DeepSeek-VL: Towards Real-World Vision-Language Understanding

Benchmark Model Rank Results
visual-question-answering-vqa-on-mmbench-cnDeepSeek-VL 7B Chat#19Accuracy: 72.8
visual-question-answering-vqa-on-seed-seedbenchDeepSeek-VL 7B Chat#23Accuracy: 70.4
optical-character-recognition-ocr-on-ocrbenchDeepSeek-VL 7B Chat#70Score: 456
visual-question-answering-on-mm-vetDeepSeek-VL#93GPT-4 score: 41.5
multimodal-reasoning-on-mathvistaDeepSeek-VL 7B Chat#114Accuracy: 36.1
sentence-completion-on-hellaswagDeepSeek-VL 7B Chat#118Accuracy: 68.4
multimodal-reasoning-on-mmmuDeepSeek-VL 7B Chat#130Accuracy: 36.6
code-generation-on-mbppDeepSeek-VL 7B Chat#142Accuracy: 35.2
arithmetic-reasoning-on-gsm8kDeepSeek-VL 7B Chat#227Accuracy: 55.0