klyang/MentaLLaMA-chat-13B

MentaLLaMA-chat-13B: Interpretable Mental Health Analysis

MentaLLaMA-chat-13B is a 13 billion parameter model from the MentaLLaMA project, developed by klyang. It is fine-tuned from Meta's LLaMA2-chat-13B using the comprehensive IMHI instruction tuning dataset, which includes 75,000 high-quality natural language instructions.

Key Capabilities

• Interpretable Mental Health Analysis: Designed to perform complex mental health analyses and provide reliable explanations for its predictions across various mental health conditions.
• Instruction-Following: Enhanced with instruction-following capabilities through fine-tuning on the IMHI dataset.
• Performance: Approaches state-of-the-art discriminative methods in correctness on the IMHI benchmark (20K test samples) while generating high-quality explanations.

Ethical Considerations & Limitations

While showing promising performance, MentaLLaMA-chat-13B is strictly intended for non-clinical research purposes. Users seeking assistance should consult professional psychiatrists or clinical practitioners. The developers acknowledge potential biases (e.g., gender gaps), incorrect predictions, inappropriate explanations, and over-generalization as existing challenges, highlighting that significant work remains for real-world mental health monitoring system applications.

Good For

• Researchers in AI and mental health exploring interpretable models.
• Developing tools for non-clinical mental health analysis and explanation generation.
• Investigating the application of LLMs in understanding and explaining mental health-related text.