LoraBlockWeightPlotHelper

https://github.com/zuntan03/lorablockweightplothelper
Admin
LoRAの階層適用を手助けするツールとチュートリアルです。
Tokens:1,657
Snippets:9
Trust Score:7.5
License:MIT
Update:1 year ago
Show doc for...
Tokens:
Raw
### Layer-wise Learning Rate Configuration Example

Source: https://github.com/zuntan03/lorablockweightplothelper/blob/main/README.md

An example configuration for layer-wise learning rates, specifying different weights for down, mid, and up layers. This setting is intended to control the learning process for different parts of the model.

```text
"down_lr_weight=0.6,0.8,1,0.8,0.8,0.6,0.6,0.6,1,0.6,0.6,0.6" "mid_lr_weight=1" "up_lr_weight=0.8,0.6,1,1,1,1,0.8,0.8,0.8,0.6,0.6,0.8"
```

--------------------------------

### Example Merged LoRA Weight Format

Source: https://github.com/zuntan03/lorablockweightplothelper/blob/main/README.md

An example of a specific LoRA weight setting that can be applied and later merged into a LoRA model. This format specifies weights for individual layers, allowing for granular control over the LoRA's effect.

```text
Gobuhime:1,0,0,1,0,0,0,0,1,1,0,0,0,1,0,0,1,1,1,1,0.5,0,0,0,0,0
```

--------------------------------

### Weighted Tag Configuration Example

Source: https://github.com/zuntan03/lorablockweightplothelper/blob/main/README.md

Demonstrates the application of weighted tags for training. Prefixes like '_a', '_k', and '_s' modify the weights of all tags, specific keywords ('gobuhime'), or user-defined tags ('greenskin'), respectively. The following numbers indicate the weight multiplier.

```text
Prefixes: '_a' for all tags, '_k' for 'gobuhime' keyword, '_s' for 'greenskin' tag. Example weights: '05' for 0.5, '15' for 1.5, '30' for 3.0. class_tokens remain unchanged.
```

--------------------------------

### LoRA Prompt Weight Tag Adjustment

Source: https://github.com/zuntan03/lorablockweightplothelper/blob/main/README.md

This example shows how to change the LoRA weight tag in a prompt from 'ALL' to 'Sub' to indicate weight reduction for individual layers. This is a common step when using tools like the LoRA Block Weight Plot Helper.

```text
1girl, gobuhime, longhair, hairornament, choker <lora:gobuhime-Defmix2:1:ALL>
↓
1girl, gobuhime, longhair, hairornament, choker <lora:gobuhime-Defmix2:1:Sub>
```

--------------------------------

### Default LoRA Block Weight Presets

Source: https://github.com/zuntan03/lorablockweightplothelper/blob/main/README.md

A list of default LoRA weight presets provided by the LoRA Block Weight Plot Helper. These presets offer starting points for layer-wise weight adjustments, ranging from no LoRA effect ('NONE') to full effect ('ALL') and various intermediate settings.

```text
NONE:0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
ALL:1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1
INS:1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0
IND:1,0,0,0,1,1,1,0,0,0,0,0,0,0,0,0,0
INALL:1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0
MIDD:1,0,0,0,1,1,1,1,1,1,1,1,0,0,0,0,0
OUTD:1,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0
OUTS:1,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1
OUTALL:1,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1
ALL0.5:0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5
```

--------------------------------

### Effective Block Analyzer Prompt Configuration

Source: https://github.com/zuntan03/lorablockweightplothelper/blob/main/README.md

This configuration is used to run the Effective Block Analyzer. It requires a specific format for the LoRA path, ending with ':XYZ>', and includes various Stable Diffusion parameters. Ensure your LoRA path is correctly formatted for the analyzer to function.

```text
1girl, gobuhime, longhair, hairornament, choker <lora:gobuhime-Defmix2:1:XYZ>
Negative prompt: EasyNegative, (nude, nipples: 1.3)
Steps: 25, Sampler: DPM++ 2M Karras, CFG scale: 8, Seed: -1, Size: 512x512, Model hash: 31c9be0ee8, Model: Defmix-v2.0, Denoising strength: 0.6, Clip skip: 2, Hires upscale: 1.5, Hires steps: 15, Hires upscaler: Latent (nearest-exact)
```

--------------------------------

### SuperMerger LoRA Merging Syntax

Source: https://github.com/zuntan03/lorablockweightplothelper/blob/main/README.md

This demonstrates the syntax used in the SuperMerger extension to specify a LoRA file and its associated weights for merging. It includes the LoRA filename, a ratio, and optionally specific block weights.

```text
LoRAファイル名:1.0:Gobuhime
```

--------------------------------

### Applying Optimized LoRA Weights with X/Y/Z Plot

Source: https://github.com/zuntan03/lorablockweightplothelper/blob/main/README.md

After determining optimized weights using the Plot Helper, they are applied to the prompt by replacing ':XYZ' with ':ALL' and setting up an X/Y/Z plot. This allows for batch generation with the new weights applied across all layers.

```text
1girl, gobuhime, longhair, hairornament, choker <lora:gobuhime-Defmix2:1:ALL>
```

```text
X軸の種類: Prompt S/R
X軸の値: ALL, Gobuhime
```

--------------------------------

### Layer-wise Learning Rate Configuration for Gobuhime

Source: https://github.com/zuntan03/lorablockweightplothelper/blob/main/README.md

A specific layer-wise learning rate configuration for the 'gobuhimeC721-Defmix2.safetensors' model. This configuration adjusts weights across down, mid, and up layers, likely to fine-tune the model's output.

```text
"down_lr_weight=0.1,0.2,0.7,0.2,0.2,0.1,0.1,0.1,0.7,0.1,0.1,0.1" "mid_lr_weight=0.7" "up_lr_weight=0.2,0.1,0.7,0.7,0.7,0.7,0.2,0.2,0.2,0.1,0.1,0.2"
```
Tokens:
Raw