Commit Graph

  • 4af2f294c6 Fix LoRA on FSDP2-sharded models: shape inference and module hooks fsdp2-dtensor-fixes Quentin Gallouédec 2026-08-21 23:15:00 +00:00
  • 41df5f482a Apply doc-builder style to the new docstring lora-moe-fused-delta Quentin Gallouédec 2026-08-21 21:53:41 +00:00
  • c0f55dfdb5 Fuse the MoE LoRA delta into a single baddbmm Quentin Gallouédec 2026-08-21 14:28:38 +00:00
  • 5d602fdbff FIX Resolve ZeRO-3 partitioned shapes in with LoRA target_parameters (#3576) main Quentin Gallouédec 2026-08-21 10:45:32 -04:00
  • 09ba20c4fc FEAT Add Super-Tuning as a new PEFT method (#3518) smellslikeml 2026-08-21 03:53:23 -07:00
  • b60f0552e7 FIX Reject duplicate adapter names early (#3559) Zhewen Tan 2026-08-20 22:22:37 +08:00
  • 86664fed1d DOC Fix GraloraConfig docstring (#3568) Rajarshi Datta 2026-08-20 19:48:58 +05:30
  • cf7256752d FIX PVeRA per-module sample_at_inference and generator device (#3563) Rajarshi Datta 2026-08-20 16:41:52 +05:30
  • a7d090d5bc STY Remove unused local variables in src/peft (#3567) Rajarshi Datta 2026-08-20 16:39:12 +05:30
  • 1f23250c92 CHORE Remove ipex config since it was removed from accelerate (#3564) Wang, Yi 2026-08-19 23:09:25 +08:00
  • cd7fb6ce4a CHORE Dockerfile installs mslk differently (#3557) Benjamin Bossan 2026-08-19 16:05:59 +02:00
  • 625c9e3f2e FIX Transformers warnings use return dict, missing inputs_embeds argument, fix causal_mask_creation_function call (#3558) Benjamin Bossan 2026-08-19 16:05:48 +02:00
  • 3a90cd4abe Fix IA3 Conv3d error message to reference Conv3d (#3505) uttam12331 2026-08-19 16:45:37 +05:30
  • 2254a93c9d CI: Bump the third-party-actions group across 1 directory with 7 updates (#3497) dependabot[bot] 2026-08-19 12:44:53 +02:00
  • a9d039ab50 DOC Fix wrong paper links for LoHa and LoKr (#3547) Benjamin Bossan 2026-08-18 15:04:16 +02:00
  • 6a687e9fba ENH Raise when a non-first adapter matches nothing (#3534) Singaraj B 2026-08-18 18:34:05 +05:30
  • b68870eadc DOC Clarify that config kwargs are not overrides (#3516) FNU AKSHANSH 2026-08-17 04:01:33 -07:00
  • 607895c990 FIX DoRA unmerge on error for Conv1D layers (#3532) Singaraj B 2026-08-14 17:11:43 +05:30
  • b8bf4075a0 method_comparison/LoKr: new hyper-params (#3535) githubnemo 2026-08-14 12:00:55 +02:00
  • 80cd114268 DOC Make guides/examples device agnostic (#3551) kaixuanliu 2026-08-14 17:54:02 +08:00
  • eba18ccf6b Method comparison: Add VBLoRA config to MetaMathQA (#3546) szzhoujiarui 2026-08-14 17:53:29 +08:00
  • e1ead75868 DOC Update CUDA windows ARM64 installation guide (#3550) Funtowicz Morgan 2026-08-14 10:22:09 +02:00
  • 82f4c0241c FIX Accept layers_to_transform=0 together with layers_pattern (#3425) Madan kumar 2026-08-13 21:59:13 +05:30
  • 1e58e3ff26 CHORE Upgrade ruff to 0.16.2 (#3545) Benjamin Bossan 2026-08-13 17:57:17 +02:00
