Kohaku-Blueleaf 9cfa68c92f [Experimental Feature] FP8 weight dtype for base model when running train_network (or sdxl_train_network) (#1057) ...

* Add fp8 support

* remove some debug prints

* Better implementation for te

* Fix some misunderstanding

* as same as unet, add explicit convert

* better impl for convert TE to fp8

* fp8 for not only unet

* Better cache TE and TE lr

* match arg name

* Fix with list

* Add timeout settings

* Fix arg style

* Add custom seperator

* Fix typo

* Fix typo again

* Fix dtype error

* Fix gradient problem

* Fix req grad

* fix merge

* Fix merge

* Resolve merge

* arrangement and document

* Resolve merge error

* Add assert for mixed precision

2024-01-20 09:46:53 +09:00

..

ipex

Cleanup

2024-01-02 11:51:29 +03:00

__init__.py

Proposed file structure rework and required file changes

2022-12-18 15:24:46 -05:00

attention_processors.py

update diffusers to 1.16 | train_network

2023-06-01 20:39:26 +09:00

config_util.py

initialize keep_tokens_separator to dataset config

2023-11-28 17:22:35 +07:00

custom_train_functions.py

Fix min-snr-gamma for v-prediction and ZSNR.

2023-11-07 23:02:25 +01:00

huggingface_util.py

fix error in huggingface_path_in_repo=None

2023-08-06 22:08:30 +09:00

hypernetwork.py

update diffusers to 1.16 | train_network

2023-06-01 20:39:26 +09:00

lpw_stable_diffusion.py

Update dependencies ref #1024

2024-01-04 19:53:25 +09:00

model_util.py

Fix an issue with saving as diffusers sd1/2 model close #1033

2024-01-04 21:43:50 +09:00

original_unet.py

Fix typo --spda (it's --sdpa)

2024-01-16 14:32:48 +02:00

sai_model_spec.py

fix crashing in saving lora with clipskip

2023-08-08 21:03:16 +09:00

sdxl_lpw_stable_diffusion.py

Add workaround for clip's bug for pooled output

2023-08-04 08:38:27 +09:00

sdxl_model_util.py

Update dependencies ref #1024

2024-01-04 19:53:25 +09:00

sdxl_original_unet.py

make separate U-Net for inference

2023-11-26 18:11:30 +09:00

sdxl_train_util.py

Fix gradients synchronization for multi-GPUs training (#989)

2023-12-07 22:01:42 +09:00

slicing_vae.py

make slicing vae compatible with latest diffusers

2023-11-26 18:12:04 +09:00

train_util.py

[Experimental Feature] FP8 weight dtype for base model when running train_network (or sdxl_train_network) (#1057)

2024-01-20 09:46:53 +09:00

utils.py

small fix

2023-04-02 23:21:17 +09:00