Embeddings 0.4.0 isubwordtokenizer encode

Lodestar.Embeddings 0.4.0. This page is frozen at that release. Read the current documentation for what main says now. A link to a decision or a migration page follows main, and leaves the archive.

ISubwordTokenizer.Encode

One string to tokens and ids.

public TokenizationResult Encode(string text)

Parameters — text is the string to encode. Empty is legal and encodes to nothing.

Returns — TokenizationResult: the token strings and their ids, the same length and in the same order.

Example — one word that splits, one that does not.

using Lodestar.Embeddings.Tokenization;

var vocab = new Dictionary<string, int>(StringComparer.Ordinal)
{
    ["[UNK]"] = 0, ["token"] = 1, ["##ize"] = 2, ["text"] = 3,
};
ISubwordTokenizer tokenizer = new WordPieceTokenizer(
    vocab, unkToken: "[UNK]", continuationPrefix: "##", maxCharsPerWord: 100, lowercase: true);

TokenizationResult encoded = tokenizer.Encode("tokenize text");

string first = encoded.Tokens[0];  // => token
string second = encoded.Tokens[1];  // => ##ize

Remarks — no special tokens are added. Encoding produces what the model's vocabulary says the text is, and wrapping that in [CLS]/[SEP] or their equivalents belongs to BatchEncoder, because which tokens wrap a sequence depends on the model rather than on the text.

A word the vocabulary cannot cover becomes the unknown token — one of it, not one per character.

Applies to — net10.0, netstandard2.0.

See also — TokenizationResult, BatchEncoder.