Introducing WordPiece and Bert tokenizers (#7275)

* Introducing WordPiece and Bert tokenizers

* Fix corner case in WordPiece

THIRD-PARTY-NOTICES.TXT

@@ -152,6 +152,25 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License.
 
+License notice for WordPiece and Bert tokenizers
+------------------------------------------------
+
+https://github.com/huggingface/transformers/blob/8e3e145b427196e014f37aa42ba890b9bc94275e/src/transformers/models/bert/tokenization_bert.py#L2
+
+Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+
 License notice for BitUtility
 ------------------------------------------
 

src/Microsoft.ML.Tokenizers/EncodedToken.cs

@@ -10,7 +10,7 @@ namespace Microsoft.ML.Tokenizers
     /// Represent the token produced from the tokenization process containing the token substring,
     /// the id associated to the token substring, and the offset mapping to the original string.
     /// </summary>
-    public readonly struct EncodedToken
+    public readonly struct EncodedToken : IEquatable<EncodedToken>
     {
         /// <summary>
         /// Gets the Id value associated to the token.
@@ -39,5 +39,8 @@ namespace Microsoft.ML.Tokenizers
             Offset = offset;
             Value = value;
         }
+
+        /// inherited
+        public bool Equals(EncodedToken other) => Id == other.Id && Value == other.Value && Offset.Equals(other.Offset);
     }
 }

src/Microsoft.ML.Tokenizers/Model/BPETokenizer.cs

@@ -87,7 +87,7 @@ namespace Microsoft.ML.Tokenizers
         /// <param name="vocabFile">The JSON file path containing the dictionary of string keys and their ids.</param>
         /// <param name="mergesFile">The file path containing the tokens's pairs list.</param>
         public static BpeTokenizer Create(string vocabFile, string? mergesFile)
-            => Create(vocabFile, mergesFile, preTokenizer: PreTokenizer.CreateWhiteSpace(), normalizer: null, unknownToken: null, continuingSubwordPrefix: null, endOfWordSuffix: null, fuseUnknownTokens: false);
+            => Create(vocabFile, mergesFile, preTokenizer: PreTokenizer.CreateWordOrNonWordPreTokenizer(), normalizer: null, unknownToken: null, continuingSubwordPrefix: null, endOfWordSuffix: null, fuseUnknownTokens: false);
 
         /// <summary>
         /// Create a new Bpe tokenizer object to use for text encoding.
@@ -131,7 +131,7 @@ namespace Microsoft.ML.Tokenizers
         /// <param name="vocabStream">The JSON stream containing the dictionary of string keys and their ids.</param>
         /// <param name="mergesStream">The stream containing the tokens's pairs list.</param>
         public static BpeTokenizer Create(Stream vocabStream, Stream? mergesStream)
-            => Create(vocabStream, mergesStream, preTokenizer: PreTokenizer.CreateWhiteSpace(), normalizer: null, addedTokens: null, unknownToken: null, continuingSubwordPrefix: null, endOfWordSuffix: null, fuseUnknownTokens: false);
+            => Create(vocabStream, mergesStream, preTokenizer: PreTokenizer.CreateWordOrNonWordPreTokenizer(), normalizer: null, addedTokens: null, unknownToken: null, continuingSubwordPrefix: null, endOfWordSuffix: null, fuseUnknownTokens: false);
 
         /// <summary>
         /// Create a new Bpe tokenizer object to use for text encoding.
@@ -225,7 +225,7 @@ namespace Microsoft.ML.Tokenizers
             FuseUnknownTokens = fuseUnknownTokens;
             ContinuingSubwordPrefix = continuingSubwordPrefix;
             EndOfWordSuffix = endOfWordSuffix;
-            _preTokenizer = preTokenizer ?? PreTokenizer.CreateWhiteSpace(); // Default to WhiteSpace pre-tokenizer
+            _preTokenizer = preTokenizer ?? PreTokenizer.CreateWordOrNonWordPreTokenizer(); // Default to WordOrNonWord pre-tokenizer
             _normalizer = normalizer;
 
