Falcon-H1-Tiny-90M-Instruct-math-safe

MATH-optimized | Safe pruning | 25% weights pruned

This model is a conservatively pruned version of tiiuae/Falcon-H1-Tiny-90M-Instruct.

Pruning Alert: The benchmarks show virtually NO quality drop! This isn't a bug -- it is a feature. The Wanda pruning algorithm is so effective at identifying unimportant weights that it can remove a large percentage of parameters without affecting performance. Think of it like pruning dead leaves from a tree -- the tree does not miss them because they were not doing anything anyway!

Performance Comparison

Category	Original	Pruned	Change
Python	0.0%	0.0%	→
Html	0.0%	0.0%	→
Trivia	30.0%	30.0%	→
Math	55.0%	55.0% ⭐	→
Reasoning	20.0%	20.0%	→
Medical	5.0%	5.0%	→
Linux	15.0%	15.0%	→
Writing	0.0%	0.0%	→

Average: 15.6% -> 15.6% (+0.0%)

Math Retention: 100.0%

Quick Start

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("CompactAI/Falcon-H1-Tiny-90M-Instruct-math-safe")
tokenizer = AutoTokenizer.from_pretrained("CompactAI/Falcon-H1-Tiny-90M-Instruct-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	tiiuae/Falcon-H1-Tiny-90M-Instruct
Specialization	Math
Prune Mode	Safe
Weight Reduction	25% weights pruned

License

This model inherits the license from the base model.