---
dataset_info:
  features:
  - name: prompt
    dtype: string
  - name: n_disks
    dtype: int64
  - name: source
    dtype: string
  - name: auxiliary
    dtype: string
  - name: target
    dtype: string
  - name: target_response
    dtype: string
  - name: optimal_response
    dtype: string
  - name: corrupted
    dtype: bool
  - name: corruption
    dtype: string
  splits:
  - name: train
    num_bytes: 185845
    num_examples: 600
  - name: grpo_train
    num_bytes: 61410
    num_examples: 200
  - name: heldout
    num_bytes: 11944
    num_examples: 40
  - name: train_instances
    num_bytes: 11897
    num_examples: 40
  - name: extrapolation
    num_bytes: 22914
    num_examples: 30
  download_size: 38858
  dataset_size: 294010
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: grpo_train
    path: data/grpo_train-*
  - split: heldout
    path: data/heldout-*
  - split: train_instances
    path: data/train_instances-*
  - split: extrapolation
    path: data/extrapolation-*
---