mdroth/huggingface-course_section-5_zst

Hugging Face course section 5 .zst datasets

You can use these datasets for whatever you want (note the Apache 2.0 license, though) but their primary purpose is to serve as a drop-in replacement for the sub-datasets of The Pile used in section 5 of the HuggingFace course.

Data sources

• PubMed-200k-RTC:
  https://www.kaggle.com/datasets/matthewjansen/pubmed-200k-rtc/download?datasetVersionNumber=5
• LegalText-classification:
  https://www.kaggle.com/datasets/shivamb/legal-citation-text-classification/download?datasetVersionNumber=1

These are Kaggle datasets. So you need to be logged into a Kaggle account to download them from Kaggle. However, you actually don't need to download (and preprocess) them from Kaggle – you can just use them as shown in the following Usage section.

Usage

To load a dataset from this repo, run

import zstandard
from datasets import load_dataset
load_dataset("json", data_files=url, split="train")

where url should be one of the following download links:

• LegalText-classification_train.jsonl.zst:
  https://huggingface.co/datasets/mdroth/PubMed-200k-RTC/resolve/main/data/LegalText-classification_train.jsonl.zst,
• LegalText-classification_train_min.jsonl.zst:
  https://huggingface.co/datasets/mdroth/PubMed-200k-RTC/resolve/main/data/LegalText-classification_train_min.jsonl.zst,
• PubMed-200k-RTC_train.jsonl.zst:
  https://huggingface.co/datasets/mdroth/PubMed-200k-RTC/resolve/main/data/PubMed-200k-RTC_train.jsonl.zst, or
• PubMed-200k-RTC_train_min.jsonl.zst:
  https://huggingface.co/datasets/mdroth/PubMed-200k-RTC/resolve/main/data/PubMed-200k-RTC_train_min.jsonl.zst.

Example:

import zstandard
from datasets import load_dataset
url = "https://huggingface.co/datasets/mdroth/PubMed-200k-RTC/resolve/main/data/LegalText-classification_train_min.jsonl.zst"
load_dataset("json", data_files=url, split="train")