README.md

# JEE NUJAN - Advanced Mathematics Model

🎯 **Specialized for JEE (Joint Entrance Examination) Mathematics**

A high-quality model optimized for JEE Advanced and JEE Main preparation, based on Qwen2-1.5B-Instruct.

## 🏗️ Architecture

**Base Model:** Qwen/Qwen2-1.5B-Instruct
**Strategy:** Single strong foundation + specialized fine-tuning
**Total Parameters:** 1.5B (optimized for quality and efficiency)
**Precision:** Float16 (high precision inference)

## 🎓 JEE Specialization

This model is specifically designed for:
- **JEE Main Mathematics** (Algebra, Calculus, Coordinate Geometry)
- **JEE Advanced Problems** (Complex analysis, Advanced calculus)
- **Step-by-step Solutions** (Clear mathematical reasoning)
- **Multiple Solution Methods** (Alternative approaches)

## 🚀 Usage

```python
from transformers import AutoTokenizer, AutoModelForCausalLM

# Load JEE NUJAN
tokenizer = AutoTokenizer.from_pretrained("shivs28/jee_nujan", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("shivs28/jee_nujan", trust_remote_code=True)

# JEE Math Problem Example
prompt = '''
JEE Problem: Find the number of real solutions of the equation:
sin(x) = x/10 in the interval [-10π, 10π]

Solve step by step:
'''

inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(
    **inputs, 
    max_length=500, 
    temperature=0.3,  # Lower temperature for mathematical accuracy
    do_sample=True,
    pad_token_id=tokenizer.pad_token_id
)

solution = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(solution)
```

## 📊 Performance Focus

- **Accuracy**: Optimized for mathematical correctness
- **Reasoning**: Clear step-by-step problem solving
- **Coverage**: JEE syllabus comprehensive
- **Speed**: Efficient inference on standard hardware

## 🎯 JEE Topics Covered

- **Algebra**: Quadratic equations, sequences, series
- **Calculus**: Limits, derivatives, integrals
- **Coordinate Geometry**: Lines, circles, parabolas
- **Trigonometry**: Identities, equations, inverse functions
- **Probability**: Combinatorics, probability distributions
- **Vector Algebra**: Dot product, cross product, planes

## ⚡ Inference Tips

```python
# For best JEE math results:
generation_config = {
    "max_length": 800,
    "temperature": 0.2,      # Low for accuracy
    "top_p": 0.9,
    "do_sample": True,
    "repetition_penalty": 1.1
}
```

## 🏆 Model Foundation

Built on Qwen2-1.5B-Instruct:
- Excellent instruction following
- Strong mathematical reasoning
- Efficient parameter usage
- Robust tokenization

---

**Note**: This is the base model. For enhanced JEE performance, see the fine-tuned version: `shivs28/jee_nujan_finetuned`