List your company

Prefill Latency on Llama-3-8B 100 Mbps

1.545Prefill Latency (s)ASTRA
-0.1731611.4244223.02234.61958May 25, 2025

Evaluation results

MethodLinks
ASTRAG=12025.051.545
ASTRAG=162025.051.572
ASTRAG=322025.051.63
BPStrategy=Block Parallelism, Nb=42025.051.753
BPStrategy=Block Parallelism, Nb=82025.052.101
SPStrategy=Sequence Parallelism2025.054.215
Llama-3-8B (Single-device)Device=Single-device2025.054.578
TPStrategy=Tensor Parallelism2025.0544.499