Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Merge AIDC-ai-business/Marcoroni-7B-v3 and rwitz/go-bruins-v2 using slerp merge from https://github.com/cg123/mergekit.
|
| 2 |
+
After that we trained DPO with HF data
|
| 3 |
+
|
| 4 |
+
*config.yaml*
|
| 5 |
+
```
|
| 6 |
+
slices:
|
| 7 |
+
- sources:
|
| 8 |
+
- model: AIDC-ai-business/Marcoroni-7B-v3
|
| 9 |
+
layer_range: [0, 32]
|
| 10 |
+
- model: rwitz/go-bruins-v2
|
| 11 |
+
layer_range: [0, 32]
|
| 12 |
+
merge_method: slerp
|
| 13 |
+
base_model: AIDC-ai-business/Marcoroni-7B-v3
|
| 14 |
+
parameters:
|
| 15 |
+
t:
|
| 16 |
+
- filter: self_attn
|
| 17 |
+
value: [0, 0.5, 0.3, 0.7, 1]
|
| 18 |
+
- filter: mlp
|
| 19 |
+
value: [1, 0.5, 0.7, 0.3, 0]
|
| 20 |
+
- value: 0.5
|
| 21 |
+
dtype: float16
|
| 22 |
+
```
|
| 23 |
+
|
| 24 |
+
You can use alpaca template.
|
| 25 |
+
```
|
| 26 |
+
template_format = """{system}
|
| 27 |
+
### Instruction:
|
| 28 |
+
{prompt}
|
| 29 |
+
|
| 30 |
+
### Response:
|
| 31 |
+
"""
|
| 32 |
+
```
|