File size: 2,711 Bytes
c9a531d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
55
56
57
58
59
60
61
62
63
64
65
66
67
68
---
base_model: []
library_name: transformers
tags:
- mergekit
- merge

---

# EXL2 quants of [ryzen88/Llama-3-70b-Arimas-story-RP-V2.1](https://huggingface.co./ryzen88/Llama-3-70b-Arimas-story-RP-V2.1)

[2.50 bits per weight](https://huggingface.co./kim512/Llama-3-70b-Arimas-story-RP-V2.1-2.5bpw-h6-exl2)  
[2.75 bits per weight](https://huggingface.co./kim512/Llama-3-70b-Arimas-story-RP-V2.1-2.75bpw-h6-exl2)  
[3.00 bits per weight](https://huggingface.co./kim512/Llama-3-70b-Arimas-story-RP-V2.1-3.0bpw-h6-exl2)  
[3.25 bits per weight](https://huggingface.co./kim512/Llama-3-70b-Arimas-story-RP-V2.1-3.25bpw-h6-exl2)  
[3.50 bits per weight](https://huggingface.co./kim512/Llama-3-70b-Arimas-story-RP-V2.1-3.5bpw-h6-exl2)  
[4.00 bits per weight](https://huggingface.co./kim512/Llama-3-70b-Arimas-story-RP-V2.1-4.0bpw-h6-exl2)  
[4.50 bits per weight](https://huggingface.co./kim512/Llama-3-70b-Arimas-story-RP-V2.1-4.5bpw-h6-exl2)  
[5.00 bits per weight](https://huggingface.co./kim512/Llama-3-70b-Arimas-story-RP-V2.1-5.0bpw-h6-exl2)  
[6.00 bits per weight](https://huggingface.co./kim512/Llama-3-70b-Arimas-story-RP-V2.1-6.0bpw-h6-exl2)  
[8.00 bits per weight](https://huggingface.co./kim512/Llama-3-70b-Arimas-story-RP-V2.1-8.0bpw-h8-exl2)  

Created using the defaults from [exllamav2 0.1.6](https://github.com/turboderp/exllamav2/releases/tag/v0.1.6) convert.py  
2.5bpw to 6.0bpw head bits = 6  
8.0bpw head bits = 8  
length = 8192  
dataset rows = 200  
measurement rows = 32  
measurement length = 8192  

# Merge_XL_model_Stock
This model switches to the Smaug instruct 32K for the base bodel.
Expanded with Giraffe and Gradient to keep a robuust long context window.
Higgs and cat for most of the story and RP aspects.
Hermes and Chinese chat are for overall intelligence and understanding.

## Merge Details
### Merge Method

This model was merged using the [Model Stock](https://arxiv.org/abs/2403.19522) merge method using \Smaug-Llama-3-70B-Instruct-32K as a base.

### Models Merged

The following models were included in the merge:
* \Llama-3-Giraffe-70B-Instruct
* \Llama-3-70B-Instruct-Gradient-262k
* \Hermes-2-Theta-Llama-3-70B
* \Higgs-Llama-3-70B
* \Llama3-70B-Chinese-Chat
* \Meta-LLama-3-Cat-A-LLama-70b

### Configuration

The following YAML configuration was used to produce this model:

```yaml
models:
    - model: \Smaug-Llama-3-70B-Instruct-32K
    - model: \Llama-3-70B-Instruct-Gradient-262k
    - model: \Llama-3-Giraffe-70B-Instruct
    - model: \Higgs-Llama-3-70B
    - model: \Llama3-70B-Chinese-Chat
    - model: \Meta-LLama-3-Cat-A-LLama-70b
    - model: \Hermes-2-Theta-Llama-3-70B
merge_method: model_stock
base_model: \Smaug-Llama-3-70B-Instruct-32K
dtype: bfloat16
```