This model is a specialized, privacy-focused mental health assistant. It was fine-tuned to provide empathetic, evidence-based responses to mental health-related inquiries, grounded in clinical guidelines[cite: 1]. The model is optimized for local deployment on consumer hardware[cite: 1].

3. Training Data

Dataset: nbertagnolli/counsel-chat[cite: 1]
Data Processing:
- Formatted using standard Llama 3 prompt templates[cite: 1].
- Filtered to prioritize empathetic, assistant-style responses[cite: 1].

4. Training Procedure

Method: QLoRA (Quantized Low-Rank Adaptation)[cite: 1]
Framework: Unsloth[cite: 1]
Parameters:
- r = 16, lora_alpha = 16, lora_dropout = 0, bias = "none"[cite: 1]
- max_steps = 60, learning_rate = 2e-4, batch_size = 2 (with 4x gradient accumulation)[cite: 1]

5. Usage

This model is designed to be used with llama.cpp or compatible UI frameworks like LM Studio.

Prompt Template:

<|start_header_id|>user<|end_header_id|>

{user_query}<|eot_id|><|start_header_id|>assistant<|end_header_id|>

6. Limitations & Ethical Considerations

Not a Doctor: This model is for support and information purposes only. It is not a substitute for professional clinical diagnosis or medical treatment.
Safety: The model is intended to provide empathetic responses but may occasionally hallucinate or provide outdated information. Users should consult mental health professionals for medical decisions.

Horiznsky/Serenity-Llama-3.2-3B-Counsel

Model Card: Serenity-Mental-Health-Assistant

1. Model Details

2. Model Description

3. Training Data

4. Training Procedure

5. Usage

Prompt Template:

6. Limitations & Ethical Considerations

Horiznsky/
Serenity-Llama-3.2-3B-Counsel