Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: llama3
|
| 3 |
+
language:
|
| 4 |
+
- ja
|
| 5 |
+
- en
|
| 6 |
+
tags:
|
| 7 |
+
- code
|
| 8 |
+
- sql
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
### モデルの説明(English explanation is below.)
|
| 12 |
+
|
| 13 |
+
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。
|
| 14 |
+
|
| 15 |
+
gguf版(今後拡充予定)は [こちら](https://huggingface.co/keitokei1994/Llama-3-Umievo-Shizuko-sqlcoder-2x8B-gguf) 。
|
| 16 |
+
|
| 17 |
+
umiyukiさんが公開している[Llama-3-Umievo-itr014-Shizuko-8b](https://huggingface.co/umiyuki/Llama-3-Umievo-itr014-Shizuko-8b) に、SQLデータセットでファインチューニングされた[rdefog/llama-3-sqlcoder-8b](https://huggingface.co/defog/llama-3-sqlcoder-8b)を合わせることで、日本語能力とSQL生成能力を両立させようとしたMoEモデルです。
|
| 18 |
+
|
| 19 |
+
### モデルの詳細
|
| 20 |
+
|
| 21 |
+
- **モデル名**: Llama-3-Umievo-Shizuko-sqlcoder-2x8B
|
| 22 |
+
- **モデルアーキテクチャ**: Mixture of Experts (MoE)
|
| 23 |
+
- **ベースモデル**: rdefog/llama-3-sqlcoder-8b, defog/llama-3-sqlcoder-8b
|
| 24 |
+
- **マージツール**: MergeKit
|
| 25 |
+
|
| 26 |
+
#### 要求スペック
|
| 27 |
+
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。
|
| 28 |
+
|
| 29 |
+
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。
|
| 30 |
+
|
| 31 |
+
- CPU: Ryzen 5 3600
|
| 32 |
+
- GPU: GeForce RTX 3060 12GB
|
| 33 |
+
- RAM: DDR4-3200 96GB
|
| 34 |
+
- OS: Windows 10
|
| 35 |
+
|
| 36 |
+
---
|
| 37 |
+
|
| 38 |
+
### Model Description
|
| 39 |
+
This model is a Mixture of Experts (MoE) language model created using the MergeKit tool.
|
| 40 |
+
The gguf version (planned to be expanded in the future) can be found [here](https://huggingface.co/keitokei1994/Llama-3-Umievo-Shizuko-sqlcoder-2x8B-gguf).
|
| 41 |
+
This MoE model aims to achieve both Japanese language ability and SQL generation capability by combining [Llama-3-Umievo-itr014-Shizuko-8b](https://huggingface.co/umiyuki/Llama-3-Umievo-itr014-Shizuko-8b), released by umiyuki, with [rdefog/llama-3-sqlcoder-8b](https://huggingface.co/defog/llama-3-sqlcoder-8b), which has been fine-tuned on an SQL dataset.
|
| 42 |
+
|
| 43 |
+
### Model Details
|
| 44 |
+
- **Model Name**: Llama-3-Umievo-Shizuko-sqlcoder-2x8B
|
| 45 |
+
- **Model Architecture**: Mixture of Experts (MoE)
|
| 46 |
+
- **Base Models**: rdefog/llama-3-sqlcoder-8b, defog/llama-3-sqlcoder-8b
|
| 47 |
+
- **Merge Tool**: MergeKit
|
| 48 |
+
|
| 49 |
+
#### Required Specifications
|
| 50 |
+
If using the Q4_K_M quantized model, it can be fully loaded on an RTX 3060 12GB.
|
| 51 |
+
The author has created the model using WSL2 and Google Colaboratory Pro, and has tested it using Llama.cpp and LMstudio.
|
| 52 |
+
- CPU: Ryzen 5 3600
|
| 53 |
+
- GPU: GeForce RTX 3060 12GB
|
| 54 |
+
- RAM: DDR4-3200 96GB
|
| 55 |
+
- OS: Windows 10
|