  • 1159d5aace Method comparison: Reduce peak memory usage of image generation task (#3395) Benjamin Bossan 2026-08-13 17:21:26 +02:00
  • ad4e565c8a FEAT Add git dirty flag to method_comparison benchmark results (#3385) Peft Jambot 2026-08-13 17:08:36 +02:00
  • b3e4d6da42 FIX Issue with key matching when dealing with disk offloading (#3468) Benjamin Bossan 2026-08-13 17:08:05 +02:00
  • d29ce74f7a FIX TST Use distributed_config kwarg for TP tests with transformers >= 5.13 (#3502) Peft Jambot 2026-08-13 17:07:44 +02:00
  • 1a270a55f5 Method comparison: Small optimizations to image-gen (#3523) Benjamin Bossan 2026-08-13 17:07:17 +02:00
  • 5b6c5e325a TST Fix some multi GPU tests (#3458) Benjamin Bossan 2026-08-13 17:07:04 +02:00
  • 4ddb88b47b Refactor state dict handling (#3490) Benjamin Bossan 2026-08-13 17:06:30 +02:00
  • a429b59491 CHORE Type hints for helpers, hotswap, constants, integrations (#3529) Rudrendu Paul 2026-08-12 03:38:53 -07:00
  • 7f56e8491a DOC Install instructions for NVIDIA Spark (ARM64) devices (#3536) Funtowicz Morgan 2026-08-12 10:20:36 +02:00
  • 505e6fd682 MetaMathQA: consistent tokenizer padding behavior (#3526) githubnemo 2026-08-11 15:32:53 +02:00
  • 5f55a6331b FIX LoRA+ correctly sets lr for embeddings (#3503) Kaif Kohari 2026-08-06 13:05:13 +01:00
  • c702de4023 CHORE Improve type annotations of PeftModel and get_peft_model (#3493) Benjamin Bossan 2026-08-05 16:19:14 +02:00
  • 2735371fb3 FIX Check merging conflicting trainable tokens adapters (#3514) Daoyuan Li 2026-08-05 05:00:34 -07:00
  • dbb064a56b TST Extend RandLoRA test matrix (#3513) Ishan Katoch 2026-08-05 17:28:13 +05:30
  • b52ed1b544 FIX Reject deletion of merged adapters (#3504) Daoyuan Li 2026-08-05 04:57:29 -07:00
  • d1536ce168 TST Add regression tests for PEFT checkpoint loading (#3489) Benjamin Bossan 2026-08-04 14:08:43 +02:00
  • 3af2ad701a FIX Crash when unloading a model whose active adapter does not use modules_to_save (#3455) Javier De Jesus 2026-08-04 12:26:13 +02:00
  • 2969e633a9 ENH Add KaSA (Knowledge-aware Singular-value Adaptation) (#3446) Robbie Walmsley 2026-08-03 16:54:59 +02:00
  • 86f5f96192 docs: API coverage tooling + fixes (#3465) githubnemo 2026-08-03 16:39:15 +02:00
  • 3f830e6149 ENH Add Riemannian-preconditioned LoRA optimizer (#3382) smellslikeml 2026-08-03 05:00:43 -07:00
  • 1ea1da1676 fix(security): remediate workflow vulnerability in .github/workflows/torch_compile_tests.yml (#3501) hf-security-analysis[bot] 2026-08-03 13:55:08 +02:00
  • 6b2eda7b54 CI: Bump the ci-actions group with 2 updates (#3496) dependabot[bot] 2026-08-03 11:18:29 +02:00
  • 781cac4673 DOC Fix DeepSpeed guide's example script path (#3494) Recoordinate 2026-08-03 21:14:58 +12:00
  • ea8ebf36da Bump version to 0.20.1.dev0 (#3484) Benjamin Bossan 2026-07-31 14:22:15 +02:00
