--- license: apache-2.0 tags: - tokenizer - bpe - multilingual - parity-aware --- # Parity-Aware BPE Tokenizer (dev-set balanced) ## Languages Trained on 21 languages, using FineWeb (train) and FLORES-200 (dev) per language: | Language | Training source | FLORES-200 code | |---|---|---| | English | `fw_edu` | `eng_Latn` | | German | `deu_Latn` | `deu_Latn` | | French | `fra_Latn` | `fra_Latn` | | Arabic | `arb_Arab` | `arb_Arab` | | Farsi | `fas_Arab` | `fas_Arab` | | Czech | `ces_Latn` | `ces_Latn` | | Chinese | `cmn_Hani` | `cmn_Hani` | | Danish | `dan_Latn` | `dan_Latn` | | Dutch | `nld_Latn` | `nld_Latn` | | Greek | `ell_Grek` | `ell_Grek` | | Hungarian | `hun_Latn` | `hun_Latn` | | Indonesian | `ind_Latn` | `ind_Latn` | | Italian | `ita_Latn` | `ita_Latn` | | Japanese | `jpn_Jpan` | `jpn_Jpan` | | Polish | `pol_Latn` | `pol_Latn` | | Portuguese | `por_Latn` | `por_Latn` | | Russian | `rus_Cyrl` | `rus_Cyrl` | | Spanish | `spa_Latn` | `spa_Latn` | | Swedish | `swe_Latn` | `swe_Latn` | | Turkish | `tur_Latn` | `tur_Latn` | | Vietnamese | `vie_Latn` | `vie_Latn` |