bees
Datasets
All datasets matching “bees”govdocs1-pdf-source
govdocs1: source PDF files
[!NOTE]
Converted versions of other document types (word, txt, etc) are available in this repo
This is ~220,000 open-access PDF documents (about 6.6M pages) from the dataset govdocs1. It wants to be OCR'd.
Uploaded as tar file pieces of ~10 GiB each due to size/file count limits with an index.csv covering details
5,000 randomly sampled PDFs are available unarchived in sample/. Hugging Face supports previewing these in-browser, for example this one… See the full description on the dataset page: https://huggingface.co/datasets/BEE-spoke-data/govdocs1-pdf-source.wikipedia-20230901.en-deduped
wikipedia - 20230901.en - deduped
purpose: train with less data while maintaining (most) of the quality
This is really more of a "high quality diverse sample" rather than "we are trying to remove literal duplicate documents". Source dataset: graelo/wikipedia.
configs
default
command:
python -m text_dedup.minhash \
--path $ds_name \
--name $dataset_config \
--split $data_split \
--cache_dir "./cache" \
--output $out_dir \
--column $text_column \… See the full description on the dataset page: https://huggingface.co/datasets/BEE-spoke-data/wikipedia-20230901.en-deduped.code_contests_instruct
Dataset Card for "code_contests_instruct"
The deepmind/code_contests dataset formatted as markdown-instruct for text generation training.
There are several different configs. Look at them. Comments:
flesch_reading_ease is computed on the description col via textstat
hq means that python2 (aka PYTHON in language column) is dropped, and keeps only rows with flesch_reading_ease 75 or greater
min-cols drops all cols except language and text
possible values for language are {'CPP'… See the full description on the dataset page: https://huggingface.co/datasets/BEE-spoke-data/code_contests_instruct.Long-Data-Col-rp_pile_pretrain
Dataset Card for "Long-Data-Col-rp_pile_pretrain"
This dataset is a subset of togethercomputer/Long-Data-Collections, namely the rp_sub.jsonl.zst and pile_sub.jsonl.zst files from the pretrain split.
Like the source dataset, we do not attempt to modify/change licenses of underlying data. Refer to the source dataset (and its source datasets) for details.
changes
as this is supposed to be a "long text dataset", we drop all rows where text contains <= 250 characters.… See the full description on the dataset page: https://huggingface.co/datasets/BEE-spoke-data/Long-Data-Col-rp_pile_pretrain.consumer-finance-complaints
BEE-spoke-data/consumer-finance-complaints
consumer-finance-complaints but in a format that actually works.
Pulled Feb 2024
TxT360-5M-sample-en
BEE-spoke-data/TxT360-5M-sample-en
english only sample from LLM360/TxT360:
min length 256 GPT-4 tokens
max length 24576 GPT-4 tokens
GPT-4 tiktoken token count:
token_count
count 5.000000e+06
mean 1.003614e+03
std 1.424231e+03
min 2.570000e+02
25% 4.020000e+02
50% 6.220000e+02
75% 1.050000e+03
max 2.457400e+04
Total count: 5018.07 M tokens