Mathematical Reasoning on MATH500 (Inference latency (s))
1.2Inference Latency (s)T2
Evaluation Results
| Method | Links | |
|---|---|---|
| T2Backbone=Llama3.1-8B2026.01 | 1.2 | |
| T2Backbone=Qwen2.5-7B2026.01 | 1.6 | |
| T2Backbone=Qwen3-4B2026.01 | 2.9 | |
| Std-PTBackbone=Llama3.1-8B2026.01 | 3 | |
| Std-PTBackbone=Qwen2.5-7B2026.01 | 3.4 | |
| ESCBackbone=Llama3.1-8B2026.01 | 4.2 | |
| DSCBackbone=Qwen2.5-7B2026.01 | 4.9 | |
| ACBackbone=Qwen2.5-7B2026.01 | 5.1 | |
| ACBackbone=Llama3.1-8B2026.01 | 5.1 | |
| DSCBackbone=Llama3.1-8B2026.01 | 5.2 | |
| ESCBackbone=Qwen3-4B2026.01 | 5.7 | |
| T2Backbone=Deepseek-R1-Distill-Qwen-7B2026.01 | 5.8 | |
| Std-PTBackbone=Qwen3-4B2026.01 | 6.5 | |
| DeepConfBackbone=Llama3.1-8B2026.01 | 7.1 | |
| DSCBackbone=Qwen3-4B2026.01 | 8.8 | |
| ESCBackbone=Qwen2.5-7B2026.01 | 9.4 | |
| ACBackbone=Qwen3-4B2026.01 | 10.1 | |
| DeepConfBackbone=Qwen2.5-7B2026.01 | 11.7 | |
| ESCBackbone=Deepseek-R1-Distill-Qwen-7B2026.01 | 13.2 | |
| Std-PTBackbone=Deepseek-R1-Distill-Qwen-7B2026.01 | 15.7 | |
| DeepConfBackbone=Qwen3-4B2026.01 | 18 | |
| ACBackbone=Deepseek-R1-Distill-Qwen-7B2026.01 | 29.7 | |
| DSCBackbone=Deepseek-R1-Distill-Qwen-7B2026.01 | 33.4 | |
| DeepConfBackbone=Deepseek-R1-Distill-Qwen-7B2026.01 | 60.6 |