List your company

Constrained LLM Decoding on Llama-3-8B

11.77Inference Time (ms)Pre³
1.3371.8142.27212.74Jun 4, 2025

Evaluation results

MethodLinks
Pre³Batch Size=16, Hardware=2xH8002025.0611.7722.49
XGrammarBatch Size=16, Hardware=2xH8002025.0615.19—
Pre³Batch Size=32, Hardware=2xH8002025.0631.1228.78
Pre³Batch Size=64, Hardware=2xH8002025.0635.8831.09
XGrammarBatch Size=32, Hardware=2xH8002025.0643.69—
Pre³Batch Size=128, Hardware=2xH8002025.0645.3230.5
XGrammarBatch Size=64, Hardware=2xH8002025.0652.07—
Pre³Batch Size=256, Hardware=2xH8002025.0664.4229.2
XGrammarBatch Size=128, Hardware=2xH8002025.0665.21—
XGrammarBatch Size=256, Hardware=2xH8002025.0690.98—
Pre³Batch Size=512, Hardware=2xH8002025.06104.4629.24
XGrammarBatch Size=512, Hardware=2xH8002025.06147.64—
Pre³Batch Size=1024, Hardware=2xH8002025.06201.1626.25
XGrammarBatch Size=1024, Hardware=2xH8002025.06272.77—