EXAONE-4.0-1.2B-math-aggressive

🎯 MATH-optimized | 📦 Aggressive pruning | ⚡ 15% weights pruned

This model is a aggressively pruned version of LGAI-EXAONE/EXAONE-4.0-1.2B, specialized for MATH tasks using activation-aware weight pruning (Wanda-style).

✨ Key Features

• Specialization: Optimized for Math tasks
• Pruning Method: Wanda-style (|W| × |activation|) importance scoring
• Size Reduction: 15% weights pruned
• Use Case: Maximum compression for edge deployment

📊 Performance Comparison

Category	Original	Pruned	Change
Python	20.0%	0.0%	↓ 20.0%
Html	6.7%	6.7%	→
Trivia	26.7%	26.7%	→
Math	60.0%	46.7% ⭐	↓ 13.3%
Reasoning	60.0%	60.0%	→
Medical	73.3%	60.0%	↓ 13.3%
Linux	93.3%	66.7%	↓ 26.7%
Writing	60.0%	20.0%	↓ 40.0%

Average: 50.0% → 35.8% (-14.2%)

Math Retention: 77.8% of original performance

🚀 Quick Start

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("CompactAI/EXAONE-4.0-1.2B-math-aggressive")
tokenizer = AutoTokenizer.from_pretrained("CompactAI/EXAONE-4.0-1.2B-math-aggressive")

# Example usage
inputs = tokenizer("Your prompt here", return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

📋 Technical Details

Property	Value
Base Model	LGAI-EXAONE/EXAONE-4.0-1.2B
Specialization	Math
Prune Mode	Aggressive
Pruning Method	Activation-based weight pruning (Wanda)
Weight Reduction	15% weights pruned

🔗 Related Models

This model is part of the EXAONE-4.0-1.2B pruned model collection. Variants:

• Safe - Conservative pruning (~10-20%), high accuracy retention
• Aggressive - Maximum compression (~40-50%), best for edge deployment

📜 License

This model inherits the license from the base model LGAI-EXAONE/EXAONE-4.0-1.2B.

---

Generated by ZANNPS [Zeto Automatic Neural Network Pruning System]