Commit Graph

  • c8af252a44 refactor Robert 2025-04-22 16:19:14 +08:00
  • 7f984f4775 # saibit 2025-04-22 16:15:12 +08:00
  • d33d5eccd1 # saibit 2025-04-22 16:12:06 +08:00
  • 7c61c0dfe0 Add autocast warpper for forward functions in deepspeed_utils.py to try aligning precision when using mixed precision in training process saibit 2025-04-22 16:06:55 +08:00
  • 26db64be17 fix: update hf_hub_download parameters to fix wd14 tagger regression Glen 2025-04-19 11:54:12 -06:00
  • 629073cd9d Add guidance scale for prompt param and flux sampling Kohya S 2025-04-16 21:50:36 +09:00
  • 06df0377f9 Merge branch 'sd3' into flux-sample-cfg Kohya S 2025-04-16 21:27:08 +09:00
  • 176baa6b95 doc: update sd3 and sdxl training guides Kohya S 2025-04-16 12:32:43 +09:00
  • b1bbd4576c doc: update sd3 LoRA, sdxl LoRA advanced Kohya S 2025-04-14 21:53:21 +09:00
  • ceb19bebf8 update docs. sdxl is transltaed, flux.1 is corrected Kohya S 2025-04-13 22:06:58 +09:00
  • 8f5a2eba3d Add documentation for LoRA training scripts for SD1/2, SDXL, FLUX.1 and SD3/3.5 models Kohya S 2025-04-11 08:07:24 +09:00
  • dbd835ee4b train: Optimize VAE encoding by handling batch sizes for images vae_batch_size_sdxl_train kohya-ss 2025-04-08 21:57:16 +09:00
  • 5a18a03ffc Merge branch 'dev' into sd3 Kohya S 2025-04-07 21:55:17 +09:00
  • 572cc3efb8 Merge branch 'main' into dev Kohya S 2025-04-07 21:48:45 +09:00
  • 52c8dec953 Merge pull request #2015 from DKnight54/uncache_vae_batch Kohya S. 2025-04-07 21:48:02 +09:00
  • 4589262f8f README.md: Update recent updates section to include IP noise gamma feature for FLUX.1 Kohya S 2025-04-06 21:34:27 +09:00
  • c56dc90b26 Merge pull request #1992 from rockerBOO/flux-ip-noise-gamma Kohya S. 2025-04-06 21:29:26 +09:00
  • 7f93e21f30 fix typo sdbds 2025-04-06 16:21:48 +08:00
  • 9f1892cc8e Merge branch 'sd3' into lumina 青龍聖者@bdsqlsz 2025-04-06 16:13:43 +08:00
  • 1a4f1ff0f1 Merge branch 'lumina' of https://github.com/sdbds/sd-scripts into lumina sdbds 2025-04-06 16:09:37 +08:00
  • 00e12eed65 update for lost change sdbds 2025-04-06 16:09:29 +08:00
  • ee0f754b08 Merge pull request #2028 from rockerBOO/patch-5 Kohya S. 2025-04-05 20:15:13 +09:00
  • 606e6875d2 Merge pull request #2022 from LexSong/fix-resize-issue Kohya S. 2025-04-05 19:28:25 +09:00
  • fd36fd1aa9 Fix resize PR link Dave Lage 2025-04-03 16:09:45 -04:00
  • 92845e8806 Merge pull request #2026 from kohya-ss/fix-finetune-dataset-resize-interpolation Kohya S. 2025-04-03 21:52:14 +09:00
  • f1423a7229 fix: add resize_interpolation parameter to FineTuningDataset constructor Kohya S 2025-04-03 21:48:51 +09:00
  • b822b7e60b Fix the interpolation logic error in resize_image() Lex Song 2025-04-02 03:32:36 +08:00
  • ede3470260 Ensure all size parameters are integers to prevent type errors Lex Song 2025-04-02 03:28:58 +08:00
  • b3c56b22bd Merge branch 'dev' into sd3 Kohya S 2025-03-31 22:05:40 +09:00
  • 583ab27b3c doc: update license information in jpeg_xl_util.py Kohya S 2025-03-31 22:02:25 +09:00
  • aa5978dffd Merge pull request #1955 from Disty0/dev Kohya S. 2025-03-31 22:00:31 +09:00
  • aaa26bb882 docs: update README to include LoRA-GGPO details for FLUX.1 training Kohya S 2025-03-30 21:18:05 +09:00
  • d0b5c0e5cf chore: formatting, add TODO comment Kohya S 2025-03-30 21:15:37 +09:00
