Update README.md
Browse files
README.md
CHANGED
@@ -1,5 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
```
|
2 |
cognitivecomputations/dolphin-2.6-mixtral-8x7b
|
3 |
mistralai/Mixtral-8x7B-v0.1
|
4 |
mistralai/Mixtral-8x7B-Instruct-v0.1
|
5 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
language:
|
4 |
+
- en
|
5 |
+
pipeline_tag: text-generation
|
6 |
+
inference: false
|
7 |
+
tags:
|
8 |
+
- dare
|
9 |
+
- super mario merge
|
10 |
+
- pytorch
|
11 |
+
- mixtral
|
12 |
+
- merge
|
13 |
+
---
|
14 |
+
|
15 |
+
# mixtral megamerge 8x7b v1
|
16 |
+
|
17 |
+
|
18 |
+
The following models were merged with DARE using [https://github.com/martyn/safetensors-merge-supermario](https://github.com/martyn/safetensors-merge-supermario)
|
19 |
+
|
20 |
+
## Mergelist
|
21 |
+
|
22 |
```
|
23 |
cognitivecomputations/dolphin-2.6-mixtral-8x7b
|
24 |
mistralai/Mixtral-8x7B-v0.1
|
25 |
mistralai/Mixtral-8x7B-Instruct-v0.1
|
26 |
```
|
27 |
+
|
28 |
+
|
29 |
+
## Merge command
|
30 |
+
|
31 |
+
```
|
32 |
+
python hf_merge.py mergelist.txt mixtral-merge-1 -p 0.1 -lambda 1.95
|
33 |
+
```
|
34 |
+
|
35 |
+
### Notes
|
36 |
+
|
37 |
+
* seems to generalize instruct styles
|
38 |
+
* MoE gates are not modified
|