operators:
- id: load_hf_python
  config:
    type: hf_source
    dataset: cfahlgren1/react-code-instructions
    split: train
    trust_remote_code: True
- id: clean_up
  config:
    type: function
    function: data_strategies.SeedCodeData.utils.react_map
  input_ids:
  - load_hf_python
- id: rename_task_column
  config:
    type: function
    function: data_strategies.commons.keep_only_columns
    function_config:
      columns_to_keep:
      - response_seed
      - instruction_seed
  input_ids:
  - clean_up
- id: sample_dataset
  config:
    type: function
    function: data_strategies.commons.uniform_sample_limited
    function_config:
      num_samples: 50000  # 1.25% of 10000 target from stage 1
  input_ids:
  - rename_task_column
- id: annotate
  config:
    type: completions
    map: chat
    map_config:
      system_message: "You are a helpful assistant."
      user_message_column: instruction_seed
      output_column: response
    model: gpt-4o-mini
    temperature: 1.0
    require_all_responses: False
    batch: True
  input_ids:
  - sample_dataset