metadata
license: apache-2.0
task_categories:
- text-generation
- question-answering
language:
- en
tags:
- fine
- finepdfs
- edu
- fineweb
- web
- pdf
- pdfs
- clean
- filtered
- cleaned
- reason
- reasoning
- logic
- high-quality
- curated
- fine data
- FW
- fineweb edu
- FW edu
- educational
- dense
- non-synthetic
size_categories:
- 100K<n<1M
PerfectPdfs
PerfectPdfs is a heavily filtered and cleaned english-only version of finepdfs-edu.
It removes all truncated rows, ensures every row has atleast a 0.95+ english score, deletes any duplicates, fixes misspellings, and keeps only rows with a 3+ edu score.
It also deletes unneeded tags for training include dump, id, offset, url, date, file_path, language (we already know it's english), minhash_cluster_size, extractor, and more.
Dataset Details
- Curated by: AInion
- Funded by: ArrowAI
- Shared by: MoreThought
- Language: english
- License: apache 2.0
Dataset Sources
- Repository: HuggingFaceFW/finepdfs-edu
- Producers: HuggingFaceFW
- License: odc-by
Statistics
| Total parquet size | Total rows | Average parquet row size |
|---|---|---|
| 5.6GB | 1,648,110 | 3.39kb |
Uses
- PreTraining SLMs on clean, heavily filtered top-tier pdfs.
- PostTraining LLMs on clean, heavily filtered top-tier pdfs.
- PreTraining data quality benefits on SLMs research.