LFM2.5-1.2B-Thinking-math-heavy

🎯 MATH-optimized | 📦 Heavy pruning | ⚡ 30% weights pruned

This model is a aggressively pruned version of LiquidAI/LFM2.5-1.2B-Thinking, 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: 30% weights pruned
• Use Case: Maximum compression while maintaining usability

📊 Performance Comparison

Category	Original	Pruned	Change
Python	0.0%	0.0%	→
Html	0.0%	0.0%	→
Trivia	93.3%	66.7%	↓ 26.7%
Math	100.0%	100.0% ⭐	→
Reasoning	N/A	N/A
Medical	86.7%	80.0%	↓ 6.7%
Linux	86.7%	73.3%	↓ 13.3%
Writing	60.0%	20.0%	↓ 40.0%

Average: 61.0% → 48.6% (-12.4%)

Math Retention: 100.0% of original performance

🚀 Quick Start

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("CompactAI/LFM2.5-1.2B-Thinking-math-heavy")
tokenizer = AutoTokenizer.from_pretrained("CompactAI/LFM2.5-1.2B-Thinking-math-heavy")

# 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	LiquidAI/LFM2.5-1.2B-Thinking
Specialization	Math
Prune Mode	Heavy
Pruning Method	Activation-based weight pruning (Wanda)
Weight Reduction	30% weights pruned

🔗 Related Models

This model is part of the LFM2.5-1.2B-Thinking pruned model collection. Other variants:

• Extra-light (minimal pruning)
• Light
• Medium-light
• Medium
• Medium-heavy
• Heavy
• Extra-heavy (maximum compression)

📜 License

This model inherits the license from the base model LiquidAI/LFM2.5-1.2B-Thinking.

---

Generated by ZANNPS [Zeto Automatic Neural Network Pruning System]