alasdairforsythe/tokenmonster
View on GitHub ↗ ★ 625 Go
Ungreedy subword tokenizer and vocabulary trainer for Python, Go, C++ & Javascript
4 mentions on Hacker News by 4 people: 2 stories, 2 comments · first 2023-06-04 · latest 2026-04-20
20082026
-
…Haiku also show the Title Case token (↑). This is similar to what the TokenMonster tokenizer does: https://github.com/alasdairforsythe/tokenmonster
comment by yorwba · 2026-04-20 · in Claude Token Counter, now with model comparisons -
The actual analysis: https://github.com/alasdairforsythe/tokenmonster/blob/main/b... > Summary of Findings: > - Comparable (50256-strict-nocapcode) TokenMonster vocabularies perform better than both GPT-2 Tokenizer and tiktoken p50k_base…
comment by alecco · 2023-09-05 · in Tokenizer benchmark comparing 16 language models pre-trained from scratch -
▲ 1
Tokenmonster: Determine tokens to optimally represents a datasetstory by anotherpaulg · 2023-06-09 · 0 comments
-
▲ 10
TokenMonster: Ungreedy tokenizer, outperforming tiktoken by 35%story by tosh · 2023-06-04 · 0 comments