  • cf08deeeba DOC Improve VeRA documentation (#3476) Ashmitha R 2026-07-31 15:30:53 +05:30
  • 9f1fe21d81 FIX Merging bugs in FRoD, HiRA, and PEANuT (#3482) Daoyuan Li 2026-07-30 01:20:18 -07:00
  • a5526d27a9 Imagegen results for new PEFT methods (#3483) v0.20.0 githubnemo 2026-07-28 15:17:43 +02:00
  • 8587a36e3e Add get base model state dict (#3000) Irakli Salia 2026-07-28 16:48:33 +04:00
  • 3116c96b5d FIX AutoPeftModel forwards revision and token (#3442) Daoyuan Li 2026-07-27 02:58:25 -07:00
  • 388ccc43f5 ENH Improve param validation for several PEFT methods (#3438) Daoyuan Li 2026-07-27 02:57:32 -07:00
  • 051b2c5d9f CI Enable test coverage only on one of the runs (#3469) Benjamin Bossan 2026-07-24 16:50:42 +02:00
  • a1d4f86369 AI: Update contribution guide based on recent experience (#3459) Benjamin Bossan 2026-07-24 16:49:03 +02:00
  • 4d49a3b5cf Fix Gemma4 prefix tuning with per_layer_config (#3471) Peft Jambot 2026-07-24 14:06:36 +02:00
  • 18ab8257bb FIX Hotswapping of grouped Conv2d weights (#3416) (#3427) Sergio Vizcaino Ferrer 2026-07-24 14:05:40 +02:00
  • 14f7e14222 fix: respect layer-specific LoRA-FA scaling in optimizer (#3315) Sergio Vizcaino Ferrer 2026-07-23 17:52:27 +02:00
  • 999b934f13 ENH Support grouped convolutions in LoHa and LoKr (#3440) Daoyuan Li 2026-07-23 05:28:45 -07:00
  • 2333bbf5f3 Add quantization support for SHiRA (#3321) githubnemo 2026-07-23 12:16:31 +02:00
  • 5dbe5cefe5 Update AdaptionPrompt results (#3466) githubnemo 2026-07-23 11:31:08 +02:00
  • 167fcb0f29 Release 0.20.0 (#3388) Benjamin Bossan 2026-07-22 16:14:05 +02:00
  • 4c672987e0 FIX MiSS with bat init error with some quantization backends (#3381) Benjamin Bossan 2026-07-22 16:13:24 +02:00
  • ca541c04c1 FIX DEFT: Device handling and unmerge precision (#3412) Daoyuan Li 2026-07-22 05:40:44 -07:00
  • aee16509f9 FIX TinyLoRA: Shared weights are now adapter-specific (#3422) Daoyuan Li 2026-07-22 05:37:51 -07:00
  • 3652122d2e FIX HiRA ConvNd layers with groups > 1 error (#3430) Daoyuan Li 2026-07-22 05:35:19 -07:00
  • 81e417b69e FIX DeLoRA: Raise proper error on unsupported layer type (#3434) Daoyuan Li 2026-07-22 05:33:53 -07:00
  • 95f2a74d17 Method comparison: Packages can be in dist-packages (#3464) Anmoldeep Singh Arora 2026-07-22 18:02:58 +05:30
  • ae623e55bf DOC Troubleshooting entry for hook pitfall (#3415) Imran Ahamed 2026-07-22 07:31:12 -05:00
  • 39096cbec0 FIX BdLoraConfig.nblocks default being a tuple (#3456) Syed Osama Ali Shah 2026-07-22 15:28:07 +03:00
  • 1025a19b11 FIX Avoid bare Exception in peft_model.py (#3462) Hyacinth-of-Security 2026-07-22 20:26:53 +08:00
  • 194cb8feb2 Method comparison: Fix adaption prompt experiment (#3463) Benjamin Bossan 2026-07-21 12:02:07 +02:00
  • 72e71c2c02 Return False from is_gptqmodel_available when gptqmodel isn't installed (#3255) Charlie Tonneslan 2026-07-21 05:39:39 -04:00
