List your company

Peak decode throughput on LLaDA 16B 16K context 2.0-mini

3,475Throughput (tokens/s)HERALD
438.21,226.62,0152,803.4Jun 19, 2026

Evaluation results

MethodLinks
HERALDBatch size (B)=1612026.063,4752.13
DenseBatch size (B)=352026.061,6291
MAGE-OffloadBatch size (B)=1282026.067960.49
InfiniGenBatch size (B)=162026.065550.34