docvqa-augmented-rl / README.md
dddraxxx's picture
Upload dataset
506b109 verified
metadata
dataset_info:
  features:
    - name: images
      list: image
    - name: data_source
      dtype: string
    - name: reward_model
      struct:
        - name: ground_truth
          dtype: string
    - name: extra_info
      struct:
        - name: all_answers
          sequence: string
        - name: answer
          dtype: string
        - name: augmentation_index
          dtype: int64
        - name: augmentation_type
          dtype: string
        - name: data_split
          dtype: string
        - name: doc_id
          dtype: int64
        - name: has_original_ocr
          dtype: bool
        - name: index
          dtype: string
        - name: ocr_results
          struct:
            - name: clockwise_orientation
              dtype: float64
            - name: height
              dtype: int64
            - name: lines
              list:
                - name: bounding_box
                  sequence: int64
                - name: text
                  dtype: string
                - name: words
                  list:
                    - name: bounding_box
                      sequence: int64
                    - name: confidence
                      dtype: string
                    - name: text
                      dtype: string
            - name: page
              dtype: int64
            - name: unit
              dtype: string
            - name: width
              dtype: int64
        - name: original_question_id
          dtype: int64
        - name: question
          dtype: string
        - name: question_id
          dtype: int64
    - name: prompt
      list:
        - name: content
          dtype: string
        - name: role
          dtype: string
    - name: ability
      dtype: string
    - name: env_name
      dtype: string
  splits:
    - name: train
      num_bytes: 887318263.496
      num_examples: 3999
  download_size: 441975037
  dataset_size: 887318263.496
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*