List your company

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

62AccuracyFaster Cascades
48.4851.9955.559.01Jun 10, 2026

Evaluation results

MethodLinks
Faster CascadesModel=Qwen 7B→72B2026.0662212.25
VIA-SDModel=Qwen 7B→72B2026.0661172.55
Speculative DecodingModel=Qwen 7B→72B2026.0660261.7
BiLDModel=Qwen 7B→72B2026.0660242.1
Cascade SDModel=Qwen 7B→72B2026.0660251.9
SWiFTModel=Qwen 7B→72B2026.0660232
CLaSPModel=Qwen 7B→72B2026.0660222.1
VIA-SDModel=Qwen 7B→14B2026.0657152.4
Faster CascadesModel=Gemma2 2B→27B2026.0656202.3
Faster CascadesModel=LLaMA2 7B→70B2026.0656222.15
Faster CascadesModel=Qwen 7B→14B2026.0656191.95
VIA-SDModel=Gemma2 2B→27B2026.0655152.5
VIA-SDModel=LLaMA2 7B→70B2026.0655182.45
Speculative DecodingModel=Gemma2 2B→27B2026.0654241.65
BiLDModel=Gemma2 2B→27B2026.0654222.15
Cascade SDModel=Gemma2 2B→27B2026.0654231.81
SWiFTModel=Gemma2 2B→27B2026.0654231.82
CLaSPModel=Gemma2 2B→27B2026.0654221.99
Speculative DecodingModel=LLaMA2 7B→70B2026.0654271.6
BiLDModel=LLaMA2 7B→70B2026.0654252.05
Cascade SDModel=LLaMA2 7B→70B2026.0654261.8
SWiFTModel=LLaMA2 7B→70B2026.0654241.9
CLaSPModel=LLaMA2 7B→70B2026.0654232
Speculative DecodingModel=Qwen 7B→14B2026.0654241.58
BiLDModel=Qwen 7B→14B2026.0654221.85
Cascade SDModel=Qwen 7B→14B2026.0654231.7
SWiFTModel=Qwen 7B→14B2026.0654211.85
CLaSPModel=Qwen 7B→14B2026.0654202
VIA-SDModel=Gemma2 2B→9B2026.0653132.33
Faster CascadesModel=Gemma2 2B→9B2026.0652171.9
VIA-SDModel=LLaMA2 7B→13B2026.0652162.2
Faster CascadesModel=LLaMA2 7B→13B2026.0651201.82
Speculative DecodingModel=Gemma2 2B→9B2026.0650211.55
BiLDModel=Gemma2 2B→9B2026.0650191.85
Cascade SDModel=Gemma2 2B→9B2026.0650201.62
SWiFTModel=Gemma2 2B→9B2026.0650181.73
CLaSPModel=Gemma2 2B→9B2026.0650171.9
Speculative DecodingModel=LLaMA2 7B→13B2026.0649261.48
BiLDModel=LLaMA2 7B→13B2026.0649231.75
Cascade SDModel=LLaMA2 7B→13B2026.0649241.6
SWiFTModel=LLaMA2 7B→13B2026.0649221.7
CLaSPModel=LLaMA2 7B→13B2026.0649211.85