Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,28 @@
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
---
|
4 |
+
This is a test merge of some Gemma-7b finetunes using task_arithmatic. So far its not working, but this is uploaded for testing and diagnosing purposes.
|
5 |
+
|
6 |
+
Merge config:
|
7 |
+
```yaml
|
8 |
+
models:
|
9 |
+
- model: gemma-7b-it-fp16
|
10 |
+
parameters:
|
11 |
+
weight: 1
|
12 |
+
- model: CorticalStack_gemma-7b-ultrachat-sft
|
13 |
+
parameters:
|
14 |
+
weight: 1
|
15 |
+
- model: cloudyu_google-gemma-7b-it-dpo-v1
|
16 |
+
parameters:
|
17 |
+
weight: 1
|
18 |
+
- model: abideen_gemma-7b-openhermes
|
19 |
+
parameters:
|
20 |
+
weight: 1
|
21 |
+
merge_method: task_arithmetic
|
22 |
+
base_model: gemma-7b-base
|
23 |
+
parameters:
|
24 |
+
normalize: true
|
25 |
+
int8_mask: true
|
26 |
+
dtype: float16
|
27 |
+
|
28 |
+
```
|