List your company

Large Language Model Inference on Llama 3.1

48.1Latency (ms)RoME
47.99248.72149.4550.179Apr 10, 2026

Evaluation results

MethodLinks
RoMEModel=Llama3.1, Parameters=8B, QKShape[BNSD]=[1, 24, 8192, 128]2026.0448.14.7
RoPEModel=Llama3.1, Parameters=8B, QKShape[BNSD]=[1, 24, 8192, 128]2026.0450.8—