|DEEP REASONING MODEL|LARGE LANGUAGE MODEL|REASONING|

Challenging the Big Data Myth: How AI Achieves Complex Reasoning with Surprisingly Few Examples, and How Does It Work

Large Reasoning Models (LRMs) learn complex reasoning with minimal data using the Less-Is-More Reasoning Hypothesis (LIMO). Long chain-of-thought (Long CoT) reasoning is efficiently trained with supervised fine-tuning (SFT) and LoRA. The LIMO model achieves 57.1% on AIME and 94.8% on MATH with just 817 samples, redefining AI training efficiency. Access the open-source LIMO suite for cutting-edge research.

image by the author using AI

I have my own definition of minimalism, which is that which is created with a minimum of means. — La Monte Young

Large reasoning models (LRMs) are the latest frontier of large language models (LLMs), and are obtained with additional training, exploiting long chain-of thoughts (Long CoTs) with reflection, backtracking, and self-validation to tackle challenging reasoning tasks. These models have shown superior capability on reasoning benchmarks but at the cost of higher computational costs. In fact, this led to the new concept of test-time computing to improve model capabilities. In other words, it is not scaling the model but how much the model “thinks” about a given question.