Sumi: Open Uniform Diffusion Language Model from Scratch

Benchmark Model Rank Results
zero-shot-question-answering-on-truthfulqa-mc2-0-shotSumi-7B#3MC2Acc: 46.6
reading-comprehension-on-raceSumi-7B#28Accuracy: 41.4
common-sense-reasoning-on-arc-easySumi-7B#55Accuracy: 70.0
multi-task-language-understanding-on-mmluSumi-7B (5-shot)#63Average (%): 51.1
common-sense-reasoning-on-winograndeSumi-7B (5-shot)#117Accuracy: 60.0
code-generation-on-humanevalSumi-7B#130Pass@1: 22.6
sentence-completion-on-hellaswagSumi-7B#151Accuracy: 60.0
code-generation-on-mbppSumi-7B (3-shot)#154Accuracy: 26.6
question-answering-on-piqaSumi-7B#157Accuracy: 66.4
arithmetic-reasoning-on-gsm8kSumi-7B (4-shot)#248Accuracy: 32.8