Update README.md
Browse filesadd merge details
README.md
CHANGED
|
@@ -14,4 +14,36 @@ ECE-TW3-JRGL-V2 is a merge of the following models using [mergekit](https://gith
|
|
| 14 |
* [Qwen/Qwen1.5-72B-Chat](https://huggingface.co/Qwen/Qwen1.5-72B-Chat)
|
| 15 |
* [vilm/Quyen-Pro-Max-v0.1](https://huggingface.co/vilm/Quyen-Pro-Max-v0.1)
|
| 16 |
|
| 17 |
-
## 🧩 Configuration
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
* [Qwen/Qwen1.5-72B-Chat](https://huggingface.co/Qwen/Qwen1.5-72B-Chat)
|
| 15 |
* [vilm/Quyen-Pro-Max-v0.1](https://huggingface.co/vilm/Quyen-Pro-Max-v0.1)
|
| 16 |
|
| 17 |
+
## 🧩 Configuration
|
| 18 |
+
```yml
|
| 19 |
+
base_model: Qwen/Qwen1.5-72B-Chat
|
| 20 |
+
dtype: bfloat16
|
| 21 |
+
merge_method: slerp
|
| 22 |
+
parameters:
|
| 23 |
+
t:
|
| 24 |
+
- filter: self_attn
|
| 25 |
+
value:
|
| 26 |
+
- 0
|
| 27 |
+
- 0.5
|
| 28 |
+
- 0.3
|
| 29 |
+
- 0.7
|
| 30 |
+
- 1
|
| 31 |
+
- filter: mlp
|
| 32 |
+
value:
|
| 33 |
+
- 1
|
| 34 |
+
- 0.5
|
| 35 |
+
- 0.7
|
| 36 |
+
- 0.3
|
| 37 |
+
- 0
|
| 38 |
+
- value: 0.5
|
| 39 |
+
slices:
|
| 40 |
+
- sources:
|
| 41 |
+
- layer_range:
|
| 42 |
+
- 0
|
| 43 |
+
- 80
|
| 44 |
+
model: Qwen/Qwen1.5-72B-Chat
|
| 45 |
+
- layer_range:
|
| 46 |
+
- 0
|
| 47 |
+
- 80
|
| 48 |
+
model: vilm/Quyen-Pro-Max-v0.1
|
| 49 |
+
```
|