Commit
•
7455bfa
1
Parent(s):
36c66a0
Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,70 @@
|
|
1 |
-
---
|
2 |
-
license: llama3
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: llama3
|
3 |
+
language:
|
4 |
+
- ja
|
5 |
+
- en
|
6 |
+
---
|
7 |
+
|
8 |
+
### モデルの説明(English explanation is below.)
|
9 |
+
|
10 |
+
このモデルは、MergeKitツールを使用して作成されたMixture of Experts (MoE) 言語モデルです。
|
11 |
+
|
12 |
+
gguf版(今後拡充予定)は [こちら](https://huggingface.co/keitokei1994/Llama-3-youko-chatvector-2x8B_v0.1-gguf) 。
|
13 |
+
|
14 |
+
元のmeta-llama/Meta-Llama-3-8B-Instructに、日本語データセットで継続事前学習されたrinna/llama-3-youko-8bを合わせることで、元のMeta-Llama-3-8B-Instructの能力を維持したまま、日本語能力を向上させようとしたMoEモデルです。
|
15 |
+
|
16 |
+
[Sdff-Ltba/LightChatAssistant-2x7B](https://huggingface.co/Sdff-Ltba/LightChatAssistant-2x7B)を参考に、以下のようなChatVector加算を行ったllama-3-youko-8bをMoEに用いています。
|
17 |
+
|
18 |
+
> rinna/llama-3-youko-8b + 0.8*(meta-llama/Meta-Llama-3-8B-Instruct - meta-llama/Meta-Llama-3-8B)
|
19 |
+
|
20 |
+
このChatVector加算操作をすでに行った状態のllama-3-youko-8bモデルが[aixsatoshi/Llama-3-youko-8b-instruct-chatvector](https://huggingface.co/aixsatoshi/Llama-3-youko-8b-instruct-chatvector)でアップロードされていたため、今回はこちらを利用させていただいております。
|
21 |
+
|
22 |
+
### モデルの詳細
|
23 |
+
|
24 |
+
- **モデル名**: Llama-3-youko-chatvector-2x8B_v0.1
|
25 |
+
- **モデルアーキテクチャ**: Mixture of Experts (MoE)
|
26 |
+
- **ベースモデル**: meta-llama/Meta-Llama-3-8B-Instruct, rinna/llama-3-youko-8b
|
27 |
+
- **マージツール**: MergeKit
|
28 |
+
|
29 |
+
#### 要求スペック
|
30 |
+
Q4_K_M量子化モデルであれば、RTX3060 12GBでフルロード可能です。
|
31 |
+
|
32 |
+
筆者はWSL2やGoogle Colaboratotry Proでの作成後、Llama.cppとLMstudioにて動作確認を行なっています。
|
33 |
+
|
34 |
+
- CPU: Ryzen 5 3600
|
35 |
+
- GPU: GeForce RTX 3060 12GB
|
36 |
+
- RAM: DDR4-3200 96GB
|
37 |
+
- OS: Windows 10
|
38 |
+
|
39 |
+
---
|
40 |
+
|
41 |
+
### Model Description
|
42 |
+
|
43 |
+
This model is a Mixture of Experts (MoE) language model created using the MergeKit tool.
|
44 |
+
|
45 |
+
The gguf version can be found [here](https://huggingface.co/keitokei1994/Llama-3-youko-chatvector-2x8B_v0.1-gguf).
|
46 |
+
|
47 |
+
By combining rinna/llama-3-youko-8b, which has been further pre-trained on Japanese datasets, with the original meta-llama/Meta-Llama-3-8B-Instruct, this MoE model aims to improve Japanese language capabilities while maintaining the abilities of the original Meta-Llama-3-8B-Instruct.
|
48 |
+
|
49 |
+
Referring to [Sdff-Ltba/LightChatAssistant-2x7B](https://huggingface.co/Sdff-Ltba/LightChatAssistant-2x7B), the llama-3-youko-8b used in the MoE has undergone the following ChatVector addition:
|
50 |
+
|
51 |
+
> rinna/llama-3-youko-8b + 0.8*(meta-llama/Meta-Llama-3-8B-Instruct - meta-llama/Meta-Llama-3-8B)
|
52 |
+
|
53 |
+
The llama-3-youko-8b model with this ChatVector addition operation already performed has been uploaded at [aixsatoshi/Llama-3-youko-8b-instruct-chatvector](https://huggingface.co/aixsatoshi/Llama-3-youko-8b-instruct-chatvector), which we have utilized in this case.
|
54 |
+
|
55 |
+
### Model Details
|
56 |
+
|
57 |
+
- **Model Name**: Llama-3-youko-chatvector-2x8B_v0.1
|
58 |
+
- **Model Architecture**: Mixture of Experts (MoE)
|
59 |
+
- **Base Models**: meta-llama/Meta-Llama-3-8B-Instruct, rinna/llama-3-youko-8b
|
60 |
+
- **Merging Tool**: MergeKit
|
61 |
+
|
62 |
+
#### Required Specifications
|
63 |
+
With the Q4_K_M quantized model, it can be fully loaded on an RTX 3060 12GB.
|
64 |
+
|
65 |
+
The author has created the model on WSL2 and Google Colaboratory Pro, and has verified its operation using Llama.cpp and LMstudio.
|
66 |
+
|
67 |
+
- CPU: Ryzen 5 3600
|
68 |
+
- GPU: GeForce RTX 3060 12GB
|
69 |
+
- RAM: DDR4-3200 96GB
|
70 |
+
- OS: Windows 10
|