Chameleon: Mixed-Modal Early-Fusion Foundation Models

Benchmark Model Rank Results
zero-shot-question-answering-on-openbookqaChameleon 34B#5Accuracy: 54.0
question-answering-on-boolqChameleon 34B#21Accuracy: 86.0
visual-question-answering-on-vqa-v2-test-devChameleon-MultiTask#34Accuracy: 69.6
common-sense-reasoning-on-winograndeChameleon 34B#39Accuracy: 78.5
sentence-completion-on-hellaswagChameleon 34B#56Accuracy: 82.7
math-word-problem-solving-on-mathChameleon 34B#154Accuracy: 22.5
arithmetic-reasoning-on-gsm8kChameleon 34B#205Accuracy: 61.4