Commit Graph

  • 100e9a42c6 [KT] Support Kimi K2.5/2.6 LoRA fine-tuning (#10826) main yyj 2026-09-09 21:12:56 +08:00
  • 31078aa10a [v1] Support multimodal Ulysses CP and memory-efficient chunk loss for SFT (#10762) xvxuopop 2026-09-09 19:22:25 +08:00
  • 673048c6a5 [v1] fix non-persistent buffer sync for init_on_rank0 (#10820) Hazeldxq 2026-09-08 16:00:35 +08:00
  • dced5f8804 [train] support HyperParallel Expert-Parallel (#10811) Pangxz 2026-09-04 16:15:23 +08:00
  • 4451765a6b [model] add Qwen3.8 model support (#10749) Hertz 2026-09-02 15:52:22 +08:00
  • d6bb97ddff [data] fix position ids after merging packed-mrope (#10783) haqishen 2026-08-31 17:13:06 +09:00
  • 6369f90cee docs: refine PenguinHarness star invitation (#10806) Yaowei Zheng 2026-08-31 15:58:16 +08:00
  • e4d79d2993 docs: encourage following and starring PenguinHarness (#10805) Yaowei Zheng 2026-08-31 15:38:00 +08:00
  • 6f38e73b82 [data] add minicpm5 template with XML tool calling (#10801) zyk_computer 2026-08-31 13:16:20 +08:00
  • 7fcf5b3b13 [v1] support LoRA with FSDPTurbo expert parallelism (#10791) xvxuopop 2026-08-27 18:50:56 +08:00
  • 273a988ebe docs: refresh PenguinHarness description (#10793) Yaowei Zheng 2026-08-27 05:14:23 +08:00
  • a18110d2f0 [docker] add NPU image tag history (#10786) xvxuopop 2026-08-25 16:53:02 +08:00
  • c4e09c7cbe [v1] support GDN Ulysses cp (#10727) cxy 2026-08-20 18:53:39 +08:00
  • ff6d4d12ee [train] Add KTransformers VLM fine-tuning support (#10760) liyuemathematician 2026-08-18 19:49:47 +08:00
  • f28afaf635 [v1] add FSDPTurbo EP/EFSDP plugin for MoE training (#10676) Hazeldxq 2026-08-13 20:45:55 +08:00
  • bc4b42cefc [train] Harden KTransformers MoE LoRA SFT integration (#10738) yyj 2026-08-13 20:43:15 +08:00
  • 199b8873d7 [train] add fa3 supported (#10742) 浮梦 2026-08-13 20:39:01 +08:00
  • 0bbe481e6e [docker] upgrade NPU images to CANN 9.1 and PyTorch 2.10 (#10729) xvxuopop 2026-08-10 11:20:41 +08:00
  • 63a89710c7 [data] pad position_ids on non-FA2 packing path (fixes rotary crash for Gemma-3/4) (#10737) haqishen 2026-08-09 17:00:10 +09:00
  • 887b850813 [assets] fix broken links in README (#10728) richboyneedcash 2026-08-06 15:09:01 +08:00
  • 84576b1408 [v1] refactor NPU kernel matching by model type (#10643) xvxuopop 2026-08-04 19:38:41 +08:00
  • 713b5a3f95 [model] add MOSS-VL support (#10708) SSSSuperC 2026-08-03 18:18:24 +08:00
  • 62ae362455 [v1] Support multimodal data training (#10656) 浮梦 2026-07-31 18:54:13 +08:00
  • 3984675dd5 fix(ci): align workflow Python version with requires-python (#10707) Kyungmin Kim 2026-07-31 17:56:00 +09:00
  • 1b47415a2f [train] Fix hyper parallel tail accumulation loss scaling (#10705) Chaoran Wei 2026-07-30 17:29:59 +08:00
  • 9ce6b663e9 [train] support megatron-bridge for PT/SFT training (#10645) sunyi0505 2026-07-27 18:45:18 +08:00
  • 2ebe7be611 [ci] pin ruff version and fix lint errors (#10681) Yaowei Zheng 2026-07-24 16:29:58 +08:00
  • 3f77101580 [v1] refactor registry plugin structure and params (#10641) Jiaqi 2026-07-24 15:23:21 +08:00
  • 19e9fe3ced [docker] improve NPU image build and distribution (#10664) xvxuopop 2026-07-24 15:22:01 +08:00
  • d0eaa10b0c [docs] update readme (#10678) Yaowei Zheng 2026-07-24 00:09:09 +08:00
  • a17afe5e1b [docs] update trend badge and promote PenguinHarness in readme (#10677) Yaowei Zheng 2026-07-23 23:52:05 +08:00
  • ef2d8f9da6 [v1] fix grad norm and lr log (#10640) HelloWorldBeginner 2026-07-17 22:50:13 +08:00
  • 5f653cb96a [v1] add muon optimizer (#10618) HelloWorldBeginner 2026-07-17 21:44:28 +08:00
