MassAlloc Attention: Let Attention Allocate Its Own Compute

Benchmark Model Rank Results
question-answering-on-piqaMALA (32B)#10Accuracy: 85.94
common-sense-reasoning-on-winograndeMALA (32B)#29Accuracy: 80.87
sentence-completion-on-hellaswagMALA (32B)#40Accuracy: 84.69
multi-task-language-understanding-on-mmlu-proMALA (32B)#78Accuracy: 68.33