87 lines
3.3 KiB
Markdown
87 lines
3.3 KiB
Markdown
# 2-Model Merger Documentation
|
||
|
||
---
|
||
|
||
## Weight-Sum
|
||
> `A * (1 - α) + B * α`. A simple linear interpolation between two models.
|
||
|
||
**Models Used:** A, B
|
||
**Parameters:**
|
||
- **Alpha:** Interpolation weight. `0.0` is 100% Model A, `1.0` is 100% Model B.
|
||
|
||
---
|
||
|
||
## Comparative-Interpolation
|
||
> Interpolates between A and B based on the relative differences in their tensor values. Creates a unique blend by deciding how much of each model to use for each individual weight.
|
||
|
||
**Models Used:** A, B
|
||
**Parameters:**
|
||
- **Alpha:** Controls the interpolation curve's strength and shape.
|
||
- **Beta:** Switches the interpolation style. `0.0` focuses on similarity, `1.0` focuses on difference.
|
||
- **Gamma:** Mixes between a randomized (binomial) interpolation at `0.0` and a smooth (linear) one at `1.0`.
|
||
|
||
---
|
||
|
||
## Power-Up (DARE)
|
||
> Adds the unique capabilities of Model B to Model A using the Drop and Rescale (DARE) technique, which often preserves the knowledge of the base model better than simple additions.
|
||
|
||
**Models Used:** A, B
|
||
**Parameters:**
|
||
- **Alpha:** The dropout rate. This is the proportion of unique weights from Model B that are *dropped* before merging. A higher value means less of B is merged.
|
||
- **Beta:** A final multiplier for the rescaled difference before it's added to Model A.
|
||
|
||
---
|
||
|
||
## SVD LoRA Extraction
|
||
> Extracts the difference between Model B and Model A and saves it as a new LoRA file. **This mode outputs a LoRA, not a full model.** The output file will be saved to your `loras` directory.
|
||
|
||
**Models Used:** A (Tuned Model), B (Base Model)
|
||
**Parameters:**
|
||
- **Alpha:** The Rank (dimension) for the LoRA's standard layers.
|
||
- **Beta:** The Rank (dimension) specifically for 3x3 convolution layers.
|
||
- **Gamma:** The clamp quantile for weight values. `0.99` is a typical value to prevent outlier weights from dominating.
|
||
|
||
---
|
||
|
||
## Layer Mismatch Handling
|
||
|
||
When merging models with different layer structures (e.g., fine-tuned models with added/removed layers, or LoRAs with partial layer coverage), the `mismatch_mode` parameter controls behavior:
|
||
|
||
| Mode | Behavior |
|
||
|------|----------|
|
||
| `skip` | Missing layers in B use A's values **(default)** |
|
||
| `zeros` | Missing layers in B are treated as zeros |
|
||
| `error` | Fail if any layer is missing |
|
||
|
||
**Common scenarios:**
|
||
- Merging a fine-tuned model with added/removed layers
|
||
- Combining LoRA-extracted differences with base models
|
||
- Cross-architecture experiments
|
||
|
||
**Note:** Extra layers in Model B that don't exist in Model A are currently ignored. The output will always have the same layer structure as Model A.
|
||
|
||
---
|
||
|
||
## Layer Filtering (Regex Patterns)
|
||
|
||
Use regex patterns to control which layers are merged:
|
||
|
||
### `exclude_patterns`
|
||
Layers matching any pattern will **keep Model A's values only** (no merge).
|
||
|
||
### `discard_patterns`
|
||
Layers matching any pattern will be **removed entirely** from the output.
|
||
|
||
**Pattern format:**
|
||
- Whitespace-separated regex patterns
|
||
- Patterns use **substring matching** (not full match)
|
||
- Example: `text_model lora` matches any key containing "text_model" OR "lora"
|
||
- Example: `layer\.[0-5]\.` matches layers 0-5 using regex syntax
|
||
|
||
**Pattern Examples:**
|
||
| Pattern | Matches |
|
||
|---------|---------|
|
||
| `text_model` | All text encoder layers |
|
||
| `\.norm` | All normalization layers |
|
||
| `attn\.(q\|k\|v)` | Query, key, value attention weights |
|
||
| `block\.[0-9]\.` | Blocks 0-9 | |