List your company

Speculative Decoding on VideoDetailCaption (~17k visual tokens)

3.91Tau (τ)SpecVLM
0.98761.74632.5053.2637Feb 17, 2026

Evaluation results

MethodLinks
SpecVLMTemperature=0, Target Model=Qwen2.5-VL-7B, Hardware=NVIDIA L20 GPUs, Batch Size=162026.023.911.2
SparrowTemperature=0, Target Model=Qwen2.5-VL-7B, Hardware=NVIDIA L20 GPUs, Batch Size=162026.023.892.04
ViSpecTemperature=0, Target Model=Qwen2.5-VL-7B, Hardware=NVIDIA L20 GPUs, Batch Size=162026.023.521.87
ViSpecTemperature=0, Target Model=Qwen2.5-VL-7B, Hardware=NVIDIA L20 GPUs, Batch Size=8, Source=Original Weights2026.022.791.53
SparrowTemperature=1, Target Model=Qwen2.5-VL-7B, Hardware=NVIDIA L20 GPUs, Batch Size=162026.022.371.27
ViSpecTemperature=1, Target Model=Qwen2.5-VL-7B, Hardware=NVIDIA L20 GPUs, Batch Size=162026.022.241.21
MSDTemperature=0, Target Model=Qwen2.5-VL-7B, Hardware=NVIDIA L20 GPUs, Batch Size=162026.021.110.52
MSDTemperature=1, Target Model=Qwen2.5-VL-7B, Hardware=NVIDIA L20 GPUs, Batch Size=162026.021.10.48