Device Overview

Models tested

14

Tokens

624,640

15

8

Model Speeds

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

Decode vs. Size

Prefill vs. Size