sd-scripts

sd-scripts copied to clipboard

what's the value for --noise_offset

1

(venv) PS G:\sd-scripts> accelerate launch --num_cpu_threads_per_process 2 train_network.py ` >> --pretrained_model_name_or_path=F:\stable-diffusion-webui\models\Stable-diffusion\dreamshaper_332BakedVaeClipFix.safetensors ` >> --train_data_dir=G:\dls\all\4 ` >> --output_dir=F:\stable-diffusion-webui\models\Lora ` >> --output_name=t1 ` >> --caption_extension=.txt ` >> --prior_loss_weight=1.0 ` >> --resolution=768,768 `...

tcflying

Dadaptation optimizer

1

--use_dadaptation_optimizer args added. with this option, dadaptation is avalilable. while using dadaptation, it recommend 1-digit float numbers( 0.1

BootsofLagrangian

Folder name system to Config File or JSON

3

The folder name system sucks. when we are literally passing so many configs each time why using the folder system. Also in the folder system we can't pass some characters...

vsatyamesc

During training, loss=nan and a broken LoRA is generated

1

Learning LoRA with the following parameters results in loss=nan. The resulting LoRA file is corrupt Is there anything I can do to improve it? xformers and others are recommended ---------------------------------------------------------...

acncagua

LearningRate-Free Learning Algorithm

33

Hi, how about [D-adaptation](https://github.com/facebookresearch/dadaptation)? This is a kind of algorithm that end-user doesn't need to set specific learning rate. In short, D-adaptation use boundedness to find proper learning rate. So,...

BootsofLagrangian

Adding Validation Loss to detect Overtraining

9

I saw that this repo has added the ability to use a validation loss to help figure out the optimal amount of training. Might be an interesting addition. https://github.com/victorchall/EveryDream2trainer/blob/main/doc/VALIDATION.md

Naegles

New optimizer implementation maybe. 3x speed boost.

1

https://github.com/lucidrains/lion-pytorch.git

x-legion

AttributeError: 'Namespace' object has no attribute 'noise_offset'

7

Hi All, Any one knows how to fix this error? Or where can I skip this 'noise_offset' option? `running training / 学習開始 num train images * repeats / 学習画像の数×繰り返し回数: 1000...

yvonnechen888

Why does the 'loss=nan' appear when finetuning model using fp16 or bf16 mixed precision in 'fine_tune.py' ?

2

Same as title. This problem has existed since the v9 version. It is related to the GPU type. It appears in 3090, A40, but doesn't appear in A100.

zakinp

Find Trigger Words from Dreambooth / Lora

2

Some Dreambooth and Lora has Trigger Words and it doesn't activate without them. https://civitai.com/models/4219/one-piece-wano-saga-style-lora (For example, this Lora needs "WANOSTYLE" to activate. But it is hard to remember all trigger...

scor2