Instruction Following on Alpaca
5.27Speedup (x)EAGLE3 + DM
Evaluation Results
| Method | Links | |||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| EAGLE3 + DMModel=Llama-3.3-70B-Instruct, Draft length (L)=92026.02 | 5.27 | 52 | — | — | — | — | — | — | 7.74 | — | — | — | — | — | — | — | — | — | — | |
| EAGLE3 + DMModel=Llama-3.3-70B-Instruct, Draft length (L)=72026.02 | 5.02 | 56 | — | — | — | — | — | — | 7.25 | — | — | — | — | — | — | — | — | — | — | |
| EAGLE3Model=Llama-3.3-70B-Instruct, Draft length (L)=92026.02 | 4.92 | 50.6 | — | — | — | — | — | — | 6.89 | — | — | — | — | — | — | — | — | — | — | |
| EAGLE3Model=Llama-3.3-70B-Instruct, Draft length (L)=72026.02 | 4.75 | 51.3 | — | — | — | — | — | — | 6.59 | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+LTDModel=Llama-3.1-8B-Instruct, Temperature=0, Decoding Mode=Greedy2026.03 | 4.53 | — | — | — | 7.31 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EAGLE3 + DMModel=Llama-3.3-70B-Instruct, Draft length (L)=52026.02 | 4.47 | 55 | — | — | — | — | — | — | 6.3 | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+LTDModel=Vicuna-13B-v1.3, Temperature=0, Decoding Mode=Greedy2026.03 | 4.4 | — | — | — | 7.01 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EAGLE3Model=Llama-3.3-70B-Instruct, Draft length (L)=52026.02 | 4.31 | 50 | — | — | — | — | — | — | 5.83 | — | — | — | — | — | — | — | — | — | — | |
| GraftModel=LLaMA3.1-8B2026.05 | 4.2 | — | — | — | — | 7.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EAGLE3Base Model=LLaMA3.3-70B2026.01 | 4.13 | — | — | — | — | — | 5.89 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ECHOModel=LLaMA3.1-8B2026.05 | 4.13 | — | — | — | — | 6.83 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EAGLE3-2KModel=LLaMA3.1-8B2026.05 | 4.06 | — | — | — | — | 7.09 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TALONModel=Llama3-8B, Temperature=02026.01 | 4.04 | — | — | — | — | 6.51 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DOUBLEBase Model=LLaMA3.3-70B2026.01 | 3.98 | — | — | — | — | — | 9.24 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+LTDModel=DeepSeek-R1-Distill-LLaMA 8B, Temperature=0, Decoding Mode=Greedy2026.03 | 3.98 | — | — | — | 6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TALONModel=Vicuna-13B, Temperature=02026.01 | 3.94 | — | — | — | — | 6.36 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EAGLE-3Model=Llama3-8B, Temperature=02026.01 | 3.92 | — | — | — | — | 6.93 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EAGLE3Base Model=LLaMA3.3-70B, Temperature=1, Decoding Setting=non-greedy2026.01 | 3.84 | — | — | — | — | — | 5.48 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DOUBLEBase Model=Deepseek-33B2026.01 | 3.84 | — | — | — | — | — | 10.24 | — | — | — | — | — | — | — | — | — | — | — | — | |
| GraftModel=Vicuna-13B2026.05 | 3.79 | — | — | — | — | 6.54 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EAGLE-3Model=Vicuna-13B, Temperature=02026.01 | 3.78 | — | — | — | — | 6.61 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| OPT-TreeModel=Vicuna-13B, Temperature=02026.01 | 3.71 | — | — | — | — | 6.56 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ECHOModel=Vicuna-13B2026.05 | 3.69 | — | — | — | — | 6.28 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EAGLE3-2KModel=Vicuna-13B2026.05 | 3.67 | — | — | — | — | 6.52 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+LTDModel=Vicuna-13B-v1.3, Temperature=12026.03 | 3.65 | — | — | — | — | — | — | — | 6.2 | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+GSModel=Vicuna-13B-v1.3, Temperature=12026.03 | 3.61 | — | — | — | — | — | — | — | 6.01 | — | — | — | — | — | — | — | — | — | — | |
