Device Overview

Models tested

26

Tokens

1,536,000

30

5

Model Speeds

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

Decode vs. Size

Prefill vs. Size