List your company

Tokenization Intrinsic Evaluation on mC4 unbalanced 60-lang (test)

0.0212Compression RatePA-BPE
0.021160.021430.02170.02197Aug 6, 2025

Evaluation results

MethodLinks
PA-BPEVocabulary size=128k2025.080.02122.6458.93.707528.328.846.6
PA-BPE (window)Vocabulary size=128k2025.080.02183.2170.13.6475110.131.446.9
PA-BPE (hybrid)Vocabulary size=128k2025.080.02192.7960.43.624521029.345
BPEVocabulary size=128k2025.080.02223.4172.83.5645114.433.548.1
PA-BPE (window+hybrid)Vocabulary size=128k2025.080.02223.3872.13.5765011.632.346.6