             _vocab = vocab ?? new Dictionary<StringSpanOrdinalKey, int>();

src/Microsoft.ML.Tokenizers/Model/BertTokenizer.cs

@@ -0,0 +1,729 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+using System;
+using System.Buffers;
+using System.Collections.Generic;
+using System.IO;
+using System.Linq;
+using System.Text;
+using System.Threading.Tasks;
+
+namespace Microsoft.ML.Tokenizers
+{
+    /// <summary>
+    /// Tokenizer for Bert model.
+    /// </summary>
+    /// <remarks>
+    /// The BertTokenizer is a based on the WordPieceTokenizer and is used to tokenize text for Bert models.
+    /// The implementation of the BertTokenizer is based on the original Bert implementation in the Hugging Face Transformers library.
+    /// https://huggingface.co/transformers/v3.0.2/model_doc/bert.html?highlight=berttokenizerfast#berttokenizer
+    /// </remarks>
+    public sealed partial class BertTokenizer : WordPieceTokenizer
+    {
+        internal BertTokenizer(
+                    Dictionary<StringSpanOrdinalKey, int> vocab,
+                    Dictionary<int, string> vocabReverse,
+                    PreTokenizer? preTokenizer,
+                    Normalizer? normalizer,
+                    IReadOnlyDictionary<string, int>? specialTokens,
+                    bool doLowerCase,
+                    bool doBasicTokenization,
+                    bool splitOnSpecialTokens,
+                    string unknownToken,
+                    string sepToken,
+                    string padToken,
+                    string clsToken,
+                    string maskToken,
+                    bool tokenizeChineseChars,
+                    bool stripAccents) : base(vocab, vocabReverse, preTokenizer, normalizer, specialTokens, unknownToken)
+        {
+            DoLowerCase = doLowerCase;
+            DoBasicTokenization = doBasicTokenization;
+            SplitOnSpecialTokens = splitOnSpecialTokens;
+
+            SepToken = sepToken;
+            SepTokenId = vocab[new StringSpanOrdinalKey(sepToken)];
+
+            PadToken = padToken;
+            PadTokenId = vocab[new StringSpanOrdinalKey(padToken)];
+
+            ClsToken = clsToken;
+            ClsTokenId = vocab[new StringSpanOrdinalKey(clsToken)];
+
+            MaskToken = maskToken;
+            MaskTokenId = vocab[new StringSpanOrdinalKey(maskToken)];
+
+            TokenizeChineseChars = tokenizeChineseChars;
+            StripAccents = stripAccents;
+        }
+
+        /// <summary>
+        /// Gets a value indicating whether the tokenizer should lowercase the input text.
+        /// </summary>
+        public bool DoLowerCase { get; }
+
+        /// <summary>
+        /// Gets a value indicating whether the tokenizer should do basic tokenization. Like clean text, normalize it, lowercasing, etc.
+        /// </summary>
+        public bool DoBasicTokenization { get; }
+
+        /// <summary>
+        /// Gets a value indicating whether the tokenizer should split on the special tokens or treat special tokens as normal text.
+        /// </summary>
+        public bool SplitOnSpecialTokens { get; }
+
+        /// <summary>
+        /// Gets the separator token, which is used when building a sequence from multiple sequences, e.g. two sequences for sequence classification or for a text and a question for question answering.
+        /// It is also used as the last token of a sequence built with special tokens.
+        /// </summary>
+        public string SepToken { get; }
+
+        /// <summary>
+        /// Gets the separator token Id
+        /// </summary>
+        public int SepTokenId { get; }
+
+        /// <summary>
+        /// Gets the token used for padding, for example when batching sequences of different lengths
+        /// </summary>
+        public string PadToken { get; }
+
+        /// <summary>
+        /// Gets padding token Id
+        /// </summary>
+        public int PadTokenId { get; }
+
+        /// <summary>
+        /// Gets the classifier token which is used when doing sequence classification (classification of the whole sequence instead of per-token classification).
+        /// It is the first token of the sequence when built with special tokens.
+        /// </summary>
+        public string ClsToken { get; }
+
+        /// <summary>
+        /// Gets the classifier token Id
+        /// </summary>
