name: feedback-example-1
operators:
  # this is representative of each subset of orca: we load the FLAN dataset and then subsample from it
  - id: load_flan_zsopt_data
    config:
      type: hf_source
      dataset: Open-Orca/FLAN
      split: train
      data_dir: flan_zsopt_data
      materialize_output: false

  - id: sample_flan_zsopt_data
    config:
      type: function
      function: data_strategies.SlimOrca.utils.sample_total_num_queries_stratified
      function_config:
        total_num_queries: 2_000_000
      materialize_output: false
    input_ids:
      - load_flan_zsopt_data