List your company

Prefill Latency on Llama-3-8B 200 Mbps

1.541Prefill Latency (s)ASTRA
0.681646.4823212.28318.08368May 25, 2025

Evaluation results

MethodLinks
ASTRAG=12025.051.541
ASTRAG=162025.051.559
BPStrategy=Block Parallelism, Nb=42025.051.586
ASTRAG=322025.051.603
BPStrategy=Block Parallelism, Nb=82025.051.762
SPStrategy=Sequence Parallelism2025.052.857
Llama-3-8B (Single-device)Device=Single-device2025.054.578
TPStrategy=Tensor Parallelism2025.0523.025