tokenizers v1 promises up to 30x faster BPE tokenization via data profiling

art_zucker · x · 2026-10-16

In tokenizers v1, BPE tokenizers will be up to 30x faster, while BERT tokenization gets about 8x. The key trick: profiling the data along with the code to locate bottlenecks. The author plans to keep optimizing the BERT tokenizer toward parity with BPE.

Related event: Hugging Face tokenizers nears v1.0 as BERT tokenizer gets ~50x speedup(7 posts)→

Original post →

More from coding & agent

coding & agent channel →