| DDDModel=Vicuna-13B2026.05 | 3.57 | — | — | — | — | 5.96 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BASTIONModel=Qwen3-4B, Decoding Mode=Stochastic Decoding, Temperature=12026.05 | 3.56 | — | — | — | — | — | — | — | 4.7 | — | — | — | — | — | — | — | — | — | — | |
| BASTIONModel=Qwen3-8B, Decoding Mode=Greedy Decoding, Temperature=02026.05 | 3.54 | — | — | — | — | — | — | — | 4.68 | — | — | — | — | — | — | — | — | — | — | |
| BASTIONModel=Qwen3-4B, Decoding Mode=Greedy Decoding, Temperature=02026.05 | 3.53 | — | — | — | — | — | — | — | 4.64 | — | — | — | — | — | — | — | — | — | — | |
| DOUBLEBase Model=LLaMA3.3-70B, Temperature=1, Decoding Setting=non-greedy2026.01 | 3.51 | — | — | — | — | — | 8.15 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+GTModel=Llama-3.1-8B-Instruct, Temperature=12026.03 | 3.51 | — | — | — | — | — | — | — | 5.12 | — | — | — | — | — | — | — | — | — | — | |
| TALONModel=DSL-8B, Temperature=02026.01 | 3.46 | — | — | — | — | 5.18 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+GSModel=Llama-3.1-8B-Instruct, Temperature=12026.03 | 3.46 | — | — | — | — | — | — | — | 5.79 | — | — | — | — | — | — | — | — | — | — | |
| BASTIONModel=Qwen3-8B, Decoding Mode=Stochastic Decoding, Temperature=12026.05 | 3.46 | — | — | — | — | — | — | — | 4.64 | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+LTDModel=Llama-3.1-8B-Instruct, Temperature=12026.03 | 3.43 | — | — | — | — | — | — | — | 5.68 | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+DDDModel=Vicuna-13B-v1.3, Temperature=12026.03 | 3.43 | — | — | — | — | — | — | — | 5 | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+Spec++Model=Llama-3.1-8B-Instruct, Temperature=12026.03 | 3.42 | — | — | — | — | — | — | — | 5.51 | — | — | — | — | — | — | — | — | — | — | |
| Eagle3Model=Vicuna-13B-v1.3, Temperature=12026.03 | 3.4 | — | — | — | — | — | — | — | 5.49 | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+GTModel=Vicuna-13B-v1.3, Temperature=12026.03 | 3.39 | — | — | — | — | — | — | — | 4.95 | — | — | — | — | — | — | — | — | — | — | |
| Eagle3Model=Llama-3.1-8B-Instruct, Temperature=12026.03 | 3.38 | — | — | — | — | — | — | — | 5.47 | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+DDDModel=Llama-3.1-8B-Instruct, Temperature=12026.03 | 3.38 | — | — | — | — | — | — | — | 4.59 | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+DiscoModel=Vicuna-13B-v1.3, Temperature=12026.03 | 3.36 | — | — | — | — | — | — | — | 5.51 | — | — | — | — | — | — | — | — | — | — | |
| DFlash+DDTreeModel=Qwen3-8B, Temperature=0.0, Node Budget=Best from {16, 32, 64, 128, 256, 512, 1024}2026.04 | 3.36 | — | — | — | 5.09 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PEARLBase Model=LLaMA3.3-70B2026.01 | 3.35 | — | — | — | — | — | 8.31 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+Spec++Model=Vicuna-13B-v1.3, Temperature=12026.03 | 3.34 | — | — | — | — | — | — | — | 5.39 | — | — | — | — | — | — | — | — | — | — | |
| DFlash+DDTreeModel=Qwen3-4B, Temperature=0.0, Node Budget=Best from {16, 32, 64, 128, 256, 512, 1024}2026.04 | 3.32 | — | — | — | 5.35 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+DiscoModel=Llama-3.1-8B-Instruct, Temperature=12026.03 | 3.3 | — | — | — | — | — | — | — | 5.18 | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+C2TModel=Vicuna-13B-v1.3, Temperature=12026.03 | 3.27 | — | — | — | — | — | — | — | 5.54 | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+C2TModel=Llama-3.1-8B-Instruct, Temperature=12026.03 | 3.24 | — | — | — | — | — | — | — | 5.47 | — | — | — | — | — | — | — | — | — | — | |
