1318return new TiktokenTokenizer( 1346=> new TiktokenTokenizer(vocabFilePath, preTokenizer, specialTokens, normalizer, cacheSize); 1366=> new TiktokenTokenizer(vocabStream, preTokenizer, specialTokens, normalizer, cacheSize); 1397return new TiktokenTokenizer(encoder, decoder, vocab, preTokenizer, specialTokens, normalizer, cacheSize); 1461return new TiktokenTokenizer(vocabStream,

22private static readonly Tokenizer _gpt4oTokenizer = TiktokenTokenizer.CreateForModel("gpt-4o");

43/// Create <see cref="TiktokenTokenizer"/> from tokenizer model file. 47public static TiktokenTokenizer FromPretrained( 53return TiktokenTokenizer.Create(File.OpenRead(modelFilePath), preTokenizer, normalizer: null, specialTokens: _specialTokens);

68var tokenizer = LlamaTokenizerHelper.FromPretrained(modelWeightFolder);

36var tokenizer = LlamaTokenizerHelper.FromPretrained(originalWeightFolder); 39var pipeline = new CausalLMPipeline<TiktokenTokenizer, LlamaForCausalLM>(tokenizer, model, device);

81public static ICausalLMPipeline<TiktokenTokenizer, LlamaForCausalLM> LoadModel(string weightFolder, string checkPointName = "model.safetensors.index.json") 91var tokenizer = LlamaTokenizerHelper.FromPretrained(originalWeightFolder); 94var pipeline = new CausalLMPipeline<TiktokenTokenizer, LlamaForCausalLM>(tokenizer, model, device);

37var tokenizer = LlamaTokenizerHelper.FromPretrained(originalWeightFolder); 40var pipeline = new CausalLMPipeline<TiktokenTokenizer, LlamaForCausalLM>(tokenizer, model, device);

1879new RegexPreTokenizer(TiktokenTokenizer.P50kBaseRegex(), CodeGenTokenizer.CodeGenSpecialTokens),

117vocabStream, mergesStream, new RegexPreTokenizer(TiktokenTokenizer.P50kBaseRegex(), CodeGenTokenizer.CodeGenSpecialTokens), normalizer: null,

1285private static TiktokenTokenizer CreateForModel( 1340public static TiktokenTokenizer Create( 1360public static TiktokenTokenizer Create( 1381public static async Task<TiktokenTokenizer> CreateAsync( 1413public static async Task<TiktokenTokenizer> CreateAsync( 1439public static TiktokenTokenizer CreateForModel( 1478public static async Task<TiktokenTokenizer> CreateForModelAsync( 1515public static TiktokenTokenizer CreateForModel(string modelName, IReadOnlyDictionary<string, int>? extraSpecialTokens = null, Normalizer? normalizer = null) 1525public static TiktokenTokenizer CreateForEncoding(string encodingName, IReadOnlyDictionary<string, int>? extraSpecialTokens = null, Normalizer? normalizer = null) 1559throw new ArgumentException($"The encoding name '{encodingName}' is not supported. The only supported encoding names are: {TiktokenTokenizer.Cl100kBaseEncodingName}, {TiktokenTokenizer.P50kBaseEncodingName}, {TiktokenTokenizer.P50kEditEncodingName}, and {TiktokenTokenizer.R50kBaseEncodingName}.", nameof(encodingName));

32return SplitText(text, TiktokenTokenizer.P50kBaseRegex()); 47return SplitText(text, TiktokenTokenizer.P50kBaseRegex());

30var exception = Record.Exception(() => TiktokenTokenizer.CreateForModel(modelName)); 53TiktokenTokenizer externalTokenizer = TiktokenTokenizer.Create(tokenizerDataFileName, preTokenizer: null, normalizer: null);

