Commit Graph

956 Commits

Author SHA1 Message Date
Dave Lage
2379bb27e8 Merge branch 'sd3' into network-wavelet-loss 2025-08-03 02:32:07 -04:00
Kohya S
96feb61c0a feat: implement modulation vector extraction for Chroma and update related methods 2025-07-30 21:34:49 +09:00
Kohya S
6c8973c2da doc: add reference link for input vector gradient requirement in Chroma class 2025-07-28 22:08:02 +09:00
Kohya S
9eda938876 Merge branch 'sd3' into feature-chroma-support 2025-07-21 13:32:22 +09:00
Kohya S.
d98400b06e Merge pull request #2138 from kohya-ss/feature-lumina-image
Feature lumina image
2025-07-21 13:21:26 +09:00
Kohya S
0b763ef1f1 feat: fix timestep for input_vec for Chroma 2025-07-20 20:53:06 +09:00
Kohya S
b4e862626a feat: add LoRA training support for Chroma 2025-07-20 19:00:09 +09:00
Kohya S
c4958b5dca feat: change img/txt order for attention and single blocks 2025-07-20 16:30:43 +09:00
Kohya S
8fd0b12d1f feat: update DoubleStreamBlock and SingleStreamBlock to handle text sequence lengths instead of mask 2025-07-20 16:00:58 +09:00
Kohya S
404ddb060d fix: inference for Chroma model 2025-07-20 14:08:54 +09:00
kohya-ss
24d2ea86c7 feat: support Chroma model in loading and inference processes 2025-07-20 12:56:42 +09:00
Dave Lage
3adbbb6e33 Add note about why we are moving it 2025-07-16 16:09:20 -04:00
rockerBOO
a7b33f3204 Fix alphas cumprod after add_noise for DDIMScheduler 2025-07-15 22:36:46 -04:00
rockerBOO
38b858926d Revert "Add import hint"
This reverts commit 9cedf18a97.
2025-07-15 19:54:41 -04:00
rockerBOO
9cedf18a97 Add import hint 2025-07-15 19:53:54 -04:00
Kohya S
e0fcb5152a feat: support Neta Lumina all-in-one weights 2025-07-15 21:34:35 +09:00
Kohya S
a96d684ffa feat: add Chroma model implementation 2025-07-15 20:44:43 +09:00
rockerBOO
121cc23f2e Moved matplotlib into functions to not require it 2025-07-14 21:26:36 -04:00
rockerBOO
8cc81e45f7 Fix wavelet loss on non-flow matching models (sd1.5, SDXL). Fix wavelet coorelation. 2025-07-14 21:20:49 -04:00
Kohya S
30295c9668 fix: update parameter names for CFG truncate and Renorm CFG in documentation and code 2025-07-13 21:00:27 +09:00
Kohya S
999df5ec15 fix: update default values for timestep_sampling and model_prediction_type in training arguments 2025-07-13 20:52:00 +09:00
Kohya S
88960e6309 doc: update lumina LoRA training guide 2025-07-13 20:49:38 +09:00
Kohya S
b4d1152293 fix: sample generation with system prompt, without TE output caching 2025-07-09 21:55:36 +09:00
Kohya S
6731d8a57f fix: update system prompt handling 2025-06-29 22:21:48 +09:00
Kohya S
884c1f37c4 fix: update to work with cache text encoder outputs (without disk) 2025-06-29 21:58:43 +09:00
Kohya S
935e0037dc feat: update lumina system prompt handling 2025-06-29 21:33:09 +09:00
rockerBOO
0e929f97b9 Revert system_prompt for dataset config 2025-06-16 16:50:18 -04:00
rockerBOO
1db78559a6 Merge branch 'sd3' into update-sd3 2025-06-16 16:43:34 -04:00
Kohya S.
3e6935a07e Merge pull request #2115 from kohya-ss/fix-flux-sampling-accelerate-error
Fix unwrap_model handling for None text_encoders in sample_images
2025-06-15 21:14:09 +09:00
Kohya S
fc40a279fa Merge branch 'dev' into sd3 2025-06-15 21:05:57 +09:00
Disty0
bcd3a5a60a Update IPEX libs 2025-06-13 16:25:16 +03:00
rockerBOO
7c83ac4369 Add avg non-zero ratio metric 2025-06-10 13:17:04 -04:00
rockerBOO
d94bed645a Add lumina tests and fix image masks 2025-06-09 21:14:51 -04:00
rockerBOO
0145efc2f2 Merge branch 'sd3' into lumina 2025-06-09 18:13:06 -04:00
Kohya S
bb47f1ea89 Fix unwrap_model handling for None text_encoders in sample_images function 2025-06-08 18:00:24 +09:00
rockerBOO
9629853d15 Fix wavelet loss not separating levels. Refactor loss to be spatial 2025-06-05 22:03:52 -04:00
rockerBOO
0af0302c38 Metrics, energy, loss 2025-05-19 19:15:23 -04:00
rockerBOO
346790a996 Merge branch 'sd3' into network-wavelet-loss 2025-05-19 19:10:55 -04:00
rockerBOO
984472ca09 Fix metrics 2025-05-04 18:58:54 -04:00
rockerBOO
7be3c5dce1 Refactor SWT to work properly and faster. Add SWT tests 2025-05-04 18:57:53 -04:00
rockerBOO
964bfcb576 Fix padding for small latents. Add DWT tests 2025-05-04 18:57:42 -04:00
rockerBOO
d5f8f7de1f Add wavelet loss fn 2025-05-04 18:57:10 -04:00
rockerBOO
56dfdae7c5 Add QuaternionWaveletTransform. Update WaveletLoss 2025-05-04 18:55:50 -04:00
rockerBOO
19ce0ae61f Add wavelet_loss_band_level_weights 2025-05-04 18:39:51 -04:00
rockerBOO
7b9e92a8cc Fix band weights via toml. Add more logging 2025-05-04 18:39:48 -04:00
rockerBOO
20a99771bf Add back in padding 2025-05-04 18:39:45 -04:00
rockerBOO
f553b7bf31 Add wavelet loss recording 2025-05-04 18:39:36 -04:00
rockerBOO
6d42b95e2b Refactor transforms, fix loss calculations
- add full conditional_loss functionality to wavelet loss
- Transforms are separate and abstracted
- Loss now doesn't include LL except the lowest level
  - ll_level_threshold allows you to control the level the ll is
    used in the loss
- band weights can now be passed in
- rectified flow calculations can be bypassed for experimentation
- Fixed alpha to 1.0 with new weighted bands producing lower loss
2025-05-04 18:39:32 -04:00
rockerBOO
64422ff4a0 Suggest the right module 2025-05-04 18:39:25 -04:00
rockerBOO
837231a5c7 Add wavelet loss 2025-05-04 18:39:17 -04:00