General LLM Acceleration on Composite Benchmarks (LongBench, RULER, MATH500)
1.56Average Decoding SpeedupPRR
Evaluation Results
| Method | Links | |
|---|---|---|
| PRRModel=GLM-4, DSA mechanism=InfLLM-v22026.06 | 1.56 | |
| PRRModel=GLM-Z1, DSA mechanism=InfLLM-v22026.06 | 1.55 | |
| PRRModel=Qwen3-14B, DSA mechanism=InfLLM-v22026.06 | 1.51 | |
| PRRModel=DeepSeek-R1, DSA mechanism=InfLLM-v22026.06 | 1.44 | |
| PRRModel=Llama3, DSA mechanism=InfLLM-v22026.06 | 1.43 | |
| PRRModel=GLM-4, DSA mechanism=Quest2026.06 | 1.42 | |
| PRRModel=GLM-Z1, DSA mechanism=Quest2026.06 | 1.42 | |
| PRRModel=DeepSeek-R1, DSA mechanism=Quest2026.06 | 1.41 | |
| PRRModel=Qwen3-14B, DSA mechanism=Quest2026.06 | 1.4 | |
| PRRModel=Llama3, DSA mechanism=Quest2026.06 | 1.38 | |
| PRRModel=Qwen3-32B, DSA mechanism=InfLLM-v22026.06 | 1.31 | |
| PRRModel=Qwen3-32B, DSA mechanism=Quest2026.06 | 1.28 |