List your company

Speculative Decoding on HumanEval

6.99Speedup FactorTAPS
1.48842.91674.3455.7733May 30, 2026May 31, 2026Jun 1, 2026Jun 2, 2026

Evaluation results

MethodLinks
TAPSHardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-4B2026.056.99—8.61———
TreeFlashTarget model=Qwen3 4B, Temperature (T)=0, Draft budget (B)=642026.066.85————9.51
TreeFlashTarget model=Qwen3 8B, Temperature (T)=0, Draft budget (B)=642026.066.79————9.4
TAPSHardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-8B2026.056.47—8.44———
TAPSHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-4B2026.056.41—8.46———
DDTreeTarget model=Qwen3 4B, Temperature (T)=0, Draft budget (B)=642026.066.36————8.55
TreeFlashTarget model=Qwen3 4B, Temperature (T)=1, Draft budget (B)=642026.066.23————8.67
DDTreeTarget model=Qwen3 8B, Temperature (T)=0, Draft budget (B)=642026.066.22————8.35
TAPSHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-8B2026.056.05—8.52———
TreeFlashTarget model=Qwen3 8B, Temperature (T)=1, Draft budget (B)=642026.065.88————8.13
DDTreeTarget model=Qwen3 4B, Temperature (T)=1, Draft budget (B)=642026.065.81————7.82
DDTreeHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-4B2026.055.56—9.44———
TreeFlashTarget model=Qwen3 4B, Temperature (T)=0, Draft budget (B)=162026.065.5————7.83
DDTreeTarget model=Qwen3 8B, Temperature (T)=1, Draft budget (B)=642026.065.4————7.26
TreeFlashTarget model=Qwen3 8B, Temperature (T)=0, Draft budget (B)=162026.065.4————7.74
DDTreeTarget model=Qwen3 4B, Temperature (T)=0, Draft budget (B)=162026.065.39————7.43
DDTreeTarget model=Qwen3 8B, Temperature (T)=0, Draft budget (B)=162026.065.25————7.2
DFlashHardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-4B2026.055.21—6.59———
DFlashHardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-8B2026.055.14—6.41———
TAPSHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.055.02—6.59———
TreeFlashTarget model=Qwen3 4B, Temperature (T)=1, Draft budget (B)=162026.064.99————7.13
DFlashTarget model=Qwen3 4B, Temperature (T)=0, Draft budget (B)=162026.064.87————6.44
DDTreeTarget model=Qwen3 4B, Temperature (T)=1, Draft budget (B)=162026.064.82————6.68
DFlashTarget model=Qwen3 8B, Temperature (T)=0, Draft budget (B)=162026.064.78————6.34
DFlashHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-4B2026.054.74—6.04———
TreeFlashTarget model=Qwen3 8B, Temperature (T)=1, Draft budget (B)=162026.064.63————6.63
TAPSHardware=NVIDIA A40-48GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.054.46—6.55———
DDTreeTarget model=Qwen3 8B, Temperature (T)=1, Draft budget (B)=162026.064.46————6.16
DFlashHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-8B2026.054.32—5.52———
DFlashTarget model=Qwen3 4B, Temperature (T)=1, Draft budget (B)=162026.064.24————5.66
DFlashTarget model=Qwen3 8B, Temperature (T)=1, Draft budget (B)=162026.063.95————5.26
DDTreeHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-8B2026.053.65—9.61———
DDTreeHardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-4B2026.053.58—9.65———
DFlashHardware=NVIDIA A40-48GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.053.51—4.88———
DFlashHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.053.48—4.9———
DDTreeHardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-8B2026.052.76—9.5———
DDTreeHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.052.74—7.28———
EAGLE3Target model=Qwen3 8B, Temperature (T)=0, Draft budget (B)=642026.062.16————3.7
EAGLE-3Hardware=NVIDIA A40-48GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.052.04—2.83———
EAGLE3Target model=Qwen3 8B, Temperature (T)=1, Draft budget (B)=642026.062.03————3.55
EAGLE3Target model=Qwen3 4B, Temperature (T)=0, Draft budget (B)=642026.062————3.47
EAGLE3Target model=Qwen3 4B, Temperature (T)=1, Draft budget (B)=642026.061.91————3.39
EAGLE-3Hardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-8B2026.051.89—3.17———
EAGLE3Target model=Qwen3 8B, Temperature (T)=0, Draft budget (B)=162026.061.87————3.18
EAGLE-3Hardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-4B2026.051.84—3.05———
