Update README.md
Browse files
README.md
CHANGED
|
@@ -15,3 +15,50 @@ tags:
|
|
| 15 |
Combine the three methods of **della**, **ties**, and **model stock** to merge the instruction model with the base model.
|
| 16 |
|
| 17 |
The aim is to solve the problems of **the decline in instruction-following ability** and **mathematical ability** caused by using only the ties merging method or only the della merging method.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
Combine the three methods of **della**, **ties**, and **model stock** to merge the instruction model with the base model.
|
| 16 |
|
| 17 |
The aim is to solve the problems of **the decline in instruction-following ability** and **mathematical ability** caused by using only the ties merging method or only the della merging method.
|
| 18 |
+
|
| 19 |
+
```yaml
|
| 20 |
+
models:
|
| 21 |
+
- model: Qwen/Qwen2.5-14B-instruct
|
| 22 |
+
parameters:
|
| 23 |
+
density: 1
|
| 24 |
+
weight: 1
|
| 25 |
+
lambda: 0.9
|
| 26 |
+
merge_method: della
|
| 27 |
+
base_model: Qwen/Qwen2.5-14B
|
| 28 |
+
parameters:
|
| 29 |
+
density: 1
|
| 30 |
+
weight: 1
|
| 31 |
+
lambda: 0.9
|
| 32 |
+
normalize: true
|
| 33 |
+
int8_mask: true
|
| 34 |
+
dtype: bfloat16
|
| 35 |
+
name: Qwen2.5-14B-della
|
| 36 |
+
```
|
| 37 |
+
```yaml
|
| 38 |
+
models:
|
| 39 |
+
- model: Qwen/Qwen2.5-14B-instruct
|
| 40 |
+
parameters:
|
| 41 |
+
density: 1
|
| 42 |
+
weight: 1
|
| 43 |
+
merge_method: ties
|
| 44 |
+
base_model: Qwen/Qwen2.5-14B
|
| 45 |
+
parameters:
|
| 46 |
+
density: 1
|
| 47 |
+
weight: 1
|
| 48 |
+
normalize: true
|
| 49 |
+
int8_mask: true
|
| 50 |
+
dtype: bfloat16
|
| 51 |
+
name: Qwen2.5-14B-ties
|
| 52 |
+
```
|
| 53 |
+
```yaml
|
| 54 |
+
merge_method: model_stock
|
| 55 |
+
base_model: Qwen/Qwen2.5-14B-instruct
|
| 56 |
+
models:
|
| 57 |
+
- model: Qwen2.5-14B-della
|
| 58 |
+
- model: Qwen2.5-14B-ties
|
| 59 |
+
dtype: bfloat16
|
| 60 |
+
tokenizer_source: base
|
| 61 |
+
int8_mask: true
|
| 62 |
+
normalize: true
|
| 63 |
+
name: Qwen2.5-14B-it-restore
|
| 64 |
+
```
|