+        public int ClsTokenId { get; }
+
+        /// <summary>
+        /// Gets the mask token used for masking values. This is the token used when training this model with masked language modeling.
+        /// This is the token which the model will try to predict.
+        /// </summary>
+        public string MaskToken { get; }
+
+        /// <summary>
+        /// Gets the mask token Id
+        /// </summary>
+        public int MaskTokenId { get; }
+
+        /// <summary>
+        /// Gets a value indicating whether the tokenizer should split the Chinese characters into tokens.
+        /// </summary>
+        public bool TokenizeChineseChars { get; }
+
+        /// <summary>
+        /// Gets a value indicating whether the tokenizer should strip accents characters.
+        /// </summary>
+        public bool StripAccents { get; }
+
+        /// <summary>
+        /// Encodes input text to token Ids.
+        /// </summary>
+        /// <param name="text">The text to encode.</param>
+        /// <param name="considerPreTokenization">Indicate whether to consider pre-tokenization before tokenization.</param>
+        /// <param name="considerNormalization">Indicate whether to consider normalization before tokenization.</param>
+        /// <returns>The list of encoded Ids.</returns>
+        public new IReadOnlyList<int> EncodeToIds(string text, bool considerPreTokenization = true, bool considerNormalization = true) =>
+            EncodeToIds(text, ReadOnlySpan<char>.Empty, addSpecialTokens: true, considerPreTokenization, considerNormalization);
+
+        /// <summary>
+        /// Encodes input text to token Ids.
+        /// </summary>
+        /// <param name="text">The text to encode.</param>
+        /// <param name="considerPreTokenization">Indicate whether to consider pre-tokenization before tokenization.</param>
+        /// <param name="considerNormalization">Indicate whether to consider normalization before tokenization.</param>
+        /// <returns>The list of encoded Ids.</returns>
+        public new IReadOnlyList<int> EncodeToIds(ReadOnlySpan<char> text, bool considerPreTokenization = true, bool considerNormalization = true) =>
+            EncodeToIds(null, text, addSpecialTokens: true, considerPreTokenization, considerNormalization);
+
+        /// <summary>
+        /// Encodes input text to token Ids.
+        /// </summary>
+        /// <param name="text">The text to encode.</param>
+        /// <param name="addSpecialTokens">Indicate whether to add special tokens to the encoded Ids.</param>
+        /// <param name="considerPreTokenization">Indicate whether to consider pre-tokenization before tokenization.</param>
+        /// <param name="considerNormalization">Indicate whether to consider normalization before tokenization.</param>
+        /// <returns>The list of encoded Ids.</returns>
+        public IReadOnlyList<int> EncodeToIds(string text, bool addSpecialTokens, bool considerPreTokenization = true, bool considerNormalization = true) =>
+            EncodeToIds(text, ReadOnlySpan<char>.Empty, addSpecialTokens, considerPreTokenization, considerNormalization);
+
+        /// <summary>
+        /// Encodes input text to token Ids.
+        /// </summary>
+        /// <param name="text">The text to encode.</param>
+        /// <param name="addSpecialTokens">Indicate whether to add special tokens to the encoded Ids.</param>
+        /// <param name="considerPreTokenization">Indicate whether to consider pre-tokenization before tokenization.</param>
+        /// <param name="considerNormalization">Indicate whether to consider normalization before tokenization.</param>
+        /// <returns>The list of encoded Ids.</returns>
+        public IReadOnlyList<int> EncodeToIds(ReadOnlySpan<char> text, bool addSpecialTokens, bool considerPreTokenization = true, bool considerNormalization = true) =>
+            EncodeToIds(null, text, addSpecialTokens, considerPreTokenization, considerNormalization);
+
+        /// <summary>
+        /// Encodes input text to token Ids.
+        /// </summary>
+        /// <param name="text">The text to encode.</param>
+        /// <param name="maxTokenCount">The maximum number of tokens to return.</param>
+        /// <param name="normalizedText">The normalized text.</param>
+        /// <param name="charsConsumed">The number of characters consumed from the input text.</param>
