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

🎯 MATH-optimized | 📦 Safe pruning | ⚡ 1% weights pruned

This model is a conservatively pruned version of LiquidAI/LFM2.5-1.2B-Thinking.

Performance Comparison

Category	Original	Pruned	Change
Python	0.0%	5.0%	↑ 5.0%
Html	25.0%	25.0%	→
Trivia	97.5%	97.5%	→
Math	100.0%	100.0% ⭐	→
Reasoning	82.5%	82.5%	→
Medical	95.0%	95.0%	→
Linux	35.0%	40.0%	↑ 5.0%
Writing	80.0%	82.5%	↑ 2.5%

Average: 64.4% → 65.9% (+1.6%)

Math Retention: 100.0%

Quick Start

from transformers import AutoModelForCausalLM, AutoTokenizer

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

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	Safe
Weight Reduction	1% weights pruned

License

This model inherits the license from the base model.