|
--- |
|
base_model: |
|
- Sao10K/Fimbulvetr-10.7B-v1 |
|
- saishf/Kuro-Lotus-10.7B |
|
library_name: transformers |
|
tags: |
|
- mergekit |
|
- merge |
|
license: cc-by-nc-4.0 |
|
--- |
|
This model is a merge of my personal favourite models, i couldn't decide between them so why not have both? Without MOE cause gpu poor :3 |
|
|
|
With my own tests it gives kuro-lotus like results without the requirement for a highly detailed character card and stays coherent when roping up to 8K context. |
|
|
|
I personally use the "Universal Light" preset in silly tavern, with "alpaca" the results can be short but are longer with "alpaca roleplay". |
|
|
|
"Universal Light" preset can be extremely creative but sometimes likes to act for user with some cards, for those i like just the "default" but any preset seems to work! |
|
|
|
Benchmarks: |
|
Average 72.73 |
|
|
|
ARC 69.54 |
|
|
|
HellaSwag 87.87 |
|
|
|
MMLU 66.99 |
|
|
|
TruthfulQA 60.95 |
|
|
|
Winogrande 84.14 |
|
|
|
GSM8K 66.87 |
|
|
|
# merge |
|
|
|
This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit). |
|
|
|
## Merge Details |
|
### Merge Method |
|
|
|
This model was merged using the SLERP merge method. |
|
|
|
### Models Merged |
|
|
|
The following models were included in the merge: |
|
* [Sao10K/Fimbulvetr-10.7B-v1](https://huggingface.co/Sao10K/Fimbulvetr-10.7B-v1) |
|
* [saishf/Kuro-Lotus-10.7B](https://huggingface.co/saishf/Kuro-Lotus-10.7B) |
|
|
|
### Configuration |
|
|
|
The following YAML configuration was used to produce this model: |
|
|
|
```yaml |
|
slices: |
|
- sources: |
|
- model: saishf/Kuro-Lotus-10.7B |
|
layer_range: [0, 48] |
|
- model: Sao10K/Fimbulvetr-10.7B-v1 |
|
layer_range: [0, 48] |
|
merge_method: slerp |
|
base_model: saishf/Kuro-Lotus-10.7B |
|
parameters: |
|
t: |
|
- filter: self_attn |
|
value: [0.6, 0.7, 0.8, 0.9, 1] |
|
- filter: mlp |
|
value: [0.4, 0.3, 0.2, 0.1, 0] |
|
- value: 0.5 |
|
dtype: bfloat16 |
|
|
|
``` |
