Question Answering on TriviaQA (Accuracy, Rejection Rate, Speedup)
62AccuracyFaster Cascades
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Faster CascadesModel=Qwen 7B→72B2026.06 | 62 | 21 | 2.25 | |
| VIA-SDModel=Qwen 7B→72B2026.06 | 61 | 17 | 2.55 | |
| Speculative DecodingModel=Qwen 7B→72B2026.06 | 60 | 26 | 1.7 | |
| BiLDModel=Qwen 7B→72B2026.06 | 60 | 24 | 2.1 | |
| Cascade SDModel=Qwen 7B→72B2026.06 | 60 | 25 | 1.9 | |
| SWiFTModel=Qwen 7B→72B2026.06 | 60 | 23 | 2 | |
| CLaSPModel=Qwen 7B→72B2026.06 | 60 | 22 | 2.1 | |
| VIA-SDModel=Qwen 7B→14B2026.06 | 57 | 15 | 2.4 | |
| Faster CascadesModel=Gemma2 2B→27B2026.06 | 56 | 20 | 2.3 | |
| Faster CascadesModel=LLaMA2 7B→70B2026.06 | 56 | 22 | 2.15 | |
| Faster CascadesModel=Qwen 7B→14B2026.06 | 56 | 19 | 1.95 | |
| VIA-SDModel=Gemma2 2B→27B2026.06 | 55 | 15 | 2.5 | |
| VIA-SDModel=LLaMA2 7B→70B2026.06 | 55 | 18 | 2.45 | |
| Speculative DecodingModel=Gemma2 2B→27B2026.06 | 54 | 24 | 1.65 | |
| BiLDModel=Gemma2 2B→27B2026.06 | 54 | 22 | 2.15 | |
| Cascade SDModel=Gemma2 2B→27B2026.06 | 54 | 23 | 1.81 | |
| SWiFTModel=Gemma2 2B→27B2026.06 | 54 | 23 | 1.82 | |
| CLaSPModel=Gemma2 2B→27B2026.06 | 54 | 22 | 1.99 | |
| Speculative DecodingModel=LLaMA2 7B→70B2026.06 | 54 | 27 | 1.6 | |
| BiLDModel=LLaMA2 7B→70B2026.06 | 54 | 25 | 2.05 | |
| Cascade SDModel=LLaMA2 7B→70B2026.06 | 54 | 26 | 1.8 | |
| SWiFTModel=LLaMA2 7B→70B2026.06 | 54 | 24 | 1.9 | |
| CLaSPModel=LLaMA2 7B→70B2026.06 | 54 | 23 | 2 | |
| Speculative DecodingModel=Qwen 7B→14B2026.06 | 54 | 24 | 1.58 | |
| BiLDModel=Qwen 7B→14B2026.06 | 54 | 22 | 1.85 | |
| Cascade SDModel=Qwen 7B→14B2026.06 | 54 | 23 | 1.7 | |
| SWiFTModel=Qwen 7B→14B2026.06 | 54 | 21 | 1.85 | |
| CLaSPModel=Qwen 7B→14B2026.06 | 54 | 20 | 2 | |
| VIA-SDModel=Gemma2 2B→9B2026.06 | 53 | 13 | 2.33 | |
| Faster CascadesModel=Gemma2 2B→9B2026.06 | 52 | 17 | 1.9 | |
| VIA-SDModel=LLaMA2 7B→13B2026.06 | 52 | 16 | 2.2 | |
| Faster CascadesModel=LLaMA2 7B→13B2026.06 | 51 | 20 | 1.82 | |
| Speculative DecodingModel=Gemma2 2B→9B2026.06 | 50 | 21 | 1.55 | |
| BiLDModel=Gemma2 2B→9B2026.06 | 50 | 19 | 1.85 | |
| Cascade SDModel=Gemma2 2B→9B2026.06 | 50 | 20 | 1.62 | |
| SWiFTModel=Gemma2 2B→9B2026.06 | 50 | 18 | 1.73 | |
| CLaSPModel=Gemma2 2B→9B2026.06 | 50 | 17 | 1.9 | |
| Speculative DecodingModel=LLaMA2 7B→13B2026.06 | 49 | 26 | 1.48 | |
| BiLDModel=LLaMA2 7B→13B2026.06 | 49 | 23 | 1.75 | |
| Cascade SDModel=LLaMA2 7B→13B2026.06 | 49 | 24 | 1.6 | |
| SWiFTModel=LLaMA2 7B→13B2026.06 | 49 | 22 | 1.7 | |
| CLaSPModel=LLaMA2 7B→13B2026.06 | 49 | 21 | 1.85 |