jee_nujan_finetuned_test

JEE NUJAN Fine-tuned - JEE Mathematics Specialist

🏆 The Ultimate JEE Mathematics AI Tutor

Fine-tuned version of JEE NUJAN specifically trained on JEE-style mathematics problems.

🎯 Fine-tuning Details

• Base Model: shivs28/jee_nujan (Qwen2-1.5B-Instruct)
• Training Data: 300 JEE-relevant mathematics problems
• Training Steps: 100 (optimized quality training)
• LoRA Rank: 16 (balanced performance)
• Effective Batch Size: 16

🏅 JEE Performance

This model excels at:

• Problem Recognition: Identifies JEE problem types instantly
• Solution Strategy: Chooses optimal solving approaches
• Step-by-step Reasoning: Clear mathematical progression
• Mathematical Accuracy: Precise calculations and reasoning
• Error Prevention: Highlights common mistakes

🚀 Usage for JEE Preparation

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("shivs28/jee_nujan_finetuned_test")
model = AutoModelForCausalLM.from_pretrained("shivs28/jee_nujan_finetuned_test")

# JEE Advanced Problem
jee_prompt = '''<|jee_problem|>
Problem: If f(x) = x³ - 3x² + 2x - 1, find the number of real roots of f(x) = 0 in the interval [0, 3].

Solution:'''

inputs = tokenizer(jee_prompt, return_tensors="pt")
outputs = model.generate(
    **inputs, 
    max_length=600, 
    temperature=0.1,
    do_sample=True,
    pad_token_id=tokenizer.pad_token_id
)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

📈 Recommended Settings

# For JEE problem solving:
generation_config = {
    "max_length": 800,
    "temperature": 0.1,     # Very low for mathematical accuracy
    "top_p": 0.95,
    "do_sample": True,
    "repetition_penalty": 1.05,
    "pad_token_id": tokenizer.pad_token_id
}

🎯 Training Improvements

• Fixed Gradient Issues: Proper prepare_model_for_kbit_training() setup
• Optimized Quantization: Correct dtype configuration
• Stable Training: Conservative learning rates and batch sizes
• JEE-Focused Data: Mathematics problems relevant to JEE syllabus

---

Ready for JEE conquest! 🎯📚