github/ai-toolkit
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity
1,300 Commits 9 Branches 0 Tags
bbaef7852aa75b6106a8314b551ec16094b2ad6e
Commit Graph

1300 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
Jaret Burkett
bbaef7852a Do not materalize weights on ostris quantizer when getting state dict. Require dequantization of weight like other quantization methods. 2026-07-16 08:55:51 -06:00
Jaret Burkett
31c45cf37d Update huggingface hub requirement. Remove forced xet removal as some models are so large it wont work without it. Hopefully the latest version of huggingface_hub doesnt have the hanging issues. 2026-07-16 08:52:24 -06:00
Jaret Burkett
e1e1996c16 Rework the logging and terminal emulator on the ui to work like an actual emulator for better output. 2026-07-16 08:50:40 -06:00
Jaret Burkett
5cb54ba9cc Allow setting weight saving flag on hidream_o1 2026-07-16 07:40:25 -06:00
fatalis
741aeb9ce0 Clear stale return-to-queue flag when starting jobs, fixes crash loop (#920) 2026-07-15 12:48:39 -06:00
Jaret Burkett
fe619405f3 Merge branch 'main' of github.com:ostris/ai-toolkit 2026-07-15 12:44:47 -06:00
DasPauluteli
a92f18bf71 krea2: don't hardcode the NVIDIA-only cuDNN SDPA backend (#933) 
* krea2: don't hardcode NVIDIA-only cuDNN SDPA backend

The krea2 attention() forced SDPBackend.CUDNN_ATTENTION, which is
NVIDIA-only. On non-NVIDIA backends (AMD ROCm, Intel XPU, Apple MPS)
every forward pass fails with 'RuntimeError: No available kernel.
Aborting execution.', so Krea 2 LoRA training cannot run at all there.

Pass a priority list [CUDNN, FLASH, EFFICIENT, MATH] instead. NVIDIA
still selects cuDNN; other backends fall back to flash/efficient/math.
Verified training end-to-end on an AMD Radeon 8060S (gfx1151, ROCm 7.2).

* Version bump

* Add set priority flag so CUDNN_ATTENTION is selected on cuda devices first.

---------

Co-authored-by: Jaret Burkett <jaretburkett@gmail.com>
 2026-07-15 12:44:34 -06:00
Jaret Burkett
4f5974ffa1 Version bump 2026-07-15 12:17:35 -06:00
Jaret Burkett
b8f8a08ba4 Fix sampling bar with anima 2026-07-15 12:17:06 -06:00
rmatif
3e6bd874c4 feat: Add Anima support (#860) 
* Add Anima training support

* Update Anima modular training

* Use sample guidance for Anima

* Fix Anima sampling

* Limit Anima LoRA targets

* Convert Anima LoRA exports

* Fix Anima local loading

* Update Anima default model

* Pin upstream Anima diffusers

* Adjust template defaults to be consistent with other models. Update README

---------

Co-authored-by: Jaret Burkett (Ostris) <jaretburkett@gmail.com>
 2026-07-15 11:59:01 -06:00
fatalis
8bbd051667 Add sample_start_step setting to configure when sampling starts (#949) 
Co-authored-by: Jaret Burkett <jaretburkett@gmail.com>
 2026-07-15 11:15:50 -06:00
Zironic
4ece17b71f Fix adapter scalar handling under torch.compile (#946) 
* Fix adapter scalar handling under torch.compile