| EAGLE-3Model=DSL-8B, Temperature=02026.01 | 3.23 | — | — | — | — | 5.64 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DFlash+DDTreeModel=Qwen3-4B, Temperature=1.0, Node Budget=Best from {16, 32, 64, 128, 256, 512, 1024}2026.04 | 3.23 | — | — | — | 4.97 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+SvipModel=Vicuna-13B-v1.3, Temperature=12026.03 | 3.21 | — | — | — | — | — | — | — | 5.39 | — | — | — | — | — | — | — | — | — | — | |
| DFlash+DDTreeModel=Qwen3-8B, Temperature=1.0, Node Budget=Best from {16, 32, 64, 128, 256, 512, 1024}2026.04 | 3.2 | — | — | — | 4.89 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+SvipModel=Llama-3.1-8B-Instruct, Temperature=12026.03 | 3.19 | — | — | — | — | — | — | — | 4.95 | — | — | — | — | — | — | — | — | — | — | |
| PEARLBase Model=LLaMA3.3-70B, Temperature=1, Decoding Setting=non-greedy2026.01 | 3.15 | — | — | — | — | — | 7.82 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DOUBLEBase Model=LLaMA2-70B2026.01 | 2.94 | — | — | — | — | — | 8.24 | — | — | — | — | — | — | — | — | — | — | — | — | |
| PEARLBase Model=Deepseek-33B2026.01 | 2.86 | — | — | — | — | — | 8.71 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+LTDModel=DeepSeek-R1-Distill-LLaMA 8B, Temperature=12026.03 | 2.83 | — | — | — | — | — | — | — | 4.76 | — | — | — | — | — | — | — | — | — | — | |
| DOMINOTemperature=0, Model=Qwen3-4B, Draft size=162026.05 | 2.8 | — | — | — | — | — | — | — | 3.88 | — | — | — | — | — | — | — | — | — | — | |
| Eagle3Model=DeepSeek-R1-Distill-LLaMA 8B, Temperature=12026.03 | 2.79 | — | — | — | — | — | — | — | 4.53 | — | — | — | — | — | — | — | — | — | — | |
| DOMINOTemperature=0, Model=Qwen3-8B, Draft size=162026.05 | 2.78 | — | — | — | — | — | — | — | 3.87 | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+GSModel=DeepSeek-R1-Distill-LLaMA 8B, Temperature=12026.03 | 2.76 | — | — | — | — | — | — | — | 4.83 | — | — | — | — | — | — | — | — | — | — | |
| No FilterData retain ratio=100%, Draft generating step (gamma)=5, Decoding temperature=12026.01 | 2.71 | — | — | — | 2.89 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SFDDData retain ratio=50%, Draft generating step (gamma)=5, Decoding temperature=12026.01 | 2.66 | — | — | — | 2.71 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PEARLBase Model=LLaMA2-70B2026.01 | 2.66 | — | — | — | — | — | 7.71 | — | — | — | — | — | — | — | — | — | — | — | — | |
| OuroborosBase Model=Deepseek-33B2026.01 | 2.64 | — | — | — | — | — | 5.3 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DOMINOTemperature=1, Model=Qwen3-4B, Draft size=162026.05 | 2.59 | — | — | — | — | — | — | — | 3.52 | — | — | — | — | — | — | — | — | — | — | |
| DOMINOTemperature=1, Model=Qwen3-8B, Draft size=162026.05 | 2.52 | — | — | — | — | — | — | — | 3.46 | — | — | — | — | — | — | — | — | — | — | |
| DFlash+DDTreeModel=Qwen3-Coder-30B-A3B-Instruct, Temperature=0.0, Node Budget=Best from {16, 32, 64, 128, 256, 512, 1024}2026.04 | 2.46 | — | — | — | 3.55 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PPLData retain ratio=50%, Draft generating step (gamma)=5, Decoding temperature=12026.01 | 2.45 | — | — | — | 2.65 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EAGLE-3Model=Qwen3-4B, Decoding Mode=Greedy Decoding, Temperature=02026.05 | 2.45 | — | — | — | — | — | — | — | 2.15 | — | — | — | — | — | — | — | — | — | — | |
