Update README.md
Browse files
README.md
CHANGED
@@ -1,5 +1,132 @@
|
|
1 |
-
---
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
base_model:
|
3 |
+
- maum-ai/Llama-3-MAAL-8B-Instruct-v0.1
|
4 |
+
- beomi/Llama-3-KoEn-8B-Instruct-preview
|
5 |
+
- asiansoul/Llama-3-Open-Ko-Linear-8B
|
6 |
+
- NousResearch/Meta-Llama-3-8B
|
7 |
+
- NousResearch/Meta-Llama-3-8B-Instruct
|
8 |
+
- ajibawa-2023/Code-Llama-3-8B
|
9 |
+
- defog/llama-3-sqlcoder-8b
|
10 |
+
- NousResearch/Hermes-2-Pro-Llama-3-8B
|
11 |
+
- Locutusque/llama-3-neural-chat-v2.2-8B
|
12 |
+
- asiansoul/Joah-Llama-3-KoEn-8B-Coder-v1
|
13 |
+
library_name: transformers
|
14 |
+
tags:
|
15 |
+
- mergekit
|
16 |
+
- merge
|
17 |
+
|
18 |
+
---
|
19 |
+
# Joah-Llama-3-KoEn-8B-Coder-v2
|
20 |
+
|
21 |
+
<a href="https://ibb.co/2Srsmn7"><img src="https://i.ibb.co/f9WnB1Y/Screenshot-2024-05-11-at-7-15-42-PM.png" alt="Screenshot-2024-05-11-at-7-15-42-PM" border="0"></a>
|
22 |
+
|
23 |
+
μ€λ λΆν° μλ‘μκ² λΉμ΄ λμ΄ μ€ μ¬λ¬λΆμ Merge Model
|
24 |
+
|
25 |
+
"μ’μ(Joah)" by AsianSoul
|
26 |
+
|
27 |
+
Soon Multi Language Model Merge based on this. First German Start (Korean / English / German) π
|
28 |
+
|
29 |
+
Where to use Joah : Medical, Korean, English, Translation, Code, Science... π₯
|
30 |
+
|
31 |
+
Strengthened SQL code & Other Sci compared to V1
|
32 |
+
|
33 |
+
## π‘ Merge Details
|
34 |
+
|
35 |
+
|
36 |
+
The performance of this merge model doesn't seem to be bad though.-> Just opinion ^^ ποΈ
|
37 |
+
|
38 |
+
This may not be a model that satisfies you. But if we continue to overcome our shortcomings,
|
39 |
+
|
40 |
+
Won't we someday find the answer we want?
|
41 |
+
|
42 |
+
Don't worry even if you don't get the results you want.
|
43 |
+
|
44 |
+
I'll find the answer for you.
|
45 |
+
|
46 |
+
Soon real PoSE to extend Llama's context length to 64k with using my merge method : [reborn](https://medium.com/@puffanddmx82/reborn-elevating-model-adaptation-with-merging-for-superior-nlp-performance-f604e8e307b2)
|
47 |
+
|
48 |
+
I have found that most of merge's model outside so far do not actually have 64k in their configs. I will improve it in the next merge with my reborn. If that doesn't work, I guess I'll have to find another way, right?
|
49 |
+
|
50 |
+
256k is not possible. My computer is running out of memory.
|
51 |
+
|
52 |
+
If you support me, i will try it on a computer with maximum specifications, also, i would like to conduct great tests by building a network with high-capacity traffic and high-speed 10G speeds for you.
|
53 |
+
|
54 |
+
### Merge Method
|
55 |
+
|
56 |
+
This model was merged using the [DARE](https://arxiv.org/abs/2311.03099) [TIES](https://arxiv.org/abs/2306.01708) merge method using [NousResearch/Meta-Llama-3-8B](https://huggingface.co/NousResearch/Meta-Llama-3-8B) as a base.
