List your company

Inference Efficiency on MoE LLMs DSV2-16B, QW3-30B, QW3-80B-I

12.46Decode Speed (tokens/sec)BITSMOE
1.21764.13637.0559.9737May 22, 2026

Evaluation results

MethodLinks
BITSMOEModel=DSV2-16B2026.0512.460.64———5.085.44
FP16Model=DSV2-16B2026.0510.390.47—29.510.6927.65—
GPTQModel=DSV2-16B2026.057.431.27—————
BITSMOEModel=QW3-30B2026.055.711.51———8.586.29
BITSMOEModel=QW3-80B-I2026.055.011.06———21.986.56
GPTQModel=QW3-30B2026.053.252.94—————
FP16Model=QW3-30B2026.053.072.35—56.951.6954—
GPTQModel=QW3-80B-I2026.052.597.32—————
FP16Model=QW3-80B-I2026.051.658.35—148.690.61144.28—