List your company

Constrained LLM Decoding on Llama-2-70B

27.2Latency (ms)Pre³
24.851240.705656.5672.4144Jun 4, 2025

Evaluation results

MethodLinks
Pre³Batch Size=16, Hardware=4xH8002025.0627.25.39
XGrammarBatch Size=16, Hardware=4xH8002025.0628.75—
Pre³Batch Size=64, Hardware=4xH8002025.0648.4814.85
Pre³Batch Size=32, Hardware=4xH8002025.0654.241.6
XGrammarBatch Size=32, Hardware=4xH8002025.0655.12—
Pre³Batch Size=128, Hardware=4xH8002025.0655.3919.48
XGrammarBatch Size=64, Hardware=4xH8002025.0656.94—
XGrammarBatch Size=128, Hardware=4xH8002025.0668.79—
Pre³Batch Size=256, Hardware=4xH8002025.0675.7211.87
XGrammarBatch Size=256, Hardware=4xH8002025.0685.92—