UNPKG

gpt-tokenizer

Version:

A pure JavaScript implementation of a BPE tokenizer (Encoder/Decoder) for GPT-2 / GPT-3 / GPT-4 and other OpenAI models

17 lines (15 loc) 550 B
/* eslint-disable no-magic-numbers */ import type { RawBytePairRanks } from '../BytePairEncodingCore.js' import { type EncodingParams } from '../modelParams.js' import { EndOfText } from '../specialTokens.js' import { R50K_TOKEN_SPLIT_REGEX } from './constants.js' export function P50KBase( bytePairRankDecoder: RawBytePairRanks, ): EncodingParams { return { expectedVocabularySize: 50_281, tokenSplitRegex: R50K_TOKEN_SPLIT_REGEX, bytePairRankDecoder, specialTokensEncoder: new Map<string, number>([[EndOfText, 50_256]]), } }