README.md

metadata

dataset_info:
  features:
    - name: problem
      dtype: string
    - name: solution
      dtype: string
    - name: response
      dtype: string
    - name: correctness
      dtype: float64
    - name: pass_rate
      dtype: float64
  splits:
    - name: train
      num_bytes: 1125571107
      num_examples: 480000
    - name: filtered
      num_bytes: 15749507
      num_examples: 7056
    - name: baseline
      num_bytes: 15749507
      num_examples: 7056
    - name: confidence
      num_bytes: 27327374
      num_examples: 7056
  download_size: 315106701
  dataset_size: 1184397495
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: filtered
        path: data/filtered-*
      - split: confidence
        path: data/confidence-*
      - split: baseline
        path: data/baseline-*

Sample from Qwen2.5-7B-Instruct

Confidence Reasoning from Qwen2.5-32B-Instruct