pochunhsu commited on
Commit
db4f493
·
verified ·
1 Parent(s): 9a775e3

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +2 -2
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。此模型基於 [Breeze 2](https://huggingface.co/MediaTek-Research/Llama-Breeze2-8B-Instruct) 骨幹網路,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
 
@@ -22,7 +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:** 聯發科技研究院
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 [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