---
language:
- en
license: apache-2.0
tags:
- medical
- healthcare
- phi-2
- tiny-llama
- medical-assistant
- lora
- finetuned
- gguf
- quantised
datasets:
- chatdoctor
- medquad
- tatsu-lab/alpaca
metrics:
- perplexity
base_model: microsoft/phi-2
pipeline_tag: text-generation
library_name: transformers
---

<div align="center">

# 🩺 Yukt-Med (Phi-2 Medical Assistant)

<img src="https://github.com/ggerganov/llama.cpp/blob/master/media/logo.png?raw=true" width="100" alt="llama.cpp Logo">
<img src="https://huggingface.co/front/assets/huggingface_logo-noborder.svg" width="80" alt="Hugging Face Logo" style="vertical-align: middle;">

<br>

[![Model Size: 1.74GB (GGUF)](https://img.shields.io/badge/Model%20Size-1.74GB%20(GGUF)-blue?style=for-the-badge&logo=huggingface)](https://huggingface.co/ayuag/yukt-med/blob/main/yukt-med-Q4_K_M.gguf)
[![Base Model: Phi-2](https://img.shields.io/badge/Base%20Model-Phi--2%20(2.7B)-green?style=for-the-badge&logo=microsoft)](https://huggingface.co/microsoft/phi-2)
[![License: Apache 2.0](https://img.shields.io/badge/License-Apache%202.0-yellow?style=for-the-badge&logo=apache)](https://opensource.org/licenses/Apache-2.0)
[![Dataset: 86K+ Examples](https://img.yarnpkg.com/v/86k+-examples.svg?style=for-the-badge&label=Dataset&color=orange)](https://huggingface.co/ayuag/yukt-med/blob/main/README.md#training-data)

<br>

<h3>Your Compact, Specialized Medical Knowledge Companion.</h3>

<p>Fine-tuned on 86,000+ curated medical interactions to provide concise, accurate, and non-diagnostic healthcare information.</p>

---

</div>

## 🌟 Overview

**Yukt-Med** is a lightweight, state-of-the-art language model designed for the medical and healthcare domain. It is fine-tuned using LoRA (Low-Rank Adaptation) on a diverse collection of healthcare datasets. 

What makes Yukt-Med unique is its balance of performance and efficiency. While powerful, it has been **quantized to 4-bit GGUF**, making it runnable on commodity hardware, mobile devices, and in offline environments.

> **💡 Perfect for:** Rapid medical information retrieval, symptom analysis support, and educational purposes. **Not for diagnosis.**

---

## 🚀 Key Features

| Feature | Description |
| :--- | :--- |
| **🧠 Specialized Brain** | Trained on ChatDoctor, MedQuad, and curated drug databases. |
| **⚡ Ultra-Efficient** | GGUF version runs smoothly on **4GB RAM** (CPU/Mobile). |
| **Instruction-Following** | Responds accurately to instructions using a specific prompt template. |
| **🌐 Production-Ready** | Available in Standard Safetensors and Compact GGUF formats. |

---

## 📊 Training Data

The model's knowledge comes from over **86,800 cleaned and structured examples**:

1.  **ChatDoctor Dataset:** Real-world patient-doctor dialogues for conversational medical advice.
2.  **MedQuad Dataset:** Large-scale Medical Question-Answering pairs.
3.  **Drugs & Side Effects:** Detailed information on pharmaceuticals.
4.  **Symptom-Disease Mapping:** Patterns for common medical conditions.

*(Note: Data was filtered to ensure high-quality, safe, and factual content.)*

---

## 🧪 Evaluation Examples

Compare Yukt-Med's focused medical responses:

| Instruction (Prompt) | Yukt-Med Response (Generation) |
| :--- | :--- |
| **### Instruction:** What are the common symptoms of a common cold?<br>**### Response:** | The common symptoms of a common cold include a runny or stuffy nose, sneezing, sore throat, cough, and mild body aches or fatigue. |
| **### Instruction:** Tell me about the side effects of Aspirin.<br>**### Response:** | The most frequent side effects are bleeding and stomach upset. |

---

## 🛠️ Usage Instructions

### 1. Using Standard Transformers (Python/GPU)

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "ayuag/yukt-med"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16, device_map="auto", trust_remote_code=True)

def ask_medical_bot(question):
    prompt = f"### Instruction: {question}\n### Response:"
    inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
    outputs = model.generate(**inputs, max_new_tokens=200, do_sample=True, temperature=0.7)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

print(ask_medical_bot("What are symptoms of Malaria?"))