @maoquan-ms made their first contribution in #1198
• @Hyaloid made their first contribution in #1220
• @Vivicai1005 made their first contribution in #1212
• @yurekami made their first contribution in #1282
• @miles-code-angel made their first contribution in #1303
• @gxlvera made their first contribution in #1141
• @vpj made their first contribution in #1306
• @kaysonyu made their first contribution in #1315
• @yzlnew made their first contribution in #1333
• @gramesh-amd made their first contribution in #279

Full Changelog: v0.2.1...v0.2.2