Motif 3: Technical Report

Benchmark Model Rank Results
multiple-choice-question-answering-mcqa-on-arc-challenge-25-shotMotif 3 Base (25-shot)#1Accuracy: 94.71
multi-task-language-understanding-on-mmluMotif 3 Base (5-shot)#4Average (%): 86.2
physics-question-answering-on-critptMotif 3#5Accuracy: 6.6
question-answering-on-humanity-s-last-exam-hleMotif 3#5Accuracy: 37.0
instruction-following-on-ifbenchMotif 3#7Accuracy: 78.2
code-generation-on-scicodeMotif 3#14Accuracy: 40.6
question-answering-on-piqaMotif 3 Base (zero-shot)#16Accuracy: 85.14
code-generation-on-mbppMotif 3 Base (3-shot)#19Accuracy: 84.6
code-generation-on-swe-bench-verifiedMotif 3 (mini-SWE-agent)#21Accuracy: 76.2
sentence-completion-on-hellaswagMotif 3 Base (10-shot)#21Accuracy: 88.3
code-generation-on-terminal-bench-2-1Motif 3 (Terminus 2)#22Accuracy: 74.9
common-sense-reasoning-on-winograndeMotif 3 Base (5-shot)#28Accuracy: 80.9
math-word-problem-solving-on-mathMotif 3 Base (4-shot CoT)#33Accuracy: 70.58
arithmetic-reasoning-on-gsm8kMotif 3 Base (8-shot CoT)#36Accuracy: 93.93
multiple-choice-question-answering-mcqa-on-gpqa-diamondMotif 3#46Accuracy: 83.4
multi-task-language-understanding-on-mmlu-proMotif 3 Base (5-shot CoT)#76Accuracy: 68.56
code-generation-on-humanevalMotif 3 Base (zero-shot)#78Pass@1: 73.7