Commit Graph

41 Commits

Author SHA1 Message Date
Filip Strand
b4c27a6fab
raise error if layer cannot be loaded
Co-authored-by: Fabio Peruzzo <39661245+Xuzzo@users.noreply.github.com>
2024-09-04 19:43:32 +02:00
elitexp
0fd03c7552
@Xuzzo - Weight Adjustment as Original Transformer
Co-authored-by: Fabio Peruzzo <39661245+Xuzzo@users.noreply.github.com>
2024-09-04 21:52:51 +05:45
Sujip Maharjan
8d8812675e feat: --apply-lora a.safetensors b.safetensors c.safetensor --lora-scales 0.5 1.0 0.5 2024-09-03 19:36:15 +05:45
Sujip Maharjan
03b929e732 feat: Add --lora-scale parameter to adjust the scaling of LoRA weights 2024-09-03 18:39:01 +05:45
elitexp
4dfb3aa892
Clean up weight handler
@xuzzo Cleanup code

Co-authored-by: Fabio Peruzzo <39661245+Xuzzo@users.noreply.github.com>
2024-09-03 18:19:21 +05:45
Sujip Maharjan
f82caa3298 Icrecream dependency removed, not required -- error 2024-09-03 13:16:35 +05:45
Sujip Maharjan
cd54ad137c Apply lora with --apply-lora <path-to-safetensors> 2024-09-03 07:38:04 +05:45
filipstrand
fdc4cb9892 Rewrite weight handling and add quantization support 2024-08-30 22:13:36 +02:00
filipstrand
1f54b94175 Simplify: Remove latent creator class and inline logic 2024-08-21 00:48:08 +02:00
filipstrand
45104ebf48 Fix width property 2024-08-20 23:22:25 +02:00
filipstrand
d0970162a5 Fix images in readme 2024-08-20 22:56:57 +02:00
filipstrand
7a277fc326 Rename flux_1_schnell folder to flux_1 2024-08-20 22:37:32 +02:00
filipstrand
4d613c7390 Move num_train_steps to ModelConfig 2024-08-20 11:45:36 +02:00
filipstrand
7795d47cd3 Add a separate runtime config 2024-08-20 10:59:28 +02:00
filipstrand
4b42e50257 Add separate model config and other small updates 2024-08-20 09:44:09 +02:00
Fabio
3710f101e5 small fix and update readme 2024-08-19 21:09:41 +02:00
Fabio
e60cd79f0c make config static and freeze after initialisation 2024-08-19 20:41:11 +02:00
Fabio
16ee26c2e9 fix merge conflicts 2024-08-18 23:33:22 +02:00
Fabio
68cbaab816 fix type and add main_dev 2024-08-18 23:31:54 +02:00
filipstrand
c50a1fb3d4 Fix height and width argument names 2024-08-18 21:49:41 +02:00
Fabio
d14f49f758 revert changes to config 2024-08-18 21:23:52 +02:00
filipstrand
5669d26274 Never override previously generated images 2024-08-18 20:46:11 +02:00
Fabio
acfc448f0c fix guidance scale 2024-08-18 17:56:31 +02:00
Fabio
6977e92061 fix sigmas shift 2024-08-18 17:15:58 +02:00
Fabio
571608b4dd fix conflicts 2024-08-18 13:32:02 +02:00
Fabio
40b40b5541 add shift to schedule 2024-08-18 13:30:19 +02:00
Fabio
14c5b2bd0e fix architecture for dev 2024-08-18 01:01:01 +02:00
Fabio
e76bb3d284 change naming 2024-08-17 22:50:35 +02:00
Fabio
65c479b091 fix height <-> width and other minor corrections 2024-08-17 22:39:52 +02:00
Fabio
7b1eba235d WIP include flux1 dev 2024-08-17 22:33:11 +02:00
Fabio Peruzzo
7661622839 allow different image resolutions 2024-08-16 16:45:49 +02:00
filipstrand
65e0232cab Move flux.py up one level 2024-08-15 21:17:33 +02:00
filipstrand
5af0763cc4 Rearrange attributes 2024-08-15 21:08:27 +02:00
filipstrand
0fae839053 Do text encoding directly and remove TextEncoder class 2024-08-15 20:55:56 +02:00
filipstrand
f0b02257e3 Update README with installation guide 2024-08-15 00:18:48 +02:00
filipstrand
e3a85df133 Automatically download models from huggingface or used cache ones 2024-08-15 00:17:30 +02:00
filipstrand
97028713e1 Update default prompt 2024-08-14 11:12:30 +02:00
filipstrand
37a6632259 Return mlx arrays directly from tokenizer call 2024-08-14 11:02:04 +02:00
filipstrand
5023437e89 Remove loop from main script 2024-08-13 23:14:15 +02:00
filipstrand
1cecc60e5e Add progress bar 2024-08-12 21:37:41 +02:00
filipstrand
83850ddbf1 Initial release 2024-08-12 20:34:17 +02:00