Device Overview

Models tested

7

Tokens

1,126,400

22

17

Model Speeds

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

Decode vs. Size

Prefill vs. Size