gpt-token-utils
Version:
Isomorphic utilities for GPT-3 tokenization and prompt building.
44 lines (43 loc) • 1.18 kB
text/typescript
/**
* @copyright Sister Software. All rights reserved.
* @author Teffen Ellis, et al.
* @license
* See LICENSE file in the project root for full license information.
*/
/**
* A map of byte-pair encodings to their corresponding tokens.
* @internal
*/
export type TokenEncodingsRecord = Record<string, number | undefined>;
declare const nodeInspectSymbol: unique symbol;
/**
* Two-way map between Unicode byte-pairs and tokens.
* @internal
*/
export declare class BytePairTokenMap {
protected _bpeTokenMap: Map<
/**
* Byte paired character(s), e.g. `'!'`, `'\u00a8'`
*/
string,
/**
* The corresponding token, e.g. `0`, `101`
*/
number>;
protected _tokenBPEMap: Map<
/**
* The corresponding token, e.g. `0`, `101`
*/
number,
/**
* Byte paired character(s), e.g. `'!'`, `'\u00a8'`
*/
string>;
constructor(tokenEncodings: TokenEncodingsRecord, nMergedSpaces?: number);
addBytePair(bytePair: string, token: number): void;
tokenToBytePair(token: number): string;
bytePairToToken(bytePair: string): number;
get size(): number;
[nodeInspectSymbol](): string;
}
export {};