+        /// <param name="considerPreTokenization">Indicate whether to consider pre-tokenization before tokenization.</param>
+        /// <param name="considerNormalization">Indicate whether to consider normalization before tokenization.</param>
+        /// <returns>The list of encoded Ids.</returns>
+        public new IReadOnlyList<int> EncodeToIds(string text, int maxTokenCount, out string? normalizedText, out int charsConsumed, bool considerPreTokenization = true, bool considerNormalization = true) =>
+            EncodeToIds(text, ReadOnlySpan<char>.Empty, maxTokenCount, addSpecialTokens: true, out normalizedText, out charsConsumed, considerPreTokenization, considerNormalization);
+
+        /// <summary>
+        /// Encodes input text to token Ids.
+        /// </summary>
+        /// <param name="text">The text to encode.</param>
+        /// <param name="maxTokenCount">The maximum number of tokens to return.</param>
+        /// <param name="normalizedText">The normalized text.</param>
+        /// <param name="charsConsumed">The number of characters consumed from the input text.</param>
+        /// <param name="considerPreTokenization">Indicate whether to consider pre-tokenization before tokenization.</param>
+        /// <param name="considerNormalization">Indicate whether to consider normalization before tokenization.</param>
+        /// <returns>The list of encoded Ids.</returns>
+        public new IReadOnlyList<int> EncodeToIds(ReadOnlySpan<char> text, int maxTokenCount, out string? normalizedText, out int charsConsumed, bool considerPreTokenization = true, bool considerNormalization = true) =>
+            EncodeToIds(null, text, maxTokenCount, addSpecialTokens: true, out normalizedText, out charsConsumed, considerPreTokenization, considerNormalization);
+
+        /// <summary>
+        /// Encodes input text to token Ids.
+        /// </summary>
+        /// <param name="text">The text to encode.</param>
+        /// <param name="maxTokenCount">The maximum number of tokens to return.</param>
+        /// <param name="addSpecialTokens">Indicate whether to add special tokens to the encoded Ids.</param>
+        /// <param name="normalizedText">The normalized text.</param>
+        /// <param name="charsConsumed">The number of characters consumed from the input text.</param>
+        /// <param name="considerPreTokenization">Indicate whether to consider pre-tokenization before tokenization.</param>
+        /// <param name="considerNormalization">Indicate whether to consider normalization before tokenization.</param>
+        /// <returns>The list of encoded Ids.</returns>
+        public IReadOnlyList<int> EncodeToIds(string text, int maxTokenCount, bool addSpecialTokens, out string? normalizedText, out int charsConsumed, bool considerPreTokenization = true, bool considerNormalization = true) =>
+            EncodeToIds(text, ReadOnlySpan<char>.Empty, maxTokenCount, addSpecialTokens, out normalizedText, out charsConsumed, considerPreTokenization, considerNormalization);
+
+        /// <summary>
+        /// Encodes input text to token Ids.
+        /// </summary>
+        /// <param name="text">The text to encode.</param>
+        /// <param name="maxTokenCount">The maximum number of tokens to return.</param>
+        /// <param name="addSpecialTokens">Indicate whether to add special tokens to the encoded Ids.</param>
+        /// <param name="normalizedText">The normalized text.</param>
+        /// <param name="charsConsumed">The number of characters consumed from the input text.</param>
+        /// <param name="considerPreTokenization">Indicate whether to consider pre-tokenization before tokenization.</param>
+        /// <param name="considerNormalization">Indicate whether to consider normalization before tokenization.</param>
+        /// <returns>The list of encoded Ids.</returns>
+        public IReadOnlyList<int> EncodeToIds(ReadOnlySpan<char> text, int maxTokenCount, bool addSpecialTokens, out string? normalizedText, out int charsConsumed, bool considerPreTokenization = true, bool considerNormalization = true) =>
+            EncodeToIds(null, text, maxTokenCount, addSpecialTokens, out normalizedText, out charsConsumed, considerPreTokenization, considerNormalization);
+