|
57 |
+
|
58 |
+
### Models Merged
|
59 |
+
|
60 |
+
The following models were included in the merge:
|
61 |
+
* [maum-ai/Llama-3-MAAL-8B-Instruct-v0.1](https://huggingface.co/maum-ai/Llama-3-MAAL-8B-Instruct-v0.1)
|
62 |
+
* [beomi/Llama-3-KoEn-8B-Instruct-preview](https://huggingface.co/beomi/Llama-3-KoEn-8B-Instruct-preview)
|
63 |
+
* [asiansoul/Llama-3-Open-Ko-Linear-8B](https://huggingface.co/asiansoul/Llama-3-Open-Ko-Linear-8B)
|
64 |
+
* [NousResearch/Meta-Llama-3-8B-Instruct](https://huggingface.co/NousResearch/Meta-Llama-3-8B-Instruct)
|
65 |
+
* [ajibawa-2023/Code-Llama-3-8B](https://huggingface.co/ajibawa-2023/Code-Llama-3-8B)
|
66 |
+
* [defog/llama-3-sqlcoder-8b](https://huggingface.co/defog/llama-3-sqlcoder-8b)
|
67 |
+
* [NousResearch/Hermes-2-Pro-Llama-3-8B](https://huggingface.co/NousResearch/Hermes-2-Pro-Llama-3-8B)
|
68 |
+
* [Locutusque/llama-3-neural-chat-v2.2-8B](https://huggingface.co/Locutusque/llama-3-neural-chat-v2.2-8B)
|
69 |
+
* [asiansoul/Joah-Llama-3-KoEn-8B-Coder-v1](https://huggingface.co/asiansoul/Joah-Llama-3-KoEn-8B-Coder-v1)
|
70 |
+
|
71 |
+
### Configuration
|
72 |
+
|
73 |
+
The following YAML configuration was used to produce this model:
|
74 |
+
|
75 |
+
```yaml
|
76 |
+
models:
|
77 |
+
- model: NousResearch/Meta-Llama-3-8B
|
78 |
+
# Base model providing a general foundation without specific parameters
|
79 |
+
|
80 |
+
- model: NousResearch/Meta-Llama-3-8B-Instruct
|
81 |
+
parameters:
|
82 |
+
density: 0.60
|
83 |
+
weight: 0.25
|
84 |
+
|
85 |
+
- model: beomi/Llama-3-KoEn-8B-Instruct-preview
|
86 |
+
parameters:
|
87 |
+
density: 0.55
|
88 |
+
weight: 0.15
|
89 |
+
|
90 |
+
- model: asiansoul/Llama-3-Open-Ko-Linear-8B
|
91 |
+
parameters:
|
92 |
+
density: 0.55
|
93 |
+
weight: 0.1
|
94 |
+
|
95 |
+
- model: maum-ai/Llama-3-MAAL-8B-Instruct-v0.1
|
96 |
+
parameters:
|
97 |
+
density: 0.55
|
98 |
+
weight: 0.1
|
99 |
+
|
100 |
+
- model: asiansoul/Joah-Llama-3-KoEn-8B-Coder-v1
|
101 |
+
parameters:
|
102 |
+
density: 0.55
|
103 |
+
weight: 0.2
|
104 |
+
|
105 |
+
- model: ajibawa-2023/Code-Llama-3-8B
|
106 |
+
parameters:
|
107 |
+
density: 0.55
|
108 |
+
weight: 0.05
|
109 |
+
|
110 |
+
- model: defog/llama-3-sqlcoder-8b
|
111 |
+
parameters:
|
112 |
+
density: 0.55
|
113 |
+
weight: 0.1
|
114 |
+
|
115 |
+
- model: Locutusque/llama-3-neural-chat-v2.2-8B
|
116 |
+
parameters:
|
117 |
+
density: 0.55
|
118 |
+
weight: 0.1
|
119 |
+
|
120 |
+
- model: NousResearch/Hermes-2-Pro-Llama-3-8B
|
121 |
+
parameters:
|
122 |
+
density: 0.55
|
123 |
+
weight: 0.05
|
124 |
+
|
125 |
+
merge_method: dare_ties
|
126 |
+
base_model: NousResearch/Meta-Llama-3-8B
|
127 |
+
parameters:
|
128 |
+
int8_mask: true
|
129 |
+
dtype: bfloat16
|
130 |
+
|
131 |
+
|
132 |
+
```
|