Inference Latency (s) on GPQA
2.7Inference Latency (s)T2
Evaluation Results
| Method | Links | |
|---|---|---|
| T2Backbone=Llama3.1-8B2026.01 | 2.7 | |
| T2Backbone=Qwen2.5-7B2026.01 | 4.7 | |
| Std-PTBackbone=Llama3.1-8B2026.01 | 7.2 | |
| Std-PTBackbone=Qwen2.5-7B2026.01 | 9.3 | |
| DSCBackbone=Llama3.1-8B2026.01 | 15 | |
| T2Backbone=Qwen3-4B2026.01 | 15.3 | |
| ESCBackbone=Llama3.1-8B2026.01 | 17.1 | |
| DSCBackbone=Qwen2.5-7B2026.01 | 20.3 | |
| ESCBackbone=Qwen2.5-7B2026.01 | 21.1 | |
| ACBackbone=Llama3.1-8B2026.01 | 21.5 | |
| ACBackbone=Qwen2.5-7B2026.01 | 25.1 | |
| DeepConfBackbone=Llama3.1-8B2026.01 | 30 | |
| DeepConfBackbone=Qwen2.5-7B2026.01 | 31.7 | |
| T2Backbone=Deepseek-R1-Distill-Qwen-7B2026.01 | 32.4 | |
| Std-PTBackbone=Qwen3-4B2026.01 | 33.2 | |
| ESCBackbone=Qwen3-4B2026.01 | 65.4 | |
| ACBackbone=Qwen3-4B2026.01 | 70.1 | |
| Std-PTBackbone=Deepseek-R1-Distill-Qwen-7B2026.01 | 72.1 | |
| DeepConfBackbone=Qwen3-4B2026.01 | 106.2 | |
| ESCBackbone=Deepseek-R1-Distill-Qwen-7B2026.01 | 120.4 | |
| DSCBackbone=Qwen3-4B2026.01 | 130.1 | |
| ACBackbone=Deepseek-R1-Distill-Qwen-7B2026.01 | 188.9 | |
| DeepConfBackbone=Deepseek-R1-Distill-Qwen-7B2026.01 | 274 | |
| DSCBackbone=Deepseek-R1-Distill-Qwen-7B2026.01 | 290.6 |