Device Overview

Models tested

8

Tokens

409,600

8

3

Model Speeds

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

Decode vs. Size

Prefill vs. Size