Sadly I get this error when trying to train a model with it

prepare dataset

prepare accelerator

Using accelerator 0.15.0 or above.

load StableDiffusion checkpoint

Traceback (most recent call last):

File "F:\kohya_ss\train_network.py", line 659, in <module>

train(args)

File "F:\kohya_ss\train_network.py", line 115, in train

text_encoder, vae, unet, = trainutil.load_target_model(args, weight_dtype)

File "F:\kohya_ss\library\train_util.py", line 2027, in load_target_model

text_encoder, vae, unet = model_util.load_models_from_stable_diffusion_checkpoint(args.v2, name_or_path)

File "F:\kohya_ss\library\model_util.py", line 877, in load_models_from_stable_diffusion_checkpoint

converted_unet_checkpoint = convert_ldm_unet_checkpoint(v2, state_dict, unet_config)

File "F:\kohya_ss\library\model_util.py", line 234, in convert_ldm_unet_checkpoint

new_checkpoint["time_embedding.linear_1.weight"] = unet_state_dict["time_embed.0.weight"]

KeyError: 'time_embed.0.weight'

Traceback (most recent call last):

File "C:\Python\lib\runpy.py", line 196, in runmodule_as_main

return runcode(code, main_globals, None,

File "C:\Python\lib\runpy.py", line 86, in runcode

exec(code, run_globals)

File "F:\kohya_ss\venv\Scripts\accelerate.exe\__main__.py", line 7, in <module>

File "C:\Users\User1\kohya_ss\venv\lib\site-packages\accelerate\commands\accelerate_cli.py", line 45, in main

args.func(args)

File "C:\Users\User1\kohya_ss\venv\lib\site-packages\accelerate\commands\launch.py", line 1104, in launch_command

simple_launcher(args)

File "C:\Users\User1\kohya_ss\venv\lib\site-packages\accelerate\commands\launch.py", line 567, in simple_launcher

raise subprocess.CalledProcessError(returncode=process.returncode, cmd=cmd)

subprocess.CalledProcessError: Command '['C:\\Users\\User1\\kohya_ss\\venv\\Scripts\\python.exe', 'train_network.py', '--pretrained_model_name_or_path=F:/stable-diffusion-webui/models/Stable-diffusion/uberRealisticPornMerge_urpmv13296LORA.safetensors', '--train_data_dir=R://STABLEDIFINPUTS//LORA 3//image', '--resolution=768,768', '--output_dir=R://STABLEDIFINPUTS//LORA 3//model', '--logging_dir=R://STABLEDIFINPUTS//LORA 3//log', '--network_alpha=128', '--save_model_as=safetensors', '--network_module=networks.lora', '--text_encoder_lr=5e-5', '--unet_lr=0.0001', '--network_dim=128', '--output_name=ma2', '--lr_scheduler_num_cycles=1', '--learning_rate=0.0001', '--lr_scheduler=constant', '--train_batch_size=1', '--max_train_steps=1482', '--save_every_n_epochs=1', '--mixed_precision=bf16', '--save_precision=bf16', '--seed=1234', '--caption_extension=.txt', '--cache_latents', '--optimizer_type=AdamW8bit', '--max_data_loader_n_workers=1', '--clip_skip=2', '--bucket_reso_steps=64', '--xformers', '--bucket_no_upscale']' returned non-zero exit status 1.

Can you make a Loha version too pls?

it asks me to use " --disable-nan-check" and then the result is just a black image.

is this overall as flexible as merging the models directly where is the catch ? or could we really save GBs of Data and just blend the models live as we wish them to be blended ?

avoiding all the repetitive data structure and waits in each merged model ?

just combining multiple 300 to 400 mb models weighting them side by side in realtime ontop of a base model ?

Could we extract those differences directly out of a existing checkpoint ?

It seems at least partly possible to reverse engineer what models where mixed together looking for specific data points and similar results.

And sometimes its interesting what you find deep down

And sometimes you wonder if that model was listed in the merge recipe you found structures of

Damn they look hot

What's the license for this? The same as URPM?

Can someone share what settings are best for the LoRa?

Like weight, Sampler, cfg, steps, ClipSkip?