README.md

---
license: apache-2.0
tags:
  - peft
  - lora
  - mental-health
  - therapy
  - transformers
  - llama
  - 4-bit
  - bitsandbytes
library_name: peft
language:
  - en
datasets:
  - kunalchamoli/mental_health_v1
model_name: Meet Lora - Your Own Therapist
model_type: llama
inference: false
pipeline_tag: text-generation
---

# Meet Lora – Your Own Therapist 🧠💬


It is a **fine-tuned LLM** built to simulate a therapist, trained on a dataset of therapy sessions.  
It leverages **Parameter-Efficient Fine-Tuning (PEFT)** using **LoRA (Low-Rank Adaptation)** on **Meta LLaMA 2 7B**.

The model was fine-tuned on a **Tesla T4 GPU**, selectively training a small number of parameters to efficiently adapt the base model to the domain of mental health and therapy.

> 💡 Dataset used is available on [Hugging Face 🤗](https://huggingface.co/datasets/kunalchamoli/mental_health_v1)

---

## 🔧 How to Use

```python
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
from peft import PeftModel

# Load the base model in 4-bit precision
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_compute_dtype=torch.float16,
    bnb_4bit_quant_type="nf4"
)

base_model = AutoModelForCausalLM.from_pretrained(
    "meta-llama/Llama-2-7b-hf",
    quantization_config=bnb_config,
    device_map="auto"
)

tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-hf")

# Load the adapter
model = PeftModel.from_pretrained(base_model, "aryan27/llama-therapy-lora")

# System prompt to guide behavior
system_prompt = (
    "You are a compassionate and thoughtful therapist. "
    "Your responses are empathetic, non-judgmental, and helpful."
)

# 🧩 Function to generate response
def generate_therapy_response(user_input: str):
    prompt = f"{system_prompt}\nUser: {user_input}\nTherapist:"

    inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=200,
            do_sample=True,
            temperature=0.7,
            top_p=0.9,
            repetition_penalty=1.1,
            pad_token_id=tokenizer.eos_token_id
        )

    response = tokenizer.decode(outputs[0], skip_special_tokens=True)

    # Strip off the prompt from the generated text
    response_only = response.split("Therapist:")[-1].strip()

    return response_only

# 💬 Example usage
response = generate_therapy_response("I feel demotivated because of my breakup")
print(response)