Text Generation
PEFT
Safetensors
Chinese
English
lora
tool-selection
tool-call
guardrail
chinese
traditional-chinese
fine-tuned
qwen2
conversational
Instructions to use GOSHUNCLE/tool_call_validator_zh with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use GOSHUNCLE/tool_call_validator_zh with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-3B-Instruct") model = PeftModel.from_pretrained(base_model, "GOSHUNCLE/tool_call_validator_zh") - Notebooks
- Google Colab
- Kaggle
Update README.md
Browse files
README.md
CHANGED
|
@@ -22,16 +22,16 @@ tags:
|
|
| 22 |
|
| 23 |
# tool_call_validator_zh
|
| 24 |
|
| 25 |
-
>
|
| 26 |
> Traditional Chinese tool-call validator (guardrail) — LoRA fine-tune of Qwen2.5-3B-Instruct
|
| 27 |
|
| 28 |
-
**🚀 [Try the live demo →](https://huggingface.co/spaces/GOSHUNCLE/tool_call_validator_zh_demo)** ·
|
| 29 |
|
| 30 |
---
|
| 31 |
|
| 32 |
## 中文說明
|
| 33 |
|
| 34 |
-
本模型是針對 **Tool Call Validation
|
| 35 |
|
| 36 |
1. 讀取使用者請求(user prompt)與多個候選工具的 description
|
| 37 |
2. 透過語意比對選出最適合的工具,或在無合適工具時拒絕匹配
|
|
@@ -65,7 +65,7 @@ tags:
|
|
| 65 |
|
| 66 |
### Performance(三層次評估)
|
| 67 |
|
| 68 |
-
|
| 69 |
|
| 70 |
| Metric | L1 base | L2 adapter | L3 +Filter |
|
| 71 |
|---|---:|---:|---:|
|
|
@@ -170,7 +170,7 @@ Invalid 時 fallback:`{signal: "abstain", confidence: "low", selected_tool: nu
|
|
| 170 |
|
| 171 |
#### 限制 C:Reasoning 中文偏正式書面語
|
| 172 |
|
| 173 |
-
訓練樣本 reasoning 風格偏向「翻譯式書面語」
|
| 174 |
|
| 175 |
### Deployment Notes(部署注意事項)
|
| 176 |
|
|
|
|
| 22 |
|
| 23 |
# tool_call_validator_zh
|
| 24 |
|
| 25 |
+
> LoRA fine-tune of Qwen2.5-3B-Instruct
|
| 26 |
> Traditional Chinese tool-call validator (guardrail) — LoRA fine-tune of Qwen2.5-3B-Instruct
|
| 27 |
|
| 28 |
+
**🚀 [Try the live demo →](https://huggingface.co/spaces/GOSHUNCLE/tool_call_validator_zh_demo)** ·
|
| 29 |
|
| 30 |
---
|
| 31 |
|
| 32 |
## 中文說明
|
| 33 |
|
| 34 |
+
本模型是針對 **Tool Call Validation** 場景微調的繁體中文模型。基於 [Qwen/Qwen2.5-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-3B-Instruct) 用 LoRA 訓練,能夠:
|
| 35 |
|
| 36 |
1. 讀取使用者請求(user prompt)與多個候選工具的 description
|
| 37 |
2. 透過語意比對選出最適合的工具,或在無合適工具時拒絕匹配
|
|
|
|
| 65 |
|
| 66 |
### Performance(三層次評估)
|
| 67 |
|
| 68 |
+
三層次評估設計:
|
| 69 |
|
| 70 |
| Metric | L1 base | L2 adapter | L3 +Filter |
|
| 71 |
|---|---:|---:|---:|
|
|
|
|
| 170 |
|
| 171 |
#### 限制 C:Reasoning 中文偏正式書面語
|
| 172 |
|
| 173 |
+
訓練樣本 reasoning 風格偏向「翻譯式書面語」,對極度口語化的輸入可能略顯生硬。
|
| 174 |
|
| 175 |
### Deployment Notes(部署注意事項)
|
| 176 |
|