Sparse Query Attention (SQA) Research
Collection
Experimental models with Sparse Query Attention layers. Reducing training time/cost by ~3-10% compared to GQA & MQA, with the same level performance
•
9 items
•
Updated
•
1