Update README.md
Browse files
README.md
CHANGED
|
@@ -38,17 +38,17 @@ and directly outputs a **structured JSON** containing a professional risk evalua
|
|
| 38 |
|
| 39 |
Key Capabilities
|
| 40 |
|
| 41 |
-
Accurately detects subtle and indirect expressions of psychological distress common in Chinese (e.g., “活着没意思”、“快受不了了”、“不如解脱”)
|
| 42 |
-
Distinguishes risk levels from mild distress to clear suicidal ideation
|
| 43 |
-
Recommends appropriate assistant strategies, with strong emphasis on escalation and resource provision when suicide risk is present
|
| 44 |
-
Handles both short single-turn inputs and very long multi-turn conversation contexts
|
| 45 |
|
| 46 |
Intended Use
|
| 47 |
|
| 48 |
-
Safety layer in Chinese mental health chatbots or counseling apps
|
| 49 |
-
Automated risk triage for online psychological support platforms
|
| 50 |
-
Early detection of depression and suicidal ideation in user conversations
|
| 51 |
-
Research on mental health AI in Chinese-language environments
|
| 52 |
|
| 53 |
Base Model
|
| 54 |
|
|
@@ -56,9 +56,9 @@ Qwen/Qwen2.5-0.5B-Instructhttps://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct
|
|
| 56 |
|
| 57 |
Fine-tuning Details
|
| 58 |
|
| 59 |
-
Adapter type: LoRA (r=16, alpha=32, targeting q/k/v/o_proj)
|
| 60 |
-
Dataset: Custom high-quality Chinese mental health risk assessment data (single-turn + multi-turn)
|
| 61 |
-
Training objective: Supervised fine-tuning with strict JSON output formatting and EOS enforcement for clean generation
|
| 62 |
|
| 63 |
```python
|
| 64 |
from peft import PeftModel, PeftConfig
|
|
|
|
| 38 |
|
| 39 |
Key Capabilities
|
| 40 |
|
| 41 |
+
- Accurately detects subtle and indirect expressions of psychological distress common in Chinese (e.g., “活着没意思”、“快受不了了”、“不如解脱”)
|
| 42 |
+
- Distinguishes risk levels from mild distress to clear suicidal ideation
|
| 43 |
+
- Recommends appropriate assistant strategies, with strong emphasis on escalation and resource provision when suicide risk is present
|
| 44 |
+
- Handles both short single-turn inputs and very long multi-turn conversation contexts
|
| 45 |
|
| 46 |
Intended Use
|
| 47 |
|
| 48 |
+
- Safety layer in Chinese mental health chatbots or counseling apps
|
| 49 |
+
- Automated risk triage for online psychological support platforms
|
| 50 |
+
- Early detection of depression and suicidal ideation in user conversations
|
| 51 |
+
- Research on mental health AI in Chinese-language environments
|
| 52 |
|
| 53 |
Base Model
|
| 54 |
|
|
|
|
| 56 |
|
| 57 |
Fine-tuning Details
|
| 58 |
|
| 59 |
+
- Adapter type: LoRA (r=16, alpha=32, targeting q/k/v/o_proj)
|
| 60 |
+
- Dataset: Custom high-quality Chinese mental health risk assessment data (single-turn + multi-turn)
|
| 61 |
+
- Training objective: Supervised fine-tuning with strict JSON output formatting and EOS enforcement for clean generation
|
| 62 |
|
| 63 |
```python
|
| 64 |
from peft import PeftModel, PeftConfig
|