Skip to main content
AIMSAI Measurement Science

Mathematics

AdaptiveStep (ASPRM) BoN

Per-sample answer correctness of the MetaMath Llama-3.1-8B and Mistral-7B policy generators on the released AdaptiveStep Best-of-N candidate pools for GSM8k and MATH500 (256 samples per problem).

1,819items
2subjects
100%observed
unknownlicense
mathematicsdomain
reasoningdomain
textmodality
Original sourcePaperBuild script← All benchmarks

Response matrix

Fit to width. Hover for subject & item; click a cell for details.

AdaptiveStep (ASPRM) BoN response matrix: AI models (rows) against items (columns)
Correct (1)Incorrect (0)Unobserved

Scale: 1 = correct · 0 = incorrect

Subjects

  1. 1MetaMath-LLaMA-8B0.6938
  2. 2MetaMath-Mistral-7B0.6186
Full data on Hugging FaceBack to the gallery