Dynin-Omni: Omnimodal Unified Large Diffusion Language Model

Benchmark Model Rank Results
video-question-answering-on-tempcompass-overallDynin-Omni (8B)#12Accuracy: 68.4
zero-shot-video-question-answer-on-video-mme-w-o-subsDynin-Omni (8B)#44Accuracy (%): 61.4
text-to-image-generation-on-genevalDynin-Omni#45Overall: 0.87
video-question-answering-on-mvbenchDynin-Omni (8B)#60Avg.: 62.0
speech-recognition-on-librispeech-test-cleanDynin-Omni (8B)#85Word Error Rate (WER): 2.1
math-word-problem-solving-on-mathDynin-Omni (8B)#87Accuracy: 49.6
speech-recognition-on-librispeech-test-otherDynin-Omni (8B)#89Word Error Rate (WER): 4.8
multimodal-reasoning-on-mmmuDynin-Omni (8B)#92Accuracy: 51.6
arithmetic-reasoning-on-gsm8kDynin-Omni (8B)#96Accuracy: 87.6