  • cd81cf38f7 CI: Bump the third-party-actions group across 1 directory with 7 updates (#3375) dependabot[bot] 2026-07-21 11:05:44 +02:00
  • 667d0c30d5 DOC Change license URL from http:// to https:// in README (#3454) MOHAMMED HANAN M T P 2026-07-20 18:03:28 +05:30
  • 51c2ff666b Update MetaMAthQA results (#3445) githubnemo 2026-07-20 14:19:43 +02:00
  • cea8213158 Create SECURITY.md (#3432) githubnemo 2026-07-14 15:51:24 +02:00
  • d787f51bba FIX Bug in forgetting metric in MetaMathQA (#3419) Benjamin Bossan 2026-07-14 15:22:56 +02:00
  • 84564b6936 Create SECURITY.md githubnemo-patch-1 githubnemo 2026-07-14 14:41:24 +02:00
  • b753abbd11 DOC Improve IA3 documentation (#3417) Dresden 2026-07-14 20:13:13 +08:00
  • 6a832dec28 Minor app embed improvements (#3322) githubnemo 2026-07-13 08:57:15 +02:00
  • 79f4c36224 FIX Device mismatch for bnb dequantization (#3418) kaixuanliu 2026-07-10 20:51:41 +08:00
  • 7a493326ac FIX GLoRA safe_merge mutating the base weight (#3411) Daoyuan Li 2026-07-10 05:50:34 -07:00
  • 5083a21a23 FIX Crash in reduce_intruder_dimension with fp16/bf16 (#3410) Daoyuan Li 2026-07-10 05:49:19 -07:00
  • 88a2195411 FIX Delete adapter removes all associated attributes (#3376) Daoyuan Li 2026-07-10 05:48:27 -07:00
  • 1598ecb8fc DOC Expand installation guide (#3406) Ihor Vitenko 2026-07-08 16:43:59 +02:00
  • 7d4fb44318 FIX Typos in examples, tests, and docs (#3398) Daoyuan Li 2026-07-08 03:57:15 -07:00
  • 0fbf3c9e23 FIX Remove unnecessary dtype casts in LoRA get_delta_weight (#3396) Hyacinth-of-Security 2026-07-08 18:56:26 +08:00
  • f553f8ce95 ENH Refactor LoRA variant resolution (#3219) Sanjana soni 2026-07-08 16:24:49 +05:30
  • 8d5f7842d9 CI Fix error with Windows loading stable diffusion models (#3399) Benjamin Bossan 2026-07-07 16:50:37 +02:00
  • b681785232 Add default target_modules for nemotron_h hybrid Mamba-MoE models (#3289) Ce Zhao 2026-07-07 03:11:27 -07:00
  • 5378052090 fix(utils): call empty_cache() after fp16→fp32 casts in prepare_model_for_kbit_training (#3293) Umut Onur Yaşar 2026-07-07 11:57:23 +03:00
  • 08c72e37db TST Skip failing Diffusers tests until v0.40.0 (#3394) Benjamin Bossan 2026-07-06 20:59:31 +02:00
  • 74481e7536 CI Update doc-builder dep, make builds faster (#3393) Mishig 2026-07-06 15:09:22 +01:00
  • d0532e0ed6 FIX Resolve regex target_modules in MoE config conversion (#3232) Stefan Wang 2026-07-06 06:12:17 -07:00
  • 563d1d5701 CI Pin kernels version (#3377) Benjamin Bossan 2026-07-06 12:24:41 +02:00
  • 4ce7fe42fc DOC Fix code blocks in README.md examples (#3392) Hyacinth-of-Security 2026-07-06 18:09:21 +08:00
  • 2de96eb91c DOC Improve VeRA conceptual explanation (#3386) Sanjay M 2026-07-06 15:38:32 +05:30
  • e4fe61b431 ENH Allow multiple adapters when using target_parameters (#3350) Benjamin Bossan 2026-07-02 14:49:41 +02:00