Device Overview

Models tested

9

Tokens

563,200

11

4

Model Speeds

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

Decode vs. Size

Prefill vs. Size