DDTreeHardware=NVIDIA A40-48GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.051.81—7.26———
EAGLE3Target model=Qwen3 8B, Temperature (T)=1, Draft budget (B)=162026.061.78————3.06
EAGLE-3Hardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-8B2026.051.75—3.05———
EAGLE3Target model=Qwen3 4B, Temperature (T)=0, Draft budget (B)=162026.061.75————3.05
EAGLE-3Hardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-4B2026.051.74—3.01———
EAGLE-3Hardware=NVIDIA A800-SXM4-80GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.051.74—2.44———
EAGLE3Target model=Qwen3 4B, Temperature (T)=1, Draft budget (B)=162026.061.7————2.96
CORAL2026.04———149.82——
EAGLE-32026.04———136.95——
EAGLE-3 KLTarget Model=LLaMA 3.1 8B Instruct, Temperature=02026.02————4.82—
EAGLE-3 KLTarget Model=LLaMA 3.1 8B Instruct, Temperature=12026.02————4.31—
EAGLE-3 KLTarget Model=LLaMA 3.3 70B Instruct, Temperature=02026.02————5.18—
EAGLE-3 KLTarget Model=LLaMA 3.3 70B Instruct, Temperature=12026.02————4.86—
EAGLE-3 KLTarget Model=GPT-OSS 20B, Temperature=02026.02————3.07—
EAGLE-3 KLTarget Model=GPT-OSS 20B, Temperature=12026.02————2.89—
EAGLE-3 KLTarget Model=GPT-OSS 120B, Temperature=02026.02————2.47—
EAGLE-3 KLTarget Model=GPT-OSS 120B, Temperature=12026.02————2.27—
EAGLE-3 KLTarget Model=Qwen3-235B-A22B-Instruct-2507, Temperature=02026.02————4.65—
EAGLE-3 KLTarget Model=Qwen3-235B-A22B-Instruct-2507, Temperature=12026.02————4.09—
EAGLE-3 Lλ LKTarget Model=LLaMA 3.1 8B Instruct, Temperature=02026.02————4.89—
EAGLE-3 Lλ LKTarget Model=LLaMA 3.1 8B Instruct, Temperature=12026.02————4.52—
EAGLE-3 Lλ LKTarget Model=LLaMA 3.3 70B Instruct, Temperature=02026.02————5.21—
EAGLE-3 Lλ LKTarget Model=LLaMA 3.3 70B Instruct, Temperature=12026.02————5.08—
EAGLE-3 Lλ LKTarget Model=GPT-OSS 20B, Temperature=02026.02————3.11—
EAGLE-3 Lλ LKTarget Model=GPT-OSS 20B, Temperature=12026.02————3.01—
EAGLE-3 Lλ LKTarget Model=GPT-OSS 120B, Temperature=02026.02————2.51—
EAGLE-3 Lλ LKTarget Model=GPT-OSS 120B, Temperature=12026.02————2.44—
EAGLE-3 Lλ LKTarget Model=Qwen3-235B-A22B-Instruct-2507, Temperature=02026.02————4.74—
EAGLE-3 Lλ LKTarget Model=Qwen3-235B-A22B-Instruct-2507, Temperature=12026.02————4.42—
EAGLE2model=Qwen2.5-Instruct-14B, Temperature=02025.03—2.843.1———
EAGLE2model=LLaMA3-Instruct-8B, Temperature=02025.03—2.763.37———
EAGLE2model=Qwen2.5-Instruct-14B, Temperature=12025.03—2.733.09———
EAGLE2model=LLaMA3-Instruct-8B, Temperature=12025.03—2.573.38———
EvoSpecStatic Vocabulary Size=8k, Dynamic Expansion Size=2562026.04———141.29——
EvoSpecStatic Vocabulary Size=8k, Dynamic Expansion Size=5122026.04———141.32——
EvoSpecStatic Vocabulary Size=8k, Dynamic Expansion Size=10242026.04———142.41——
EvoSpecStatic Vocabulary Size=16k, Dynamic Expansion Size=2562026.04———156.5——
EvoSpecStatic Vocabulary Size=16k, Dynamic Expansion Size=5122026.04———156.39——
EvoSpecStatic Vocabulary Size=16k, Dynamic Expansion Size=10242026.04———156.31——
EvoSpecStatic Vocabulary Size=32k, Dynamic Expansion Size=2562026.04———154.39——
EvoSpecStatic Vocabulary Size=32k, Dynamic Expansion Size=5122026.04———153.23——
EvoSpecStatic Vocabulary Size=32k, Dynamic Expansion Size=10242026.04———153.11——
FR-SpecStatic Vocabulary Size=32k2026.04———137.06——
MTP KLTarget Model=DeepSeek-V3-0324 (685B), Temperature=02026.02————4.74—
MTP KLTarget Model=DeepSeek-V3-0324 (685B), Temperature=12026.02————4.39—
MTP Lλ LKTarget Model=DeepSeek-V3-0324 (685B), Temperature=02026.02————4.77—
MTP Lλ LKTarget Model=DeepSeek-V3-0324 (685B), Temperature=12026.02————4.64—
MTP originalTarget Model=DeepSeek-V3-0324 (685B), Temperature=02026.02————3.28—
MTP originalTarget Model=DeepSeek-V3-0324 (685B), Temperature=12026.02————3.17—
PLDmodel=Qwen2.5-Instruct-14B, Temperature=02025.03—1.490.55———
PLDmodel=LLaMA3-Instruct-8B, Temperature=02025.03—1.380.51———
PLDmodel=Qwen2.5-Instruct-14B, Temperature=12025.03—1.530.58———
PLDmodel=LLaMA3-Instruct-8B, Temperature=12025.03—1.320.45———
RASD(PLD)model=Qwen2.5-Instruct-14B, Temperature=02025.03—2.983.44———
RASD(PLD)model=LLaMA3-Instruct-8B, Temperature=02025.03—2.793.55———