* Fix instance where error could happen when merging in a lora to the base model

---------

Co-authored-by: Rydén Johan <johan.ryden@bostad.uppsala.se>
Co-authored-by: Jaret Burkett <jaretburkett@gmail.com>
 2026-07-15 11:07:55 -06:00
PlagueKind
e44c34a955 fix lokr speed and convrot compile (#945) 2026-07-15 10:46:30 -06:00
Jaret Burkett
30162c0602 Improvements for captioner quantization to speed it up. Block compile on captioners. 2026-07-15 10:25:47 -06:00
Jaret Burkett
e28727d5cb Made a fused GEMV kernel for convrot unpacking to increase speed further. Fix bug in test script that made train time add additional grads to bf16. 2026-07-15 10:24:52 -06:00
Jaret Burkett
691ddf434e Add Qwen3.6 VL captioner. 2026-07-15 07:02:33 -06:00
Jaret Burkett
18da85153b Disable xet by default. Seems to be causing a lot of hanging issues. 2026-07-14 10:42:03 -06:00
Jaret Burkett
cf0db39ede Fix some errors for specific quants. Enable saving quantizations. 2026-07-14 07:25:09 -06:00
Jaret Burkett
abba6b5845 Show better errors on captioner 2026-07-14 07:19:03 -06:00
Jaret Burkett
8b5bf25b13 Add convrot quants to captioner 2026-07-14 07:03:10 -06:00
Jaret Burkett
676b4f3c4c Add Automagic3 optimizer to the ui 2026-07-14 06:39:10 -06:00
Jaret Burkett
0d53e5e1f9 Fix compile with lokr on convrot 2026-07-13 19:24:57 -06:00
Jaret Burkett
a5f857ddb0 Added patch from Fatalis to fix lokr offloading with convrot 2026-07-13 18:48:56 -06:00
Jaret Burkett
28f2c0acbe Move z_image over to the new modeling class 2026-07-13 17:12:30 -06:00
Jaret Burkett
dcb3b329b2 Fix issue with the graph with sparse data when shrinking smoothing 2026-07-13 17:10:38 -06:00
Jaret Burkett
1f7d608e20 Show sparse datapoints on the loss graph. 2026-07-13 10:59:57 -06:00
Jaret Burkett
7602e476eb Exclude sensative layers from quantization in krea 2026-07-13 10:14:53 -06:00
Jaret Burkett
28b05ee4ed Added convrotbitnet 1.58 bit quantization 2026-07-13 07:52:48 -06:00
Jaret Burkett
a259fa07cd Make convrot ui clearer 2026-07-13 06:44:27 -06:00
Jaret Burkett
0b62e516cc Version Bump 2026-07-13 06:34:21 -06:00
Jaret Burkett
b6ff367633 Convrot improvements. Add argitrary bit packed byte quantizations. 2026-07-13 06:33:54 -06:00
Jaret Burkett
64663c8575 Add Boogu to the readme. 2026-07-11 21:03:00 -06:00
Jaret Burkett
4625406093 WIP working on convrot offloading 2026-07-11 15:37:28 -06:00
Jaret Burkett
1d1e21177a Add ostris linear layer to linear layer searches. 2026-07-11 14:21:16 -06:00
Jaret Burkett
095d6e7418 Add convrot4 and convrot8 quantizations to the ui. A lot of improvements to the convrot quantization methods. 2026-07-11 13:49:41 -06:00
Jaret Burkett
933ca1c517 Apply w8a8 on the forward when training with convrot for more speed. Fix issue where quantizing a model had a pin memory leak. 2026-07-11 10:53:50 -06:00
Jaret Burkett
065ac27353 Fix issue with compiling convrot on some models 2026-07-11 09:26:12 -06:00
Jaret Burkett
96a3a06111 Added initial experimental support for convrot4 and convrot8 quantization. 2026-07-10 11:01:14 -06:00
Jaret Burkett
6fac83d068 Fix casting issue with orbit quant 2026-07-10 10:23:38 -06:00
Jaret Burkett
71c75357eb Add cached conditioning recovery to wan 22 5b model 2026-07-10 10:22:24 -06:00
Jaret Burkett
ad07b06de5 Use cached first frame for wan22_5 model 2026-07-10 10:21:34 -06:00
Jaret Burkett
886c2aec57 Allow for vae tiling onle without low vram on wan models with a model kwarg 2026-07-10 09:42:52 -06:00
Jaret Burkett
fe82487187 Add tiling on vae decode for qwen image models when low vram flag is on 2026-07-10 09:07:29 -06:00
Jaret Burkett
e7951ad29e Exclude sensative layers from quantization on wan models 2026-07-10 08:56:05 -06:00
Jaret Burkett
883d60eb71 Do vae tiling when decoding wan models with low vram active. 2026-07-10 08:02:32 -06:00
Jaret Burkett
fed9357234 Fix ui to show kv cache for krea2 edit raw 2026-07-09 15:48:46 -06:00
Jaret Burkett
5a9b5bde3f Added experimental orbit quant 2026-07-09 15:14:51 -06:00
Jaret Burkett
a4bbe167ce Added reference token attention isolation (kv_cache) for Krea2 edit training. Same training cost with significant inference speed up. 2x inference speedup. 2026-07-09 12:00:55 -06:00
Jaret Burkett
6233efe1bb Fix sampling issue on zimage turbo and krea2 turbo models 2026-07-08 12:34:39 -06:00
Jaret Burkett
dd08579eda Add ability to pull control images from same folder group 2026-07-08 05:21:31 -06:00