List your company

General LLM Acceleration on Composite Benchmarks (LongBench, RULER, MATH500)

1.56Average Decoding SpeedupPRR
1.26881.34441.421.4956Jun 29, 2026

Evaluation results

MethodLinks
PRRModel=GLM-4, DSA mechanism=InfLLM-v22026.061.56
PRRModel=GLM-Z1, DSA mechanism=InfLLM-v22026.061.55
PRRModel=Qwen3-14B, DSA mechanism=InfLLM-v22026.061.51
PRRModel=DeepSeek-R1, DSA mechanism=InfLLM-v22026.061.44
PRRModel=Llama3, DSA mechanism=InfLLM-v22026.061.43
PRRModel=GLM-4, DSA mechanism=Quest2026.061.42
PRRModel=GLM-Z1, DSA mechanism=Quest2026.061.42
PRRModel=DeepSeek-R1, DSA mechanism=Quest2026.061.41
PRRModel=Qwen3-14B, DSA mechanism=Quest2026.061.4
PRRModel=Llama3, DSA mechanism=Quest2026.061.38
PRRModel=Qwen3-32B, DSA mechanism=InfLLM-v22026.061.31
PRRModel=Qwen3-32B, DSA mechanism=Quest2026.061.28