32public static Tokenizer GPT4 { get; } = TiktokenTokenizer.CreateForModel("gpt-4", _specialTokens); 33public static Tokenizer GPT2 { get; } = TiktokenTokenizer.CreateForModel("gpt2"); 34public static Tokenizer P50kBase { get; } = TiktokenTokenizer.CreateForModel("text-davinci-003"); 35public static Tokenizer R50kBase { get; } = TiktokenTokenizer.CreateForModel("ada"); 36public static Tokenizer P50kEdit { get; } = TiktokenTokenizer.CreateForModel("text-davinci-edit-001"); 37public static Tokenizer GPT4o { get; } = TiktokenTokenizer.CreateForModel("gpt-4o"); 38public static Tokenizer GPT5 { get; } = TiktokenTokenizer.CreateForModel("gpt-5"); 39public static Tokenizer Phi4 { get; } = TiktokenTokenizer.CreateForModel("phi-4"); 40public static TiktokenTokenizer GptOss { get; } = TiktokenTokenizer.CreateForModel("gpt-oss-20b"); 48Assert.True(GPT4 is TiktokenTokenizer); 49IReadOnlyDictionary<string, int>? specialTokens = (GPT4 as TiktokenTokenizer)!.SpecialTokens; 53string assemblyName = typeof(TiktokenTokenizer).Assembly.FullName!; 64Tokenizer tokenizer = TiktokenTokenizer.Create(tokenizerDataFileName, GPT4.PreTokenizer, null, specialTokens); 69tokenizer = TiktokenTokenizer.Create(stream, GPT4.PreTokenizer, null, specialTokens); 73tokenizer = await TiktokenTokenizer.CreateAsync(tokenizerDataFileName, GPT4.PreTokenizer, normalizer: null, specialTokens); 78tokenizer = await TiktokenTokenizer.CreateAsync(stream, GPT4.PreTokenizer, normalizer: null, specialTokens); 84tokenizer = TiktokenTokenizer.CreateForModel("gpt-4", stream); 90tokenizer = await TiktokenTokenizer.CreateForModelAsync("gpt-3.5-turbo", stream); 94tokenizer = TiktokenTokenizer.CreateForModel("gpt-4"); 121TiktokenTokenizer tiktoken = (tokenizer as TiktokenTokenizer)!; 122TiktokenTokenizer externalTokenizer = TiktokenTokenizer.Create(tokenizerDataFileName, tokenizer.PreTokenizer, null, tiktoken.SpecialTokens); 146TestDecodingWithSpan((tokenizer as TiktokenTokenizer)!, encoded.ToArray(), text); 163private void TestDecodingWithSpan(TiktokenTokenizer tokenizer, int[] ids, string expectedDecoded) 199TestDecodingWithSpan((GPT4 as TiktokenTokenizer)!, encoded.ToArray(), text); 230TestDecodingWithSpan((gpt4Tokenizer as TiktokenTokenizer)!, encoded.ToArray(), text); 242TestDecodingWithSpan((GPT4 as TiktokenTokenizer)!, encoded.ToArray(), text); 277TestDecodingWithSpan((GPT4 as TiktokenTokenizer)!, encoded.ToArray(), text); 289foreach (TiktokenTokenizer tokenizer in new[] { GPT4o, GptOss, GPT5 }) 342TestDecodingWithSpan((GPT2 as TiktokenTokenizer)!, encoded.ToArray(), text); 361TestDecodingWithSpan((P50kBase as TiktokenTokenizer)!, encoded.ToArray(), text); 380TestDecodingWithSpan((P50kEdit as TiktokenTokenizer)!, encoded.ToArray(), text); 399TestDecodingWithSpan((R50kBase as TiktokenTokenizer)!, encoded.ToArray(), text); 477Tokenizer tokenizer = TiktokenTokenizer.CreateForModel(modelName); 478Assert.True(tokenizer is TiktokenTokenizer); 491Tokenizer tokenizer = TiktokenTokenizer.CreateForEncoding(encodingName); 492Assert.True(tokenizer is TiktokenTokenizer); 506Tokenizer tokenizer1 = TiktokenTokenizer.CreateForModel(modelName); 508Assert.True(tokenizer is TiktokenTokenizer); 509Assert.True(tokenizer1 is TiktokenTokenizer); 511TiktokenTokenizer tiktoken = (tokenizer as TiktokenTokenizer)!; 512TiktokenTokenizer tiktoken1 = (tokenizer1 as TiktokenTokenizer)!; 523Assert.Throws<ArgumentNullException>(() => TiktokenTokenizer.CreateForEncoding(null!)); 524Assert.Throws<ArgumentException>(() => TiktokenTokenizer.CreateForEncoding("r50k_base_")); 525Assert.Throws<ArgumentException>(() => TiktokenTokenizer.CreateForEncoding("p50k_base_")); 526Assert.Throws<ArgumentException>(() => TiktokenTokenizer.CreateForEncoding("p50k_edit_")); 527Assert.Throws<ArgumentException>(() => TiktokenTokenizer.CreateForEncoding("cl100k_base_")); 528Assert.Throws<ArgumentException>(() => TiktokenTokenizer.CreateForEncoding("o200k_base_")); 529Assert.Throws<ArgumentException>(() => TiktokenTokenizer.CreateForEncoding("o200k_harmony_")); 553Tokenizer tokenizer = TiktokenTokenizer.CreateForModel(name); 554Assert.True(tokenizer is TiktokenTokenizer); 558int entriesCount = GetEncoder((tokenizer as TiktokenTokenizer)!)!.Count; 835private static IReadOnlyDictionary<ReadOnlyMemory<byte>, int>? GetEncoder(TiktokenTokenizer tiktoken) 836=> typeof(TiktokenTokenizer).GetProperty("Encoder", BindingFlags.Instance | BindingFlags.NonPublic)?.GetValue(tiktoken) as IReadOnlyDictionary<ReadOnlyMemory<byte>, int>; 838private static IReadOnlyDictionary<int, ReadOnlyMemory<byte>>? GetDecoder(TiktokenTokenizer tiktoken) 839=> typeof(TiktokenTokenizer).GetProperty("Decoder", BindingFlags.Instance | BindingFlags.NonPublic)?.GetValue(tiktoken) as IReadOnlyDictionary<int, ReadOnlyMemory<byte>>; 841private static IReadOnlyDictionary<string, int>? GetVocabulary(TiktokenTokenizer tiktoken) 842=> typeof(TiktokenTokenizer).GetProperty("Vocabulary", BindingFlags.Instance | BindingFlags.NonPublic)?.GetValue(tiktoken) as IReadOnlyDictionary<string, int>;