  • d1049d650a [docs] Add AMD GPU Cloud link (#10649) GaoYuYang 2026-07-15 17:52:24 +08:00
  • 8489928769 [fix]update license check, update transformers (#10632) 浮梦 2026-07-13 17:30:36 +08:00
  • b61140db3e [v1] replace custom template system with apply_chat_template (#10598) 浮梦 2026-07-10 21:15:44 +08:00
  • ea31c43d80 [v1] improve getting started guide with comprehensive content (#10626) Hyacinth-of-Security 2026-07-08 19:55:23 +08:00
  • 76a0391ddd [misc] fix ray initialization comment typo (#10628) Karunanidhi Mishra 2026-07-07 04:30:58 -05:00
  • 445163ab5e [misc] fix typos in comments and help text (#10633) Karunanidhi Mishra 2026-07-07 04:30:34 -05:00
  • d58ec6a0bc [deps] exclude broken transformers release (#10634) Karunanidhi Mishra 2026-07-07 04:30:26 -05:00
  • 5987a8dd68 [webui] add seed controls for reproducibility (#10629) Karunanidhi Mishra 2026-07-07 04:30:09 -05:00
  • a61cfa692a [readme] Revise bitsandbytes installation instructions in README (#10621) zhangzhengshan 2026-07-03 13:16:11 +08:00
  • 7a83d28ce3 [readme] Revise bitsandbytes installation instructions (#10622) zhangzhengshan 2026-07-03 13:15:43 +08:00
  • c8a082e0e3 [fix] Fixes Qwen3-VL prompt expansion for multiple videos (#10518) luca-888 2026-07-02 11:06:42 +08:00
  • a48af5cc69 [data] clarify _nothink suffix warning for reasoning-only models (#10613) GSCSD1 2026-06-30 17:15:29 +08:00
  • c383c0d067 [model] add Qwen-AgentWorld-35B-A3B support (#10615) souljoy 2026-06-30 17:15:09 +08:00
  • 50ff45176a [v1] set flash_attn to flash_attention_2 for ulysses CP example (#10616) HelloWorldBeginner 2026-06-30 16:56:12 +08:00
  • 9c0b4b3835 [v1][feature] add dpo trainer (#10544) codingma 2026-06-26 15:32:10 +08:00
  • b7615dbdc9 [v1] Fix device mesh, fix lora for reward model and fix sp (#10555) jiaqiw09 2026-06-25 20:05:56 +08:00
  • 666ee0ca78 [fix] redundant transformers check (#10602) Artyom Iudin 2026-06-24 11:34:59 +03:00
  • aca54c7f17 [model] add Hy-MT2-1.8B/7B support (#10605) souljoy 2026-06-24 16:34:53 +08:00
  • 48aa9ef084 [docs] update supported models list for MiniCPM 4/5 (#10603) souljoy 2026-06-24 15:04:15 +08:00
  • c928c1cb21 [assets] update llamafactory sft skill guidance (#10600) GaoYuYang 2026-06-23 17:22:56 +08:00
  • c35b7d7f55 [assets] add llamafactory sft skills (#10597) GaoYuYang 2026-06-22 17:01:21 +08:00
  • 802bcfe969 [feat] support HyperParallel Context Parallel feature (#10559) Chaoran Wei 2026-06-22 07:40:44 +08:00
  • 8792f06161 [webui] Fix WebUI training hang from subprocess log pipe (#10584) summernight 2026-06-17 15:36:40 +08:00
  • 8669a22e9c [fix] fix liger kernel patch for npu (#10583) jiaqiw09 2026-06-16 18:21:52 +08:00
  • 897a44386c [docs] add DataFlow and DataFlex blog tutorials (#10582) Hao Liang 2026-06-16 14:20:36 +08:00
  • 7a1e9630f2 [fix] update ascend doc link (#10572) jiaqiw09 2026-06-15 13:55:53 +08:00
  • cabe59a343 [model] add MiniCPM5-1B-Chat (#10558) souljoy 2026-06-10 16:18:27 +08:00
  • 9ca4026efe [model] handle unsloth model loading fallback during checkpoint resume (#7156) (#10551) Co-Cl2 2026-06-09 01:01:01 +08:00
  • 0b7aaf8f6a [fix] correctly place new token embeddings when embedding is padded (#10547) Ximing Xing 2026-06-05 10:47:51 +08:00
  • 8a4f6a3da5 [model] add gemma-4-12B-it (#10549) codingma 2026-06-04 23:43:20 +08:00
  • 409e8a477f [model] Patch GDN for NPU (#10504) A1waysBeenHere 2026-06-04 16:39:02 +08:00
  • 053d43c0ac [feat] support HyperParallel PT training and activation optimization (#10370) Cui-yshoho 2026-06-02 22:39:32 +08:00
  • a98a1ef101 [docs] fix README citation typo (#10540) Zhao73 2026-06-01 21:04:53 +08:00