| Top-1 ProbabilityData retain ratio=50%, Draft generating step (gamma)=5, Decoding temperature=12026.01 | 2.44 | — | — | — | 2.66 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Energy ScoreData retain ratio=50%, Draft generating step (gamma)=5, Decoding temperature=12026.01 | 2.44 | — | — | — | 2.64 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TALONModel=Qwen3-8B, Temperature=02026.01 | 2.44 | — | — | — | — | 3.39 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EntropyData retain ratio=50%, Draft generating step (gamma)=5, Decoding temperature=12026.01 | 2.43 | — | — | — | 2.64 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DFlash+DDTreeModel=Qwen3-Coder-30B-A3B-Instruct, Temperature=1.0, Node Budget=Best from {16, 32, 64, 128, 256, 512, 1024}2026.04 | 2.43 | — | — | — | 3.51 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GraftModel=Qwen3-32B2026.05 | 2.41 | — | — | — | — | 2.93 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EAGLE-3Model=Qwen3-4B, Decoding Mode=Stochastic Decoding, Temperature=12026.05 | 2.41 | — | — | — | — | — | — | — | 2.16 | — | — | — | — | — | — | — | — | — | — | |
| HYDRAModel=Vicuna-13B, Temperature=02026.01 | 2.4 | — | — | — | — | 3.51 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+LTDModel=Qwen3-32B, Temperature=0, Decoding Mode=Greedy2026.03 | 2.38 | — | — | — | 2.85 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RandomData retain ratio=50%, Draft generating step (gamma)=5, Decoding temperature=12026.01 | 2.37 | — | — | — | 2.59 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MarginData retain ratio=50%, Draft generating step (gamma)=5, Decoding temperature=12026.01 | 2.35 | — | — | — | 2.48 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Graft-DFLASH(16)Base Model=Qwen3-8B, Block Size=162026.05 | 2.33 | — | — | — | — | 3.34 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GraftModel=Qwen3-8B2026.05 | 2.29 | — | — | — | — | 3.56 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DFlashModel=Qwen3-8B, Decoding Mode=Greedy Decoding, Temperature=02026.05 | 2.28 | — | — | — | — | — | — | — | 3.15 | — | — | — | — | — | — | — | — | — | — | |
| ECHOModel=Qwen3-32B2026.05 | 2.25 | — | — | — | — | 2.76 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+LTDModel=Qwen3-14B, Temperature=0, Decoding Mode=Greedy2026.03 | 2.24 | — | — | — | 2.87 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpsBase Model=LLaMA3.3-70B2026.01 | 2.23 | — | — | — | — | — | 4.61 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DFlashTemperature=0, Model=Qwen3-8B, Draft size=162026.05 | 2.23 | — | — | — | — | — | — | — | 3.1 | — | — | — | — | — | — | — | — | — | — | |
| DFlashTemperature=0, Model=Qwen3-4B, Draft size=162026.05 | 2.21 | — | — | — | — | — | — | — | 3.12 | — | — | — | — | — | — | — | — | — | — | |
| DARTTemperature=0, Model=Qwen3-8B, Draft size=602026.05 | 2.21 | — | — | — | — | — | — | — | 2.84 | — | — | — | — | — | — | — | — | — | — | |
| ECHOModel=Qwen3-8B2026.05 | 2.19 | — | — | — | — | 3.32 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DFlashTemperature=1, Model=Qwen3-4B, Draft size=162026.05 | 2.18 | — | — | — | — | — | — | — | 3.03 | — | — | — | — | — | — | — | — | — | — | |
| DARTTemperature=1, Model=Qwen3-8B, Draft size=602026.05 | 2.18 | — | — | — | — | — | — | — | 2.84 | — | — | — | — | — | — | — | — | — | — | |
| Graft-EAGLE3(60)Base Model=Qwen3-8B, Budget=602026.05 | 2.17 | — | — | — | — | 3.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpsBase Model=Deepseek-33B2026.01 | 2.16 | — | — | — | — | — | 4.22 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Eagle3+LTDModel=Qwen3 32B, Temperature=12026.03 | 2.16 | — | — | — | — | — | — | — | 2.59 | — | — | — | — | — | — | — | — | — | — | |
| DARTTemperature=0, Model=Qwen3-4B, Draft size=602026.05 | 2.16 | — | — | — | — | — | — | — | 2.78 | — | — | — | — | — | — | — | — | — | — | |
| DARTTemperature=1, Model=Qwen3-4B, Draft size=602026.05 | 2.16 | — | — | — | — | — | — | — | 2.78 | — | — | — | — | — | — | — | — | — | — |