Inference Efficiency on TQA
0.05Relative Execution Time OverheadPerplexity
Evaluation Results
| Method | Links | |
|---|---|---|
| PerplexityModel=Qwen2.5-14B2026.03 | 0.05 | |
| PerplexityModel=Llama-2-7B2026.03 | 0.066 | |
| PerplexityModel=Qwen2.5-7B2026.03 | 0.094 | |
| PerplexityModel=Qwen2.5-3B2026.03 | 0.096 | |
| PerplexityModel=Mistral-7B2026.03 | 0.102 | |
| PerplexityModel=Llama-3-8B2026.03 | 0.13 | |
| STCModel=Llama-2-7B2026.03 | 2.329 | |
| STCModel=Mistral-7B2026.03 | 2.908 | |
| STCModel=Qwen2.5-14B2026.03 | 5.866 | |
| STCModel=Llama-3-8B2026.03 | 6.833 | |
| STCModel=Qwen2.5-3B2026.03 | 6.946 | |
| STCModel=Qwen2.5-7B2026.03 | 7.268 |