Can Large Models Teach Student Models to Solve Mathematical Problems Like Human Beings? A Reasoning Distillation Method via Multi-LoRA Interaction

Open paper

Benchmark	Model	Rank	Results
math-word-problem-solving-on-math	LoRID (Qwen2.5-Math)	#49	Accuracy: 61.2
math-word-problem-solving-on-math	LoRID (DeepSeekMath)	#70	Accuracy: 54.8
math-word-problem-solving-on-math	LoRID (Mistral)	#123	Accuracy: 38.7
math-word-problem-solving-on-math	LoRID (LLaMA-2)	#149	Accuracy: 25.2
arithmetic-reasoning-on-gsm8k	LoRID (Qwen2.5-Math)	#61	Accuracy: 91.7
arithmetic-reasoning-on-gsm8k	LoRID (DeepSeekMath)	#73	Accuracy: 90.0
arithmetic-reasoning-on-gsm8k	LoRID (Mistral)	#129	Accuracy: 84.2
arithmetic-reasoning-on-gsm8k	LoRID (LLaMA-2)	#167	Accuracy: 78.5