LLM Inference Acceleration on MT-bench
5.5818Mean Accepted Length (tau)CATS w/ EAGLE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CATS w/ EAGLEModel=Llama2-7b, Decoding Strategy=Relaxed (temperature = 0.7), Decoding Structure=tree base2026.05 | 5.5818 | 5.58 | |
| CATS w/ EAGLEModel=Llama2-13b, Decoding Strategy=Relaxed (temperature = 0.7), Decoding Structure=tree base2026.05 | 5.3499 | 5.28 | |
| EAGLE-self, treeModel=Llama2-7b, Decoding Strategy=Relaxed (temperature = 0.7), Decoding Structure=tree base2026.05 | 5.2409 | 4.52 | |
| CATSModel=Llama2-7b, Decoding Strategy=Relaxed (temperature = 0.7), Decoding Structure=chain base2026.05 | 5.2004 | 5.2 | |
| EAGLE-self, treeModel=Llama2-13b, Decoding Strategy=Relaxed (temperature = 0.7), Decoding Structure=tree base2026.05 | 5.1051 | 5.05 | |
| CATS w/ EAGLEModel=Llama2-7b, Decoding Strategy=Greedy, Decoding Structure=tree base2026.05 | 5.0634 | 5.06 | |
| EAGLE-selfModel=Llama2-7b, Decoding Strategy=Relaxed (temperature = 0.7), Decoding Structure=chain base2026.05 | 4.9756 | 4.29 | |
| KangarooModel=Llama2-7b, Decoding Strategy=Relaxed (temperature = 0.7), Decoding Structure=chain base2026.05 | 4.9609 | 4.28 | |
| CATSModel=Llama2-13b, Decoding Strategy=Relaxed (temperature = 0.7), Decoding Structure=chain base2026.05 | 4.9133 | 4.87 | |
| CATS w/ EAGLEModel=Llama2-13b, Decoding Strategy=Greedy, Decoding Structure=tree base2026.05 | 4.8191 | 4.75 | |
| EAGLE-self, treeModel=Llama2-7b, Decoding Strategy=Greedy, Decoding Structure=tree base2026.05 | 4.7199 | 4.07 | |
| EAGLE-selfModel=Llama2-13b, Decoding Strategy=Relaxed (temperature = 0.7), Decoding Structure=chain base2026.05 | 4.7032 | 4.68 | |
| KangarooModel=Llama2-13b, Decoding Strategy=Relaxed (temperature = 0.7), Decoding Structure=chain base2026.05 | 4.6991 | 4.68 | |
| CATSModel=Llama2-7b, Decoding Strategy=Greedy, Decoding Structure=chain base2026.05 | 4.6181 | 4.62 | |
| EAGLE-self, treeModel=Llama2-13b, Decoding Strategy=Greedy, Decoding Structure=tree base2026.05 | 4.4699 | 4.42 | |
| CATSModel=Llama2-13b, Decoding Strategy=Greedy, Decoding Structure=chain base2026.05 | 4.2752 | 4.24 | |
| CATS w/ EAGLEModel=Vicuna-13b, Decoding Strategy=Relaxed (temperature = 0.7), Decoding Structure=tree base2026.05 | 4.236 | 4.18 | |
| KangarooModel=Llama2-7b, Decoding Strategy=Greedy, Decoding Structure=chain base2026.05 | 4.193 | 3.61 | |
| EAGLE-selfModel=Llama2-7b, Decoding Strategy=Greedy, Decoding Structure=chain base2026.05 | 4.1813 | 3.6 | |
| EAGLE-selfModel=Llama2-13b, Decoding Strategy=Greedy, Decoding Structure=chain base2026.05 | 3.9754 | 3.96 | |
| KangarooModel=Llama2-13b, Decoding Strategy=Greedy, Decoding Structure=chain base2026.05 | 3.9369 | 3.92 | |
| CATS w/ EAGLEModel=Vicuna-13b, Decoding Strategy=Greedy, Decoding Structure=tree base2026.05 | 3.6897 | 3.64 | |
| EAGLE-self, treeModel=Vicuna-13b, Decoding Strategy=Relaxed (temperature = 0.7), Decoding Structure=tree base2026.05 | 3.5232 | 3.49 | |
| CATSModel=Vicuna-13b, Decoding Strategy=Relaxed (temperature = 0.7), Decoding Structure=chain base2026.05 | 3.2421 | 3.22 | |
| EAGLE-self, treeModel=Vicuna-13b, Decoding Strategy=Greedy, Decoding Structure=tree base2026.05 | 3.0423 | 3.01 | |
| EAGLE-selfModel=Vicuna-13b, Decoding Strategy=Relaxed (temperature = 0.7), Decoding Structure=chain base2026.05 | 2.8444 | 2.83 | |
| CATSModel=Vicuna-13b, Decoding Strategy=Greedy, Decoding Structure=chain base2026.05 | 2.7946 | 2.77 | |
| MedusaModel=Vicuna-13b, Decoding Strategy=Relaxed (temperature = 0.7), Decoding Structure=chain base2026.05 | 2.7905 | 2.37 | |
| KangarooModel=Vicuna-13b, Decoding Strategy=Relaxed (temperature = 0.7), Decoding Structure=chain base2026.05 | 2.7894 | 2.78 | |
| MedusaModel=Vicuna-13b, Decoding Strategy=Greedy, Decoding Structure=chain base2026.05 | 2.5833 | 2.19 | |
| KangarooModel=Vicuna-13b, Decoding Strategy=Greedy, Decoding Structure=chain base2026.05 | 2.4184 | 2.41 | |
| EAGLE-selfModel=Vicuna-13b, Decoding Strategy=Greedy, Decoding Structure=chain base2026.05 | 2.3831 | 2.37 | |
| MedusaModel=Llama2-13b, Decoding Strategy=Relaxed (temperature = 0.7), Decoding Structure=chain base2026.05 | 2.0646 | 1.75 | |
| MedusaModel=Llama2-13b, Decoding Strategy=Greedy, Decoding Structure=chain base2026.05 | 1.9737 | 1.68 | |
| MedusaModel=Llama2-7b, Decoding Strategy=Relaxed (temperature = 0.7), Decoding Structure=chain base2026.05 | 1.8667 | 1.87 | |
| MedusaModel=Llama2-7b, Decoding Strategy=Greedy, Decoding Structure=chain base2026.05 | 1.7817 | 1.78 |