  • 59d98e45a9 Merge pull request #1974 from rockerBOO/lora-ggpo Kohya S. 2025-03-30 21:07:31 +09:00
  • 3149b2771f Merge pull request #2018 from kohya-ss/resize-interpolation-small-fix Kohya S. 2025-03-30 20:52:25 +09:00
  • 96a133c998 README.md: update recent updates section to include new interpolation method for resizing images Kohya S 2025-03-30 20:45:06 +09:00
  • 1f432e2c0e use PIL for lanczos and box Kohya S 2025-03-30 20:40:29 +09:00
  • 9e9a13aa8a Merge pull request #1936 from rockerBOO/resize-interpolation Kohya S. 2025-03-30 20:37:34 +09:00
  • 93a4efabb5 Merge branch 'sd3' into resize-interpolation Kohya S. 2025-03-30 19:30:56 +09:00
  • 381303d64f Update train_network.py DKnight54 2025-03-29 02:26:18 +08:00
  • 0181b7a042 Remove progress bar avg norms rockerBOO 2025-03-27 03:28:33 -04:00
  • 182544dcce Remove pertubation seed rockerBOO 2025-03-26 14:23:04 -04:00
  • e64dc05c2a Supplement the input parameters to correctly convert the flux model to BFL format; fixes #1996 laolongboy 2025-03-24 23:33:25 +08:00
  • 8ebe858f89 Merge branch 'dev' into sd3 Kohya S 2025-03-24 22:02:16 +09:00
  • a0f11730f7 Merge pull request #1966 from sdbds/faster_fix_sdxl Kohya S. 2025-03-24 21:53:42 +09:00
  • 30008168e3 Merge pull request #24 from rockerBOO/lumina-fix-max-norms 青龍聖者@bdsqlsz 2025-03-22 14:59:10 +08:00
  • 1481217eb2 Merge pull request #25 from rockerBOO/lumina-fix-non-cache-image-vae-encode 青龍聖者@bdsqlsz 2025-03-22 14:58:52 +08:00
  • 61f7283167 Fix non-cache vae encode rockerBOO 2025-03-21 20:38:43 -04:00
  • 2ba1cc7791 Fix max norms not applying to noise rockerBOO 2025-03-21 20:17:22 -04:00
  • 6364379f17 Merge branch 'dev' into sd3 Kohya S 2025-03-21 22:07:50 +09:00
  • 5253a38783 Merge branch 'main' into dev Kohya S 2025-03-21 22:07:03 +09:00
  • 8f4ee8fc34 doc: update README for latest v0.9.1 Kohya S 2025-03-21 22:05:48 +09:00
  • 367f348430 Merge pull request #1964 from Nekotekina/main Kohya S. 2025-03-21 21:59:03 +09:00
  • 89f0d27a59 Set sigmoid_scale to default 1.0 rockerBOO 2025-03-20 15:10:33 -04:00
  • d40f5b1e4e Revert "Scale sigmoid to default 1.0" rockerBOO 2025-03-20 15:09:50 -04:00
  • 8aa126582e Scale sigmoid to default 1.0 rockerBOO 2025-03-20 15:09:11 -04:00
  • e8b3254858 Add flux_train_utils tests for get get_noisy_model_input_and_timesteps rockerBOO 2025-03-20 15:01:15 -04:00
  • 16cef81aea Refactor sigmas and timesteps rockerBOO 2025-03-20 14:32:56 -04:00
  • d151833526 docs: update README with recent changes and specify version for pytorch-optimizer Kohya S 2025-03-20 22:05:29 +09:00
  • 936d333ff4 Merge pull request #1985 from gesen2egee/pytorch-optimizer Kohya S. 2025-03-20 22:01:03 +09:00
  • f974c6b257 change order to match upstream rockerBOO 2025-03-19 14:27:43 -04:00
  • 5d5a7d2acf Fix IP noise calculation rockerBOO 2025-03-19 13:50:04 -04:00
  • 1eddac26b0 Separate random to a variable, and make sure on device rockerBOO 2025-03-19 00:49:42 -04:00
  • 8e6817b0c2 Remove double noise rockerBOO 2025-03-19 00:45:13 -04:00
  • d93ad90a71 Add perturbation on noisy_model_input if needed rockerBOO 2025-03-19 00:37:27 -04:00
  • 7197266703 Perturbed noise should be separate of input noise rockerBOO 2025-03-19 00:25:51 -04:00
  • 5b210ad717 update prodigyopt and prodigy-plus-schedule-free gesen2egee 2025-03-19 10:49:06 +08:00