  • 8ef7335b6a [misc] set dev version (#10533) Yaowei Zheng 2026-05-31 00:16:07 +08:00
  • 7af909522a [version] release v0.9.5 (#10532) v0.9.5 Yaowei Zheng 2026-05-30 23:57:09 +08:00
  • e016d2480e [fix] Fix NPU FusedMoE and RMSNorm (#10512) xvxuopop 2026-05-30 21:42:54 +08:00
  • 7d719182c9 [model] fix non-packing batch (bsz>1) for Qwen3.5 with flash attention (#10529) jiaqiw09 2026-05-30 21:41:41 +08:00
  • 01398eb18d [v1] fix padding free with sp (#10513) jiaqiw09 2026-05-26 23:49:21 +08:00
  • 8e68764b65 [v1] Implement dynamic padding-free stretrgy for batching (#10507) cxy 2026-05-25 20:40:21 +08:00
  • 16ff5a23cb [fix] use getattr for profiler attrs to support MCA TrainingArguments (#10506) Copilot 2026-05-21 17:26:29 +08:00
  • bdcb92d035 [v1] Add FlashAttention selection and implement normal / padding-free / dynamic batching (#10469) jiaqiw09 2026-05-21 17:14:19 +08:00
  • 7e20db5735 [v1] support liger_kernel (#10493) sunyi0505 2026-05-21 11:44:56 +08:00
  • 2322bf1cc2 [v1] add cuda fused moe kernel, implementing with triton (#10481) 浮梦 2026-05-20 20:49:42 +08:00
  • 368c48968f [callback] add torch profiler callback (#10463) 浮梦 2026-05-20 20:47:52 +08:00
  • 8b5ea65770 [v1] support reward training stage (#10431) 浮梦 2026-05-20 20:46:52 +08:00
  • 40e786d016 [data] add missing return statement in MiniCPM V Plugin (#10500) Dennis Huang 2026-05-20 01:50:00 +08:00
  • 6b9df75ab9 [docker] update npu docker (#10479) xvxuopop 2026-05-13 20:56:43 +08:00
  • ca50f22c38 [fix] Fix MiniCPM-V-4.6 image preprocessing behavior (#10478) 马境远 2026-05-12 11:35:23 +08:00
  • 53e77a9bfa [model] support MiniCPM-V-4.6 (#10472) 马境远 2026-05-08 18:14:34 +08:00
  • 55bd4944b6 [fix] fix qwen3_6 template doc (#10470) 浮梦 2026-05-08 11:47:02 +08:00
  • 7e09152275 fix(data/converter): handle None tool_calls in OpenAI-style messages (#10455) Tai An 2026-05-07 02:44:41 -07:00
  • 1e503a982d [assets] correct typo in examples/README_zh.md (#10462) simulikeit 2026-05-07 00:42:01 +08:00
  • 8752280dd7 [data] Optimize QwenVL video dataset preprocessing (#10404) luca-888 2026-05-03 18:36:56 +08:00
  • 468723c5d9 [packing] fix GDN crash when meeting dummy image (#10453) Kingsley 2026-05-01 12:10:13 +08:00
  • 887ee2b121 [refactor] Add KTransformers AMX MoE SFT support via Accelerate (#10430) Peilin Li 2026-05-01 01:47:58 +08:00
  • 6b08b948c9 [misc] bump transformers version upperbound (#10446) Kingsley 2026-05-01 01:30:11 +08:00
  • f7f3bfcbd7 [model] support Hy3-Preview (#10432) Hertz 2026-04-29 23:21:13 +08:00
  • 3475198d1e [fa2] fix IMA when train qwen3_5 (#10448) Kingsley 2026-04-29 20:20:55 +08:00
  • 50945ef850 [v1] fix device_mesh and sp for fsdp2 (#10429) sunyi0505 2026-04-28 11:20:11 +08:00
  • 2f0bef207a [export] handle NotImplementedError in export_model for transformers>=5.0 (fixes #10410) (#10438) Octopus 2026-04-27 23:36:23 +08:00
  • 2092abc217 [npu] add Qwen3.5 support with Partial RoPE and Hybrid Attention (#10421) curnane-lab 2026-04-27 23:36:07 +08:00
  • 99464b3d03 [misc] code lint (#10439) Kingsley 2026-04-27 14:07:31 +08:00
  • 9a0cfdccfa [v1] fix init on meta in transformers v5 (#10414) jiaqiw09 2026-04-27 00:37:09 +08:00
  • c8890c32db [data] support discard history cot for multiturn (#10435) Kingsley 2026-04-27 00:32:44 +08:00
  • 79c8332e4c [train] add qwen35 patch for neat_packing (#10436) Kingsley 2026-04-27 00:31:49 +08:00
  • e0bc3c1971 [v1] fix epoch and steps (#10422) jiaqiw09 2026-04-23 17:29:06 +08:00
  • ecca167eb4 [model] support qwen3.6 models (#10415) 浮梦 2026-04-22 19:44:01 +08:00