pszemraj's picture
filter with pszemraj/xtremedistil-l6-h256-OCR-quality
e46d9ab verified
|
raw
history blame
1.96 kB
metadata
language:
  - en
license: cc0-1.0
size_categories:
  - 1K<n<10K
source_datasets: storytracer/LoC-PD-Books
task_categories:
  - text-generation
  - feature-extraction
dataset_info:
  - config_name: default
    features:
      - name: lccn
        dtype: string
      - name: title
        dtype: string
      - name: author
        dtype: string
      - name: year
        dtype: int64
      - name: page_count
        dtype: int64
      - name: filename
        dtype: string
      - name: text
        dtype: string
      - name: label
        dtype: string
      - name: score
        dtype: float64
    splits:
      - name: train
        num_bytes: 2788098633.628336
        num_examples: 8816
    download_size: 1435586557
    dataset_size: 2788098633.628336
  - config_name: en-clean
    features:
      - name: lccn
        dtype: string
      - name: title
        dtype: string
      - name: author
        dtype: string
      - name: year
        dtype: int64
      - name: page_count
        dtype: int64
      - name: filename
        dtype: string
      - name: text
        dtype: string
      - name: score
        dtype: float64
    splits:
      - name: train
        num_bytes: 1906155961.9587114
        num_examples: 6399
    download_size: 1055862380
    dataset_size: 1906155961.9587114
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
  - config_name: en-clean
    data_files:
      - split: train
        path: en-clean/train-*
tags:
  - books

LoC-PD-Books: preprocessed

This is the storytracer/LoC-PD-Books dataset with the following preprocessing steps:

  • apply clean-text package keeping casing and newlines
  • drop OCR garbled text in first few lines of each example
  • fix (most) 'hard' newlines w/ regex similar to gutenberg clean
  • 'grade' first 512 tokens of each book with this quantized model; keep examples from labels clean (all) and mild gibberish w/ score 0.9 or higher