Upload README.md with huggingface_hub
Browse files
README.md
ADDED
@@ -0,0 +1,49 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
base_model:
|
3 |
+
- ReadyArt/Forgotten-Safeword-24B
|
4 |
+
- TheDrummer/Cydonia-24B-v2
|
5 |
+
library_name: transformers
|
6 |
+
tags:
|
7 |
+
- mergekit
|
8 |
+
- merge
|
9 |
+
|
10 |
+
---
|
11 |
+
# merge
|
12 |
+
|
13 |
+
This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
|
14 |
+
|
15 |
+
## Merge Details
|
16 |
+
### Merge Method
|
17 |
+
|
18 |
+
This model was merged using the [DARE TIES](https://arxiv.org/abs/2311.03099) merge method using [ReadyArt/Forgotten-Safeword-24B](https://huggingface.co/ReadyArt/Forgotten-Safeword-24B) as a base.
|
19 |
+
|
20 |
+
### Models Merged
|
21 |
+
|
22 |
+
The following models were included in the merge:
|
23 |
+
* [TheDrummer/Cydonia-24B-v2](https://huggingface.co/TheDrummer/Cydonia-24B-v2)
|
24 |
+
|
25 |
+
### Configuration
|
26 |
+
|
27 |
+
The following YAML configuration was used to produce this model:
|
28 |
+
|
29 |
+
```yaml
|
30 |
+
merge_method: dare_ties
|
31 |
+
base_model: ReadyArt/Forgotten-Safeword-24B
|
32 |
+
models:
|
33 |
+
- model: ReadyArt/Forgotten-Safeword-24B
|
34 |
+
parameters:
|
35 |
+
weight: 0.5
|
36 |
+
density: 0.7 # Balanced parameter retention
|
37 |
+
- model: TheDrummer/Cydonia-24B-v2
|
38 |
+
parameters:
|
39 |
+
weight: 0.5
|
40 |
+
density: 0.7 # Mirror density for symmetry
|
41 |
+
parameters:
|
42 |
+
int8_mask: true
|
43 |
+
normalize: true
|
44 |
+
lambda: 1.0 # Full task vector application
|
45 |
+
filter: [] # No layer-specific adjustments
|
46 |
+
dtype: bfloat16
|
47 |
+
tokenizer_source: union
|
48 |
+
|
49 |
+
```
|