Commit Graph

257 Commits

Author SHA1 Message Date
Jaret Burkett
193c1b2dfa Add a watcher to constantly check for stop signal from the UI. This will force a stop within 2 seconds instead of having to wait on a long hung process. 2025-08-31 16:58:01 -06:00
Jaret Burkett
056711d4ed Fix issue with wan22 14b that woudl load both transformers temporarily resulting in oom on 24GB. 2025-08-28 13:06:31 -06:00
Jaret Burkett
1f541bc5d8 Changes to handle a different DFE arch 2025-08-27 11:05:16 -06:00
Jaret Burkett
119653c3f2 Force width, height, and num frames to always be the proper sizes for Wan 2.2 models 2025-08-25 10:33:28 -06:00
Jaret Burkett
e1fd411665 Added support for Chroma1 official release. Will still use single file verstion instead of the diffusers version. 2025-08-23 09:06:28 -06:00
Jaret Burkett
aa99784b89 Add control to prompot encodings in the trainer when not cached 2025-08-21 16:52:13 -06:00
Jaret Burkett
bf2700f7be Initial support for finetuning qwen image. Will only work with caching for now, need to add controls everywhere. 2025-08-21 16:41:17 -06:00
Jaret Burkett
83deaec417 Minor bug fixes 2025-08-21 08:05:34 -06:00
Jaret Burkett
d2bbe1872c Add support for fine tuning Wan 2.2 I2V 14B 2025-08-18 11:43:32 -06:00
Jaret Burkett
b3e666daf4 Fix issue with wan22 14b where timesteps were generated not in the current boundary. 2025-08-16 21:16:48 -06:00
Jaret Burkett
6fffadfc0e Fixed a bug that prevented training just one stage of Wan 2.2 14b 2025-08-16 18:07:21 -06:00
Jaret Burkett
8ea2cf00f6 Added training to the ui. Still testing, but everything seems to be working. 2025-08-16 05:51:37 -06:00
Jaret Burkett
1c96b95617 Fix issue where sometimes the transformer does not get loaded properly. 2025-08-14 14:24:41 -06:00
Jaret Burkett
3413fa537f Wan22 14b training is working, still need tons of testing and some bug fixes 2025-08-14 13:03:27 -06:00
Jaret Burkett
259d68d440 Added a flushg during sampling to prevent spikes on low vram qwen 2025-08-12 12:57:18 -06:00
Jaret Burkett
77b10d884d Add support for training with an accuracy recovery adapter with qwen image 2025-08-12 08:21:36 -06:00
Jaret Burkett
4ad18f3d00 Clip max token embeddings to the max rope length for qwen image to solve for an error for super long captions > 1024 2025-08-10 08:44:41 -06:00
Jaret Burkett
f0105c33a7 Fixed issue that sometimes happens in qwen image where text seq length is wrong 2025-08-09 16:33:05 -06:00
Jaret Burkett
bb6db3d635 Added support for caching text embeddings. This is just initial support and will probably fail for some models. Still needs to be ompimized 2025-08-07 10:27:55 -06:00
Jaret Burkett
4c4a10d439 Remove vision model from qwen text encoder as it is not needed for image generation currently 2025-08-06 11:40:02 -06:00
Jaret Burkett
14ccf2f3ce Refactor qwen5b model code to be qwen 5b specific 2025-08-06 10:54:56 -06:00
Jaret Burkett
5d8922fca2 Add ability to designate a dataset as i2v or t2v for models that support it 2025-08-06 09:29:47 -06:00
Jaret Burkett
93202c7a2b Training working for Qwen Image 2025-08-04 21:14:30 +00:00
Jaret Burkett
9da8b5408e Initial but untested support for qwen_image 2025-08-04 13:29:37 -06:00
Jaret Burkett
a558d5b68f Move transformer back to device on aggresive wan 2.2 pipeline after generation. 2025-07-29 09:13:47 -06:00
Jaret Burkett
1d1199b15b Fix bug that prevented training wan 2.2 with batch size greater than 1 2025-07-29 09:06:25 -06:00
Jaret Burkett
ca7c5c950b Add support for Wan2.2 5B 2025-07-29 05:31:54 -06:00
Jaret Burkett
cefa2ca5fe Added initial support for Hidream E1 training 2025-07-27 15:12:56 -06:00
Daniel Verdu
a77ba5a089 fix: Guidance incorrect shape 2025-07-18 12:49:18 +02:00
Jaret Burkett
611969ec1f Allow control image for omnigen training and sampling 2025-07-09 13:54:55 -06:00
Jaret Burkett
bbb57de6ec Speed up omnigen TE loading 2025-07-05 09:32:00 -06:00
Jaret Burkett
5906a76666 Fixed issue with flux kontext forcing generation image sizes 2025-06-29 05:38:20 -06:00
Jaret Burkett
57a81bc0db Update base model version for kontext meta 2025-06-28 14:48:36 -06:00
Jaret Burkett
01a3c8a9b1 Fix device issue 2025-06-26 19:14:25 -06:00
Jaret Burkett
4f91cb7148 Fix issue with gradient checkpointing and flux kontext 2025-06-26 19:03:12 -06:00
Jaret Burkett
446b0b6989 Remove revision for kontext 2025-06-26 16:46:58 -06:00
Jaret Burkett
60ef2f1df7 Added support for FLUX.1-Kontext-dev 2025-06-26 15:24:37 -06:00
Jaret Burkett
8d9c47316a Work on mean flow. Minor bug fixes. Omnigen improvements 2025-06-26 13:46:20 -06:00
Jaret Burkett
84c6edca7e Merge branch 'main' into dev 2025-06-25 14:10:25 -06:00
Jaret Burkett
19ea8ecc38 Added support for finetuning OmniGen2. 2025-06-25 13:58:16 -06:00
Jaret Burkett
18513ec866 Merged in from main 2025-06-24 10:56:54 -06:00
Jaret Burkett
f3eb1dff42 Add a config flag to trigger fast image size db builder. Add config flag to set unconditional prompt for guidance loss 2025-06-24 08:51:29 -06:00
Jaret Burkett
ba1274d99e Added a guidance burning loss. Modified DFE to work with new model. Bug fixes 2025-06-23 08:38:27 -06:00
Jaret Burkett
8602470952 Updated diffusion feature extractor 2025-06-19 15:36:10 -06:00
Jaret Burkett
1cc663a664 Performance optimizations for pre processing the batch 2025-06-17 07:37:41 -06:00
Jaret Burkett
1c2b7298dd More work on mean flow loss. Moved it to an adapter. Still not functioning properly though. 2025-06-16 07:17:35 -06:00
Jaret Burkett
c0314ba325 Fixed some issues with training mean flow algo. Still testing WIP 2025-06-16 07:14:59 -06:00
Jaret Burkett
cbf04b8d53 Fixed some issues with training mean flow algo. Still testing WIP 2025-06-14 12:24:00 -06:00
Jaret Burkett
fc83eb7691 WIP on mean flow loss. Still a WIP. 2025-06-12 08:00:51 -06:00
Jaret Burkett
eefa93f16e Various code to support experiments. 2025-06-09 11:19:21 -06:00