Device Overview

Models tested

7

Tokens

921,600

18

11

Model Speeds

Results include trials with 4,096 input tokens and 1,024 output tokens only.

Decode vs. Size

Prefill vs. Size