sentencepiece
Text Tokenization using Byte Pair Encoding and Unigram Modelling
The R binding to Google's tokenizer has shipped nothing but compiler fixes since 2021.
◆Recent moves
- 6mo ago
Build switched to C++17
Switches Makevars to C++17, completing the standard migration begun in 0.2.4. Build configuration only.
View source ↗ - 8mo ago
C++11 dropped; deprecated std::iterator replaced in protobuf-lite
Drops C++11 from Makevars and replaces a std::iterator use in the vendored protobuf-lite tree that C++17 deprecates. Compiler conformance in third-party code.
View source ↗ - 3y ago
R CMD check warning from 0.2.2 cleared
Clears an R CMD check warning introduced by 0.2.2, in a section of vendored protobuf-lite the package does not even use. Published within a minute of 0.2.2.
View source ↗ - 3y ago
sprintf replaced with snprintf for M1 Mac checks
Replaces sprintf with snprintf to satisfy an R CMD check note on M1 Macs. The change that 0.2.3 immediately had to patch.
View source ↗ - 4y ago
- 4y ago
Vendored sentencepiece upgraded to v0.1.96; wordpiece fix
The last functional release: upgrades the vendored library to sentencepiece v0.1.96 and fixes wordpiece handling of one-character words. Its notes also repeat the 0.1.3 changelog, which added BPEembed and subword regularisation via the nbest and alpha options.
View source ↗