List your company

Question Answering on WebQuestions (Accuracy, Rejection Rate, Speedup)

37AccuracyFaster Cascades
25.5628.5331.534.47Jun 10, 2026

Evaluation results

MethodLinks
Faster CascadesModel=Qwen 7B→72B2026.0637222
VIA-SDModel=Qwen 7B→72B2026.0637152.45
Faster CascadesModel=Qwen 7B→72B, Method=Faster Cascades, Speedup=2.00×2026.063722—
VIA-SDModel=Qwen 7B→72B, Method=VIA-SD (ours), Speedup=2.45×2026.063715—
Speculative DecodingModel=Qwen 7B→72B2026.0636281.6
BiLDModel=Qwen 7B→72B2026.0636261.75
Cascade SDModel=Qwen 7B→72B2026.0636251.85
SWiFTModel=Qwen 7B→72B2026.0636212.05
CLaSPModel=Qwen 7B→72B2026.0636202.15
Speculative DecodingModel=Qwen 7B→72B, Method=Speculative Decoding, Speedup=1.60×2026.063628—
Cascade SDModel=Qwen 7B→72B, Method=Cascade SD, Speedup=1.85×2026.063625—
Faster CascadesModel=Gemma2 2B→27B2026.0633221.81
VIA-SDModel=Qwen 7B→14B2026.0633112.05
Faster CascadesModel=Gemma2 2B→27B, Method=Faster Cascades, Speedup=1.81×2026.063322—
VIA-SDModel=Qwen 7B→14B, Method=VIA-SD (ours), Speedup=2.05×2026.063311—
Speculative DecodingModel=Gemma2 2B→27B2026.0632271.55
BiLDModel=Gemma2 2B→27B2026.0632261.63
Cascade SDModel=Gemma2 2B→27B2026.0632241.71
SWiFTModel=Gemma2 2B→27B2026.0632211.89
CLaSPModel=Gemma2 2B→27B2026.0632201.97
VIA-SDModel=Gemma2 2B→27B2026.0632142.32
Faster CascadesModel=LLaMA2 7B→70B2026.0632241.85
Faster CascadesModel=Qwen 7B→14B2026.0632151.72
Speculative DecodingModel=Gemma2 2B→27B, Method=Speculative Decoding, Speedup=1.55×2026.063227—
Cascade SDModel=Gemma2 2B→27B, Method=Cascade SD, Speedup=1.71×2026.063224—
VIA-SDModel=Gemma2 2B→27B, Method=VIA-SD (ours), Speedup=2.32×2026.063214—
Faster CascadesModel=LLaMA2 7B→70B, Method=Faster Cascades, Speedup=1.85×2026.063224—
Faster CascadesModel=Qwen 7B→14B, Method=Faster Cascades, Speedup=1.72×2026.063215—
Speculative DecodingModel=LLaMA2 7B→70B2026.0631301.5
BiLDModel=LLaMA2 7B→70B2026.0631281.65
Cascade SDModel=LLaMA2 7B→70B2026.0631271.72
SWiFTModel=LLaMA2 7B→70B2026.0631231.92
CLaSPModel=LLaMA2 7B→70B2026.0631222
VIA-SDModel=LLaMA2 7B→70B2026.0631162.3
Speculative DecodingModel=Qwen 7B→14B2026.0631201.5
BiLDModel=Qwen 7B→14B2026.0631181.65
Cascade SDModel=Qwen 7B→14B2026.0631191.58
SWiFTModel=Qwen 7B→14B2026.0631161.7
CLaSPModel=Qwen 7B→14B2026.0631151.78
Speculative DecodingModel=LLaMA2 7B→70B, Method=Speculative Decoding, Speedup=1.50×2026.063130—
Cascade SDModel=LLaMA2 7B→70B, Method=Cascade SD, Speedup=1.72×2026.063127—
VIA-SDModel=LLaMA2 7B→70B, Method=VIA-SD (ours), Speedup=2.30×2026.063116—
Speculative DecodingModel=Qwen 7B→14B, Method=Speculative Decoding, Speedup=1.50×2026.063120—
Cascade SDModel=Qwen 7B→14B, Method=Cascade SD, Speedup=1.58×2026.063119—
Faster CascadesModel=Gemma2 2B→9B2026.0628131.65
VIA-SDModel=Gemma2 2B→9B2026.0628101.82
Faster CascadesModel=Gemma2 2B→9B, Method=Faster Cascades, Speedup=1.65×2026.062813—
VIA-SDModel=Gemma2 2B→9B, Method=VIA-SD (ours), Speedup=1.82×2026.062810—
Speculative DecodingModel=Gemma2 2B→9B2026.0627171.5
BiLDModel=Gemma2 2B→9B2026.0627151.59
Cascade SDModel=Gemma2 2B→9B2026.0627161.53
SWiFTModel=Gemma2 2B→9B2026.0627151.59
CLaSPModel=Gemma2 2B→9B2026.0627141.62
Faster CascadesModel=LLaMA2 7B→13B2026.0627171.62
VIA-SDModel=LLaMA2 7B→13B2026.0627131.88
Speculative DecodingModel=Gemma2 2B→9B, Method=Speculative Decoding, Speedup=1.50×2026.062717—
Cascade SDModel=Gemma2 2B→9B, Method=Cascade SD, Speedup=1.53×2026.062716—
Faster CascadesModel=LLaMA2 7B→13B, Method=Faster Cascades, Speedup=1.62×2026.062717—
VIA-SDModel=LLaMA2 7B→13B, Method=VIA-SD (ours), Speedup=1.88×2026.062713—
Speculative DecodingModel=LLaMA2 7B→13B2026.0626221.42
BiLDModel=LLaMA2 7B→13B2026.0626201.55
Cascade SDModel=LLaMA2 7B→13B2026.0626211.48
SWiFTModel=LLaMA2 7B→13B2026.0626191.58
CLaSPModel=LLaMA2 7B→13B2026.0626181.65
Speculative DecodingModel=LLaMA2 7B→13B, Method=Speculative Decoding, Speedup=1.42×2026.062622—
Cascade SDModel=LLaMA2 7B→13B, Method=Cascade SD, Speedup=1.48×2026.062621—