metadata
license: mit
datasets:
- vibhorag101/phr-mental-therapy-dataset-conversational-format
language:
- en
base_model:
- unsloth/Llama-3.2-3B-Instruct
tags:
- text-generation-inference
- unsloth
Overview
The chatbot has been fine-tuned on the PHR Therapy Dataset using LLaMA 3.2 3B Instruct, enhancing its ability to engage in meaningful and supportive conversations.
Features
- Empathetic Responses: Trained to understand and respond with emotional intelligence.
- Context Awareness: Retains context over multiple interactions.
- Mental Health Focus: Provides supportive and non-judgmental responses based on therapy-related dialogues.
- Efficient Inference: Optimized for deployment with reduced latency.
Model Fine-Tuning Details
- Base Model: LLaMA 3.2 3B Instruct
- Dataset: PHR Therapy Dataset (contains therapist-patient conversations for empathetic response generation)
- Fine-Tuning Framework: Unsloth (optimized training for efficiency)
- Training Environment: Google colab free version
- Optimization Techniques:
- LoRA (Low-Rank Adaptation) for parameter-efficient tuning
- Mixed Precision Training for speed and memory efficiency
- Supervised Fine-Tuning (SFT) on therapist-patient interactions
Installation
Using ollama
ollama run hf.co/Ishan93/Fine_tuned_ver2
Usage
Using Google Colab or other notebooks
from llama_cpp import Llama
llm = Llama.from_pretrained(
repo_id="Ishan93/Fine_tuned_ver2",
filename="Fine_tuned_ver2.gguf",
)