Mathematics on AIME (MAT, Speedup)
3.57MAT ScoreLogitSpec
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LogitSpecBackbone=Llama-3.1-8B-Instruct, Temperature (T)=0, Tree capacity (K)=642025.07 | 3.57 | 3.09 | |
| SAM DecodingBackbone=Llama-3.1-8B-Instruct, Temperature (T)=0, Tree capacity (K)=642025.07 | 3.11 | 2.73 | |
| Token RecyclingBackbone=Llama-3.1-8B-Instruct, Temperature (T)=0, Tree capacity (K)=642025.07 | 3.1 | 2.82 | |
| AR (Greedy)Backbone=Llama-3.1-8B-Instruct, Temperature (T)=0, Tree capacity (K)=642025.07 | 1 | 1 |