Device Overview

Models tested

13

Tokens

768,000

15

4

Model Speeds

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

Decode vs. Size

Prefill vs. Size