1254return new TiktokenTokenizer( 1282=> new TiktokenTokenizer(vocabFilePath, preTokenizer, specialTokens, normalizer, cacheSize); 1302=> new TiktokenTokenizer(vocabStream, preTokenizer, specialTokens, normalizer, cacheSize); 1333return new TiktokenTokenizer(encoder, decoder, vocab, preTokenizer, specialTokens, normalizer, cacheSize); 1397return new TiktokenTokenizer(vocabStream,

23private 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,

1221private static TiktokenTokenizer CreateForModel( 1276public static TiktokenTokenizer Create( 1296public static TiktokenTokenizer Create( 1317public static async Task<TiktokenTokenizer> CreateAsync( 1349public static async Task<TiktokenTokenizer> CreateAsync( 1375public static TiktokenTokenizer CreateForModel( 1414public static async Task<TiktokenTokenizer> CreateForModelAsync( 1451public static TiktokenTokenizer CreateForModel(string modelName, IReadOnlyDictionary<string, int>? extraSpecialTokens = null, Normalizer? normalizer = null) 1461public static TiktokenTokenizer CreateForEncoding(string encodingName, IReadOnlyDictionary<string, int>? extraSpecialTokens = null, Normalizer? normalizer = null) 1491throw 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 Phi4 { get; } = TiktokenTokenizer.CreateForModel("phi-4"); 46Assert.True(GPT4 is TiktokenTokenizer); 47IReadOnlyDictionary<string, int>? specialTokens = (GPT4 as TiktokenTokenizer)!.SpecialTokens; 51string assemblyName = typeof(TiktokenTokenizer).Assembly.FullName!; 62Tokenizer tokenizer = TiktokenTokenizer.Create(tokenizerDataFileName, GPT4.PreTokenizer, null, specialTokens); 67tokenizer = TiktokenTokenizer.Create(stream, GPT4.PreTokenizer, null, specialTokens); 71tokenizer = await TiktokenTokenizer.CreateAsync(tokenizerDataFileName, GPT4.PreTokenizer, normalizer: null, specialTokens); 76tokenizer = await TiktokenTokenizer.CreateAsync(stream, GPT4.PreTokenizer, normalizer: null, specialTokens); 82tokenizer = TiktokenTokenizer.CreateForModel("gpt-4", stream); 88tokenizer = await TiktokenTokenizer.CreateForModelAsync("gpt-3.5-turbo", stream); 92tokenizer = TiktokenTokenizer.CreateForModel("gpt-4"); 119TiktokenTokenizer tiktoken = (tokenizer as TiktokenTokenizer)!; 120TiktokenTokenizer externalTokenizer = TiktokenTokenizer.Create(tokenizerDataFileName, tokenizer.PreTokenizer, null, tiktoken.SpecialTokens); 144TestDecodingWithSpan((tokenizer as TiktokenTokenizer)!, encoded.ToArray(), text); 161private void TestDecodingWithSpan(TiktokenTokenizer tokenizer, int[] ids, string expectedDecoded) 197TestDecodingWithSpan((GPT4 as TiktokenTokenizer)!, encoded.ToArray(), text); 228TestDecodingWithSpan((gpt4Tokenizer as TiktokenTokenizer)!, encoded.ToArray(), text); 240TestDecodingWithSpan((GPT4 as TiktokenTokenizer)!, encoded.ToArray(), text); 275TestDecodingWithSpan((GPT4 as TiktokenTokenizer)!, encoded.ToArray(), text); 301TestDecodingWithSpan((GPT4o as TiktokenTokenizer)!, encoded.ToArray(), text); 309TestDecodingWithSpan((GPT4o as TiktokenTokenizer)!, encoded.ToArray(), text); 337TestDecodingWithSpan((GPT2 as TiktokenTokenizer)!, encoded.ToArray(), text); 356TestDecodingWithSpan((P50kBase as TiktokenTokenizer)!, encoded.ToArray(), text); 375TestDecodingWithSpan((P50kEdit as TiktokenTokenizer)!, encoded.ToArray(), text); 394TestDecodingWithSpan((R50kBase as TiktokenTokenizer)!, encoded.ToArray(), text); 455Tokenizer tokenizer = TiktokenTokenizer.CreateForModel(modelName); 456Assert.True(tokenizer is TiktokenTokenizer); 468Tokenizer tokenizer = TiktokenTokenizer.CreateForEncoding(encodingName); 469Assert.True(tokenizer is TiktokenTokenizer); 482Tokenizer tokenizer1 = TiktokenTokenizer.CreateForModel(modelName); 484Assert.True(tokenizer is TiktokenTokenizer); 485Assert.True(tokenizer1 is TiktokenTokenizer); 487TiktokenTokenizer tiktoken = (tokenizer as TiktokenTokenizer)!; 488TiktokenTokenizer tiktoken1 = (tokenizer1 as TiktokenTokenizer)!; 499Assert.Throws<ArgumentNullException>(() => TiktokenTokenizer.CreateForEncoding(null!)); 500Assert.Throws<ArgumentException>(() => TiktokenTokenizer.CreateForEncoding("r50k_base_")); 501Assert.Throws<ArgumentException>(() => TiktokenTokenizer.CreateForEncoding("p50k_base_")); 502Assert.Throws<ArgumentException>(() => TiktokenTokenizer.CreateForEncoding("p50k_edit_")); 503Assert.Throws<ArgumentException>(() => TiktokenTokenizer.CreateForEncoding("cl100k_base_")); 504Assert.Throws<ArgumentException>(() => TiktokenTokenizer.CreateForEncoding("o200k_base_")); 526Tokenizer tokenizer = TiktokenTokenizer.CreateForModel(name); 527Assert.True(tokenizer is TiktokenTokenizer); 531int entriesCount = GetEncoder((tokenizer as TiktokenTokenizer)!)!.Count; 761private static IReadOnlyDictionary<ReadOnlyMemory<byte>, int>? GetEncoder(TiktokenTokenizer tiktoken) 762=> typeof(TiktokenTokenizer).GetProperty("Encoder", BindingFlags.Instance | BindingFlags.NonPublic)?.GetValue(tiktoken) as IReadOnlyDictionary<ReadOnlyMemory<byte>, int>; 764private static IReadOnlyDictionary<int, ReadOnlyMemory<byte>>? GetDecoder(TiktokenTokenizer tiktoken) 765=> typeof(TiktokenTokenizer).GetProperty("Decoder", BindingFlags.Instance | BindingFlags.NonPublic)?.GetValue(tiktoken) as IReadOnlyDictionary<int, ReadOnlyMemory<byte>>; 767private static IReadOnlyDictionary<string, int>? GetVocabulary(TiktokenTokenizer tiktoken) 768=> typeof(TiktokenTokenizer).GetProperty("Vocabulary", BindingFlags.Instance | BindingFlags.NonPublic)?.GetValue(tiktoken) as IReadOnlyDictionary<string, int>;