+        private IReadOnlyList<int> EncodeToIds(string? text, ReadOnlySpan<char> textSpan, int maxTokenCount, bool addSpecialTokens, out string? normalizedText, out int charsConsumed, bool considerPreTokenization = true, bool considerNormalization = true)
+        {
+            if (addSpecialTokens)
+            {
+                if (maxTokenCount < 2)
+                {
+                    charsConsumed = 0;
+                    normalizedText = null;
+                    return Array.Empty<int>();
+                }
+
+                IReadOnlyList<int> ids = text is null ?
+                                            base.EncodeToIds(textSpan, maxTokenCount - 2, out normalizedText, out charsConsumed, considerPreTokenization, considerNormalization) :
+                                            base.EncodeToIds(text, maxTokenCount - 2, out normalizedText, out charsConsumed, considerPreTokenization, considerNormalization);
+
+                if (ids is not List<int> list)
+                {
+                    list = new List<int>(ids);
+                }
+
+                list.Insert(0, ClsTokenId);
+                list.Add(SepTokenId);
+
+                return list;
+            }
+
+            return text is null ?
+                    base.EncodeToIds(textSpan, maxTokenCount, out normalizedText, out charsConsumed, considerPreTokenization, considerNormalization) :
+                    base.EncodeToIds(text, maxTokenCount, out normalizedText, out charsConsumed, considerPreTokenization, considerNormalization);
+        }
+
+        private IReadOnlyList<int> EncodeToIds(string? text, ReadOnlySpan<char> textSpan, bool addSpecialTokens, bool considerPreTokenization = true, bool considerNormalization = true)
+        {
+            IReadOnlyList<int> ids = text is null ? base.EncodeToIds(textSpan, considerPreTokenization, considerNormalization) : base.EncodeToIds(text, considerPreTokenization, considerNormalization);
+
+            if (addSpecialTokens)
+            {
+                if (ids is not List<int> list)
+                {
+                    list = new List<int>(ids);
+                }
+
+                list.Insert(0, ClsTokenId);
+                list.Add(SepTokenId);
+
+                return list;
+            }
+
+            return ids;
+        }
+
+        /// <summary>
+        /// Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and adding special tokens. A BERT sequence has the following format:
+        ///     - single sequence: `[CLS] tokenIds0 [SEP]`
+        ///     - pair of sequences: `[CLS] tokenIds0 [SEP] tokenIds1 [SEP]`
+        /// </summary>
+        /// <param name="tokenIds0">List of IDs to which the special tokens will be added.</param>
+        /// <param name="tokenIds1">Optional second list of IDs for sequence pairs.</param>
+        /// <returns>The list of IDs with special tokens added.</returns>
+        /// <exception cref="ArgumentNullException">When <paramref name="tokenIds0"/> is null.</exception>
+        public IReadOnlyList<int> BuildInputsWithSpecialTokens(IEnumerable<int> tokenIds0, IEnumerable<int>? tokenIds1 = null)
+        {
+            if (tokenIds0 is null)
+            {
+                throw new ArgumentNullException(nameof(tokenIds0));
+            }
+
+            // Add 2 for [CLS] and [SEP] tokens. Add 1 for [SEP] token if tokenIds1 is not null.
+            int capacity = tokenIds0.Count() + 2 + (tokenIds1 is null ? 0 : tokenIds1.Count() + 1);
+            List<int> ids = new List<int>(capacity: capacity) { ClsTokenId };
+            ids.AddRange(tokenIds0);
+            ids.Add(SepTokenId);
+
+            if (tokenIds1 is not null)
+            {
+                ids.AddRange(tokenIds1);
+                ids.Add(SepTokenId);
+            }
+
+            return ids;
+        }
+
+        /// <summary>
+        /// Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and adding special tokens. A BERT sequence has the following format:
+        ///     - single sequence: `[CLS] tokenIds0 [SEP]`
+        ///     - pair of sequences: `[CLS] tokenIds0 [SEP] tokenIds1 [SEP]`
+        /// </summary>
+        /// <param name="tokenIds0">List of IDs to which the special tokens will be added.</param>
+        /// <param name="buffer">The buffer to write the token IDs with special tokens added.</param>
+        /// <param name="written">The number of elements written to the buffer.</param>
