Data for Wordpiece-Style Tokenization

Provides data to be used by the wordpiece algorithm in order to tokenize text into somewhat meaningful chunks. Included vocabularies were retrieved from < https://huggingface.co/bert-base-cased/resolve/main/vocab.txt> and < https://huggingface.co/bert-base-uncased/resolve/main/vocab.txt> and parsed into an R-friendly format.

wordpiece.data

The goal of wordpiece.data is to provide stable, versioned data for use in the {wordpiece} tokenizer package.

Installation

You can install the released version of wordpiece.data from CRAN with:

install.packages("wordpiece.data")

And the development version from GitHub with:

# install.packages("remotes")
remotes::install_github("macmillancontentscience/wordpiece.data")

Dataset Creation

The datasets included in this package were retrieved from huggingface (specifically, cased and uncased). They were then processed using the {wordpiece} package. This is a bit circular, because this package is a dependency for the wordpiece package.

vocab_txt <- tempfile(fileext = ".txt")
download.file(
  url = "https://huggingface.co/bert-base-cased/resolve/main/vocab.txt", 
  destfile = vocab_txt
)
parsed_vocab <- wordpiece::load_vocab(vocab_txt)
rds_filename <- paste0(
  paste(
    "wordpiece",
    "cased",
    length(parsed_vocab),
    sep = "_"
  ),
  ".rds"
)
saveRDS(parsed_vocab, here::here("inst", "rds", rds_filename))
unlink(vocab_txt)

vocab_txt <- tempfile(fileext = ".txt")
download.file(
  url = "https://huggingface.co/bert-base-uncased/resolve/main/vocab.txt", 
  destfile = vocab_txt
)
parsed_vocab <- wordpiece::load_vocab(vocab_txt)
rds_filename <- paste0(
  paste(
    "wordpiece",
    "uncased",
    length(parsed_vocab),
    sep = "_"
  ),
  ".rds"
)
saveRDS(parsed_vocab, here::here("inst", "rds", rds_filename))
unlink(vocab_txt)

Example

You likely won’t ever need to use this package directly. It contains a function to load data used by {wordpiece}.

library(wordpiece.data)

head(wordpiece_vocab())
#> [1] "[PAD]"     "[unused0]" "[unused1]" "[unused2]" "[unused3]" "[unused4]"

Code of Conduct

Please note that the wordpiece.data project is released with a Contributor Code of Conduct. By contributing to this project, you agree to abide by its terms.

Disclaimer

This is not an officially supported Macmillan Learning product.

Contact information

Questions or comments should be directed to Jonathan Bratt ([email protected]) and Jon Harmon ([email protected]).