Math Problem Solving on Math Benchmarks LIMO curation (test)
72.6AccuracyLALP
Evaluation results
| Method | Links | |
|---|---|---|
| LALPStudent Model=Qwen2.5-32B-Instruct, Selection Rule=Highest local log probability2025.10 | 72.6 | |
| RandomStudent Model=Qwen2.5-32B-Instruct, Selection Rule=Random2025.10 | 65.1 | |
| GALPStudent Model=Qwen2.5-32B-Instruct, Selection Rule=Highest global log probability2025.10 | 63.2 | |
| Local LowestStudent Model=Qwen2.5-32B-Instruct, Selection Rule=Local Lowest (sanity check)2025.10 | 62.3 | |
| Original ModelStudent Model=Qwen2.5-32B-Instruct2025.10 | 44.5 | |
| LALPStudent Model=Qwen2.5-7B-Instruct, Selection Rule=Highest local log probability2025.10 | 44 | |
| GALPStudent Model=Qwen2.5-7B-Instruct, Selection Rule=Highest global log probability2025.10 | 41.2 | |
| RandomStudent Model=Qwen2.5-7B-Instruct, Selection Rule=Random2025.10 | 40.7 | |
| Local LowestStudent Model=Qwen2.5-7B-Instruct, Selection Rule=Local Lowest (sanity check)2025.10 | 39.9 | |
| Original ModelStudent Model=Qwen2.5-7B-Instruct2025.10 | 35.3 |