CBT-Copilot

CBT-Copilot is a fine-tuned version of meta-llama/Llama-3.2-3B-Instruct, specifically designed to simulate compassionate and supportive dialogues in the style of Cognitive Behavioral Therapy (CBT). This model provides empathetic, structured therapeutic conversations while maintaining professional boundaries.

Model Overview

• Base Model: meta-llama/Llama-3.2-3B-Instruct
• Fine-tuning Method: LoRA (Low-Rank Adaptation)
• Dataset: Lumiiree/therapod-dpo
• Model Size: 3B parameters
• Specialty: Cognitive Behavioral Therapy conversations
• Deployment: Compatible with transformers, vLLM, and other inference frameworks

Quick Start

Installation

pip install transformers torch accelerate

Basic Usage

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

# Load model and tokenizer
model_name = "thillaic/CBT-Copilot"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.float16,
    device_map="auto"
)

# Generate response
def generate_cbt_response(prompt, max_length=512):
    inputs = tokenizer.encode(prompt, return_tensors="pt")
    
    with torch.no_grad():
        outputs = model.generate(
            inputs,
            max_length=max_length,
            temperature=0.7,
            do_sample=True,
            pad_token_id=tokenizer.eos_token_id
        )
    
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return response[len(prompt):].strip()

# Example usage
prompt = "I've been feeling overwhelmed with work lately and can't seem to manage my stress."
response = generate_cbt_response(prompt)
print(response)

Web Interface with Streamlit

For an interactive web interface, you can use the included Streamlit app:

# Install additional dependencies
pip install streamlit

# Run the web interface
streamlit run app.py

This will launch a user-friendly web interface where you can have conversations with CBT-Copilot through your browser.

Screenshot – Example Question:

Screenshot – Example Response:

Using with vLLM (Recommended for Production)

from vllm import LLM, SamplingParams

# Initialize the model
llm = LLM(model="thillaic/CBT-Copilot")

# Set sampling parameters
sampling_params = SamplingParams(
    temperature=0.7,
    top_p=0.9,
    max_tokens=512
)

# Generate responses
prompts = ["I'm struggling with negative thoughts about myself."]
outputs = llm.generate(prompts, sampling_params)

for output in outputs:
    print(output.outputs[0].text)

Model Details

Training Configuration

The model was fine-tuned using the following configuration:

• LoRA Parameters:

  • r=8
  • lora_alpha=16
  • target_modules=["q_proj", "v_proj"]
  • lora_dropout=0.05

• Training Parameters:

  • learning_rate=2e-5
  • batch_size=1 (with gradient accumulation)
  • epochs=1
  • optimizer=paged_adamw_8bit

Dataset

Fine-tuned on the Lumiiree/therapod-dpo dataset, which contains high-quality therapeutic conversation examples structured for CBT-style interactions.

Use Cases

• Mental Health Support: Provide initial emotional support and coping strategies
• Educational Tool: Demonstrate CBT techniques and therapeutic communication
• Research: Study therapeutic dialogue patterns and intervention strategies
• Training: Help train mental health professionals in CBT communication styles

Important Disclaimers

This model is for educational and research purposes only. It is NOT a replacement for professional mental health treatment.

• This AI model cannot provide clinical diagnosis or treatment
• For serious mental health concerns, please consult licensed mental health professionals
• The model's responses should be used as supplementary support, not primary treatment
• Always prioritize professional medical advice for mental health issues

Performance & Evaluation

The model demonstrates strong performance in:

• Empathetic response generation
• CBT technique application
• Maintaining therapeutic boundaries
• Providing structured support

Detailed evaluation metrics and benchmarks will be added as they become available.

Fine-tuning Your Own Version

If you want to fine-tune this model further or create your own version:

# Clone this repository
git clone https://github.com/yourusername/CBT-Copilot
cd CBT-Copilot

# Install dependencies
pip install transformers accelerate peft trl datasets bitsandbytes

# Run fine-tuning
python model_finetuning.py

See model_finetuning.py for the complete training script.

Deployment Options

Local Deployment

• Use transformers library for small-scale applications
• Compatible with consumer GPUs (8GB+ VRAM recommended)

Production Deployment

• vLLM: Recommended for high-throughput serving
• Text Generation Inference: Hugging Face's production-ready solution
• Ollama: For local, user-friendly deployment

Cloud Deployment

• Deploy on Hugging Face Inference Endpoints
• Compatible with AWS SageMaker, Google Cloud AI Platform
• Docker containers available for scalable deployment

Contributing

We welcome contributions! Please feel free to:

• Report bugs and issues
• Suggest improvements
• Submit pull requests
• Share evaluation results

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

Acknowledgments

• Meta AI for the base Llama 3.2 model
• Lumiiree for the therapod-dpo dataset
• The open-source community for tools and libraries used in this project

---

Built for the mental health and AI communities