  • b81bcd0b01 Move IP noise gamma to noise creation to remove complexity and align noise for target loss rockerBOO 2025-03-18 21:36:55 -04:00
  • 6f4d365775 zeros_like because we are adding rockerBOO 2025-03-18 18:53:34 -04:00
  • a4f3a9fc1a Use ones_like rockerBOO 2025-03-18 18:44:21 -04:00
  • b425466e7b Fix IP noise gamma to use random values rockerBOO 2025-03-18 18:42:35 -04:00
  • c8be141ae0 Apply IP gamma to noise fix rockerBOO 2025-03-18 15:42:18 -04:00
  • 0b25a05e3c Add IP noise gamma for Flux rockerBOO 2025-03-18 15:40:40 -04:00
  • 3647d065b5 Cache weight norms estimate on initialization. Move to update norms every step rockerBOO 2025-03-18 14:25:09 -04:00
  • 620a06f517 Check for uppercase file extension too Disty0 2025-03-17 17:44:29 +03:00
  • 564ec5fb7f use extend instead of += Disty0 2025-03-17 17:41:03 +03:00
  • 7e90cdd47a use bytearray and add typing hints Disty0 2025-03-17 17:26:08 +03:00
  • c898e4e536 fix: optimize weight device swapping with no_grad context kohya-ss 2025-03-17 21:28:02 +09:00
  • e5b5c7e1db Update requirements.txt gesen2egee 2025-03-15 13:29:32 +08:00
  • 7482784f74 Merge pull request #23 from rockerBOO/lumina-lora 青龍聖者@bdsqlsz 2025-03-09 21:04:45 +08:00
  • ea53290f62 Add LoRA-GGPO for Flux rockerBOO 2025-03-06 00:00:38 -05:00
  • 75933d70a1 Merge pull request #1960 from kohya-ss/sd3_safetensors_merge Kohya S. 2025-03-05 23:28:38 +09:00
  • aa2bde7ece docs: add utility script for merging SD3 weights into a single .safetensors file Kohya S 2025-03-05 23:24:52 +09:00
  • e8c15c7167 Remove log rockerBOO 2025-03-04 02:30:08 -05:00
  • 9fe8a47080 Undo dropout after up rockerBOO 2025-03-04 02:28:56 -05:00
  • 1f22a94cfe Update embedder_dims, add more flexible caption extension rockerBOO 2025-03-04 02:21:05 -05:00
  • 5e45df722d update gemma2 train attention layer sdbds 2025-03-04 08:07:33 +08:00
  • 09c4710d1e Merge pull request #22 from rockerBOO/sage_attn 青龍聖者@bdsqlsz 2025-03-03 10:26:02 +08:00
  • 3f49053c90 fatser fix bug for SDXL super SD1.5 assert cant use 32 sdbds 2025-03-02 19:32:06 +08:00
  • dfe1ab6c50 Merge pull request #21 from rockerBOO/lumina-torch-dynamo-gemma2 青龍聖者@bdsqlsz 2025-03-02 18:31:13 +08:00
  • b6e4194ea5 Merge pull request #20 from rockerBOO/lumina-system-prompt-special-token 青龍聖者@bdsqlsz 2025-03-02 18:30:49 +08:00
  • b5d1f1caea Merge pull request #19 from rockerBOO/lumina-block-swap 青龍聖者@bdsqlsz 2025-03-02 18:30:37 +08:00
  • d6c3e6346e Merge pull request #18 from rockerBOO/fix-sample-batch-norms 青龍聖者@bdsqlsz 2025-03-02 18:30:24 +08:00
  • 800d068e37 Merge pull request #17 from rockerBOO/lumina-cache-text-encoder-outputs 青龍聖者@bdsqlsz 2025-03-02 18:30:08 +08:00
  • 3817b65b45 Merge pull request #16 from rockerBOO/lumina 青龍聖者@bdsqlsz 2025-03-02 18:29:44 +08:00
  • a69884a209 Add Sage Attention for Lumina rockerBOO 2025-03-01 20:37:45 -05:00
  • acdca2abb7 Fix [occasionally] missing text encoder attn modules Ivan Chikish 2025-03-01 17:06:17 +03:00
  • ba5251168a fix: save tensors as is dtype, add save_precision option Kohya S 2025-03-01 10:31:39 +09:00
  • cad182d29a fix torch compile/dynamo for Gemma2 rockerBOO 2025-02-28 18:30:16 -05:00
  • a2daa87007 Add block swap for uncond (neg) for sample images rockerBOO 2025-02-28 14:22:39 -05:00