56 lines
1.7 KiB
Markdown
56 lines
1.7 KiB
Markdown
|
|
# Documentation
|
|
- Class name: Automatic CFG
|
|
- Category: model_patches/automatic_cfg/presets
|
|
- Output node: False
|
|
|
|
Automatic CFG节点通过基于'boost'标志调整参数来为模型应用动态配置，从而增强其性能。这种调整旨在优化模型的运行，可能会提高效率和输出质量。
|
|
|
|
# Input types
|
|
## Required
|
|
- model
|
|
- model参数代表将应用动态配置的模型。它对于定义进行调整和优化的基础至关重要。
|
|
- Comfy dtype: MODEL
|
|
- Python dtype: torch.nn.Module
|
|
- boost
|
|
- boost标志决定了配置调整的强度。启用boost会应用更激进的优化策略，可能会显著提升模型性能。
|
|
- Comfy dtype: BOOLEAN
|
|
- Python dtype: bool
|
|
|
|
# Output types
|
|
- model
|
|
- 输出是输入模型的修改版本，通过动态配置调整进行了优化，以提高性能。
|
|
- Comfy dtype: MODEL
|
|
- Python dtype: torch.nn.Module
|
|
|
|
|
|
## Usage tips
|
|
- Infra type: `CPU`
|
|
- Common nodes: unknown
|
|
|
|
|
|
## Source code
|
|
```python
|
|
class simpleDynamicCFG:
|
|
@classmethod
|
|
def INPUT_TYPES(s):
|
|
return {"required": {
|
|
"model": ("MODEL",),
|
|
"boost" : ("BOOLEAN", {"default": True}),
|
|
}}
|
|
RETURN_TYPES = ("MODEL",)
|
|
FUNCTION = "patch"
|
|
|
|
CATEGORY = "model_patches/automatic_cfg/presets"
|
|
|
|
def patch(self, model, boost):
|
|
advcfg = advancedDynamicCFG()
|
|
m = advcfg.patch(model,
|
|
skip_uncond = boost,
|
|
uncond_sigma_start = 15, uncond_sigma_end = 1,
|
|
automatic_cfg = "hard" if boost else "soft"
|
|
)[0]
|
|
return (m, )
|
|
|
|
```
|