Device Overview

Models tested

17

Tokens

870,400

17

1

Model Speeds

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

Decode vs. Size

Prefill vs. Size