NahiyanAnwar/ mental-health-ai:latest

17 3 days ago

Compassionate mental health support AI fine-tuned from Qwen 3 (14B) with 128K context. 

ollama run NahiyanAnwar/mental-health-ai

Details

3 days ago

b151c7759b0a · 9.0GB ·

model
qwen2
·
14.8B
·
Q4_K_M
template
{{ if .System }}<|im_start|>system {{ .System }}<|im_end|> {{ end }}{{ range .Messages }}{{ if eq .R
system
You are a compassionate mental health support assistant. Your purpose is to provide emotional suppor
params
{ "num_ctx": 131072, "num_predict": 1024, "repeat_penalty": 1.1, "stop": [ "

Readme

Compassionate Mental Health AI Assistant

A supportive, empathetic AI assistant fine-tuned from the Qwen 3 (14B) Instruct base architecture using LoRA adapters. The model is specifically optimized for cognitive behavioral therapy (CBT) coaching, empathetic validation, and active listening, with an expanded context window of 128K tokens (131,072) to handle long, multi-turn therapeutic conversations.

⚠️ Safety & Crisis Disclaimer

[!IMPORTANT] This AI is NOT a licensed therapist, clinical medical device, or replacement for professional mental health care. - It cannot diagnose mental health conditions, prescribe medication, or formulate clinical treatment plans. - It is designed solely for supportive conversation, psychoeducation, and basic coping strategy guidance.

If you are experiencing thoughts of self-harm, suicide, or any acute crisis, please reach out to professional services immediately: - In the US: Call or text 988 to reach the Suicide & Crisis Lifeline, or text HOME to 741741 for the Crisis Text Line. - In the UK: Call 111 or contact Samaritans at 116 123. - International: Find resources in your area at findahelpline.com. - Emergency: Call your local emergency services (e.g., 911, 999, 112) or go to the nearest hospital.

🌟 Key Features & Guidelines

  • Empathetic Active Listening: Validates feelings first before jumping into problem-solving.
  • Evidence-Based Coping Strategies: Guides users through progressive muscle relaxation, grounding exercises (like 5-4-3-2-1), and CBT-based thought challenging.
  • Strict Clinical Boundaries: Gracefully refuses requests for medical diagnosis or drug prescriptions, referring the user to human professionals.
  • Extended Memory: Supports up to a 128,000 token context window to remember details across long chat histories.

📚 Datasets & Fine-Tuning

The model was fine-tuned using Unsloth on structured, high-quality emotional support and counseling datasets: - ESConv (Emotional Support Conversations) - CounselChat (Verified responses from licensed therapists) - Empathetic Dialogues - MentalChat16K

💻 How to Use

To run this model locally using Ollama:

”`bash ollama run nahiyananwar/mental-health-ai