File size: 1,723 Bytes
8501a37 d1a9083 4cfabbf 8501a37 d1a9083 8501a37 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 |
---
base_model:
- mistralai/Mixtral-8x7B-Instruct-v0.1
- NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO
tags:
- mergekit
- merge
---
# lesser-hermes
## Why?
Hermes is very, um, Hermes-y. I wanted to dilute it so I could use it as an ingredient
for other things. Sampling Hermes is a pain in the ass, it either sounds super model-esque
or it loses all instructability. Hence, dilution back to the root.
This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit). We've been using this as one of the experimental ingredients to help stabilize the monkey-typewriter merges, and it's kinda okay at that.
Note that modern mergekit handles MoE just fine, now. But back in the day it did a horrible job and only the fork worked properly.
## Merge Details
### Merge Method
This model was merged using the [DARE](https://arxiv.org/abs/2311.03099) [TIES](https://arxiv.org/abs/2306.01708) merge method using [mistralai/Mixtral-8x7B-Instruct-v0.1](https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1) as a base.
### Models Merged
The following models were included in the merge:
* [NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO](https://huggingface.co/NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO)
### Configuration
The following YAML configuration was used to produce this model:
```yaml
models:
# dont bagel me bro
- model: NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO
parameters:
density: 0.25
weight: 0.3
- model: mistralai/Mixtral-8x7B-Instruct-v0.1
parameters:
density: 0.5
weight: 1
merge_method: dare_ties
base_model: mistralai/Mixtral-8x7B-Instruct-v0.1
parameters:
#normalize: false
#int8_mask: true
dtype: bfloat16
```
|