Solving Quantitative Reasoning Problems with Language Models

Benchmark Model Rank Results
math-word-problem-solving-on-mathMinerva 62B (maj5@256)#43Accuracy: 64.9Parameters (Billions): 62
math-word-problem-solving-on-mathMinerva 540B (maj1@k, k=64)#85Accuracy: 50.3
math-word-problem-solving-on-mathMinerva 8B (maj5@256)#97Accuracy: 47.6Parameters (Billions): 8
math-word-problem-solving-on-mathMinerva 62B (maj1@k, k=64)#115Accuracy: 43.4Parameters (Billions): 62
math-word-problem-solving-on-mathMinerva 540B#129Accuracy: 33.6Parameters (Billions): 540
math-word-problem-solving-on-mathMinerva 62B (4-shot)#145Accuracy: 27.6Parameters (Billions): 62
math-word-problem-solving-on-mathMinerva 8B (maj1@k, k=64)#148Accuracy: 25.4Parameters (Billions): 8
math-word-problem-solving-on-mathdavinci-002 175B#163Accuracy: 19.1Parameters (Billions): 175
math-word-problem-solving-on-mathMinerva 8B#169Accuracy: 14.1Parameters (Billions): 8
math-word-problem-solving-on-mathPaLM 540B#180Accuracy: 8.8Parameters (Billions): 540
math-word-problem-solving-on-mathPaLM 8B (fine-tuned)#190Accuracy: 5.6Parameters (Billions): 8
math-word-problem-solving-on-mathPaLM 62B#194Accuracy: 4.4Parameters (Billions): 62
math-word-problem-solving-on-mathPaLM 8B#199Accuracy: 1.5Parameters (Billions): 8
arithmetic-reasoning-on-gsm8kMinerva 62B (maj5@100)#85Accuracy: 89Parameters (Billion): 62
arithmetic-reasoning-on-gsm8kMinerva 540B (CoT)#166Accuracy: 78.5Parameters (Billion): 540
arithmetic-reasoning-on-gsm8kMinerva 62B (maj1@100)#201Accuracy: 68.5Parameters (Billion): 62
arithmetic-reasoning-on-gsm8kMinerva 8B (maj5@100)#221Accuracy: 56.8Parameters (Billion): 8
arithmetic-reasoning-on-gsm8kPaLM 540B (8-shot)#223Accuracy: 56.5Parameters (Billion): 540
arithmetic-reasoning-on-gsm8kMinerva 62B (8-shot)#231Accuracy: 52.4Parameters (Billion): 62
arithmetic-reasoning-on-gsm8kPaLM 62B (8-shot)#256Accuracy: 33.0Parameters (Billion): 62
arithmetic-reasoning-on-gsm8kMinerva 8B-maj1@k (8-shot)#260Accuracy: 28.4Parameters (Billion): 8
arithmetic-reasoning-on-gsm8kMinerva 8B (8-shot)#269Accuracy: 16.2Parameters (Billion): 8
arithmetic-reasoning-on-gsm8kPaLM 8B (8-shot)#278Accuracy: 4.1Parameters (Billion): 8