2128ac1e08
fixed issue with embed name, save whole config to dir instead of just process so it can be easily shared. Only make one config, no timesteps
Jaret Burkett2023-09-09 12:24:08 -06:00
be804c9cf5
Save embeddings as their trigger to match auto and comfy style loading. Also, FINALLY found why gradients were wonkey and fixed it. The root problem is dropping out of network state before backward pass.
Jaret Burkett2023-09-09 12:02:07 -06:00
3feb663a51
Switched to new bucket system that matched sdxl trained buckets. Fixed requirements. Updated embeddings to work with sdxl. Added method to train lora with an embedding at the trigger. Still testing but works amazingly well from what I can see
Jaret Burkett2023-09-07 13:06:18 -06:00
436bf0c6a3
Added experimental concept replacer, replicate converter, bucket maker, and other goodies
Jaret Burkett2023-09-06 18:50:32 -06:00
33267e117c
Reworked bucket loader to scale buckets to pixels amounts not just minimum size. Makes the network more consistant
Jaret Burkett2023-08-30 14:52:12 -06:00
d401348c2e
Make data loader resiliant to bad headers in meta
Jaret Burkett2023-08-29 18:56:06 -06:00
836fee47a6
Fixed some mismatched weights by adjusting tolerance. The mismatch ironically made the models better lol
Jaret Burkett2023-08-29 15:20:03 -06:00
14ff51ceb4
fixed issues with converting and saving models. Cleaned keys. Improved testing for cycle load saving.
Jaret Burkett2023-08-29 12:31:19 -06:00
714854ee86
Hude rework to move the batch to a DTO to make it far more modular to the future ui
Jaret Burkett2023-08-29 10:22:19 -06:00
a008d9e63b
Fixed issue with loadin models after resume function added. Added additional flush if not training text encoder to clear out vram before grad accum
Jaret Burkett2023-08-28 17:56:30 -06:00
9b164a8688
Fixed issue with bucket dataloader corpping in too much. Added normalization capabilities to LoRA modules. Testing effects, but should prevent them from burning and also make them more compatable with stacking many LoRAs
Jaret Burkett2023-08-27 09:40:01 -06:00
6bd3851058
Fixed issue with prompt token replace adding more than one replacement
Jaret Burkett2023-08-26 18:52:23 -06:00
fd338e67bb
Fixed bug with dataloader not seperating mulitple datasets
Jaret Burkett2023-08-26 18:07:24 -06:00
8105c05c12
Added bucketting capabilities to dataloader. Finally have full planned capability. noice
Jaret Burkett2023-08-26 16:36:32 -06:00
2cb27c3f57
Merge branch 'main' of github.com:ostris/ai-toolkit
Jaret Burkett2023-08-26 08:55:09 -06:00
3367ab6b2c
Moved SD batch processing to a shared method and added it for use in slider training. Still testing if it affects quality over sampling
Jaret Burkett2023-08-26 08:55:00 -06:00
90eedb78bf
Added multiplier jitter, min_snr, ability to choose sdxl encoders to use, shuffle generator, and other fun
Jaret Burkett2023-08-19 05:54:22 -06:00
80e2f4a2a4
Merge branch 'main' of github.com:ostris/ai-toolkit
Jaret Burkett2023-08-18 11:45:00 -06:00
d51c4ca704
Added ability to use two seperate folders for datasets when doing image reference sliders
Jaret Burkett2023-08-18 11:44:33 -06:00
355c80df07
Added ability to use civit ai url ar model name and built a model downloader and cache manager for it
Jaret Burkett2023-08-13 13:09:51 -06:00
383bad958d
Added a way to run as a library by passing job dict
Jaret Burkett2023-08-13 09:54:39 -06:00
196b693cf0
Worked on reference slider script. It is working well currently. Still going to tune it a bit before a writeup though
Jaret Burkett2023-08-12 17:59:24 -06:00
fd95e7b60c
Merge branch 'main' of github.com:ostris/ai-toolkit
Jaret Burkett2023-08-12 05:59:58 -06:00
8c90fa86c6
Complete reqork of how slider training works and optimized it to hell. Can run entire algorythm in 1 batch now with less VRAM consumption than a quarter of it used to take
Jaret Burkett2023-08-05 18:46:08 -06:00
7e4e660663
Added extensions and an example extension that merges models
Jaret Burkett2023-08-04 09:37:24 -06:00
Jaret Burkett
Fundamentum