Update README.md
Browse files
README.md
CHANGED
|
@@ -12,7 +12,7 @@ tags:
|
|
| 12 |
|
| 13 |
[GitHub](https://github.com/mtkresearch/TS-Bench.git) | [Paper](https://arxiv.org/abs/2603.07286)
|
| 14 |
|
| 15 |
-
**Breeze Guard 26** 是一個 80 億參數的台灣華語安全分類器,專門用於偵測使用者輸入中的有害內容。此模型
|
| 16 |
|
| 17 |
**Breeze Guard 26** is an 8B safety classifier for Taiwanese Mandarin, designed to detect harmful content in user prompts. Built on the [Breeze 2](https://huggingface.co/MediaTek-Research/Llama-Breeze2-8B-Instruct) backbone and fine-tuned on 12,000 human-verified samples targeting Taiwan-specific safety risks.
|
| 18 |
|
|
@@ -22,7 +22,7 @@ tags:
|
|
| 22 |
- **基礎模型:** Breeze 2 8B Instruct
|
| 23 |
- **語言:** 台灣華語(繁體中文),並支援基本英文
|
| 24 |
- **授權:** apache-2.0
|
| 25 |
-
- **開發者:** 聯發
|
| 26 |
|
| 27 |
- **Model Type:** Safety classifier (prompt-level harmfulness detection)
|
| 28 |
- **Base Model:** Breeze 2 8B Instruct
|
|
|
|
| 12 |
|
| 13 |
[GitHub](https://github.com/mtkresearch/TS-Bench.git) | [Paper](https://arxiv.org/abs/2603.07286)
|
| 14 |
|
| 15 |
+
**Breeze Guard 26** 是一個 80 億參數的台灣華語安全分類器,專門用於偵測使用者輸入中的有害內容。此模型以 [Breeze 2](https://huggingface.co/MediaTek-Research/Llama-Breeze2-8B-Instruct) 為基底,並使用 12,000 筆經人工驗證、針對台灣特定安全風險的資料進行微調。
|
| 16 |
|
| 17 |
**Breeze Guard 26** is an 8B safety classifier for Taiwanese Mandarin, designed to detect harmful content in user prompts. Built on the [Breeze 2](https://huggingface.co/MediaTek-Research/Llama-Breeze2-8B-Instruct) backbone and fine-tuned on 12,000 human-verified samples targeting Taiwan-specific safety risks.
|
| 18 |
|
|
|
|
| 22 |
- **基礎模型:** Breeze 2 8B Instruct
|
| 23 |
- **語言:** 台灣華語(繁體中文),並支援基本英文
|
| 24 |
- **授權:** apache-2.0
|
| 25 |
+
- **開發者:** 聯發創新基地
|
| 26 |
|
| 27 |
- **Model Type:** Safety classifier (prompt-level harmfulness detection)
|
| 28 |
- **Base Model:** Breeze 2 8B Instruct
|