+        /// <param name="tokenIds1">Optional second list of IDs for sequence pairs.</param>
+        /// <returns>The status of the operation.</returns>
+        /// <exception cref="ArgumentNullException">When <paramref name="tokenIds0"/> is null.</exception>
+        public OperationStatus BuildInputsWithSpecialTokens(IEnumerable<int> tokenIds0, Span<int> buffer, out int written, IEnumerable<int>? tokenIds1 = null)
+        {
+            if (tokenIds0 is null)
+            {
+                throw new ArgumentNullException(nameof(tokenIds0));
+            }
+
+            // Add 2 for [CLS] and [SEP] tokens. Add 1 for [SEP] token if tokenIds1 is not null.
+            int capacity = tokenIds0.Count() + 2 + (tokenIds1 is null ? 0 : tokenIds1.Count() + 1);
+            if (buffer.Length < capacity)
+            {
+                written = 0;
+                return OperationStatus.DestinationTooSmall;
+            }
+
+            written = 0;
+            buffer[written++] = ClsTokenId;
+            foreach (int id in tokenIds0)
+            {
+                buffer[written++] = id;
+            }
+            buffer[written++] = SepTokenId;
+
+            if (tokenIds1 is not null)
+            {
+                foreach (int id in tokenIds1)
+                {
+                    buffer[written++] = id;
+                }
+
+                buffer[written++] = SepTokenId;
+            }
+
+            return OperationStatus.Done;
+        }
+
+        /// <summary>
+        /// Retrieve sequence tokens mask from a IDs list.
+        /// </summary>
+        /// <param name="tokenIds0">List of IDs.</param>
+        /// <param name="tokenIds1">Optional second list of IDs for sequence pairs.</param>
+        /// <param name="alreadyHasSpecialTokens">Indicate whether or not the token list is already formatted with special tokens for the model.</param>
+        /// <returns>A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.</returns>
+        /// <exception cref="ArgumentNullException"></exception>
+        public IReadOnlyList<int> GetSpecialTokensMask(IEnumerable<int> tokenIds0, IEnumerable<int>? tokenIds1 = null, bool alreadyHasSpecialTokens = false)
+        {
+            if (tokenIds0 is null)
+            {
+                throw new ArgumentNullException(nameof(tokenIds0));
+            }
+
+            int capacity = alreadyHasSpecialTokens ?
+                        tokenIds0.Count() + (tokenIds1?.Count() ?? 0) :
+                        tokenIds0.Count() + 2 + (tokenIds1 is null ? 0 : 1);    // Add 2 for [CLS] and [SEP] tokens. Add 1 for [SEP] token if tokenIds1 is not null.
+
+            List<int> mask = new List<int>(capacity: capacity);
+
+            if (!alreadyHasSpecialTokens)
+            {
+                mask.Add(1); // CLS
+                mask.AddRange(Enumerable.Repeat(0, tokenIds0.Count()));
+                mask.Add(1); // SEP
+                if (tokenIds1 is not null)
+                {
+                    mask.AddRange(Enumerable.Repeat(0, tokenIds1.Count()));
+                    mask.Add(1); // SEP
+                }
+
+                return mask;
+            }
+
+            foreach (int id in tokenIds0)
+            {
+                mask.Add(id == ClsTokenId || id == SepTokenId || id == PadTokenId || id == MaskTokenId || id == UnknownTokenId ? 1 : 0);
+            }
+
+            if (tokenIds1 is not null)
+            {
+                foreach (int id in tokenIds1)
+                {
+                    mask.Add(id == ClsTokenId || id == SepTokenId || id == PadTokenId || id == MaskTokenId || id == UnknownTokenId ? 1 : 0);
+                }
+            }
+
+            return mask;
+        }
+
+        /// <summary>
+        /// Retrieve sequence tokens mask from a IDs list.
+        /// </summary>
+        /// <param name="tokenIds0">List of IDs.</param>
+        /// <param name="buffer">The buffer to write the mask. The integers written values are in the range [0, 1]: 1 for a special token, 0 for a sequence token.</param>
+        /// <param name="written">The number of elements written to the buffer.</param>
+        /// <param name="tokenIds1">Optional second list of IDs for sequence pairs.</param>
+        /// <param name="alreadyHasSpecialTokens">Indicate whether or not the token list is already formatted with special tokens for the model.</param>
+        /// <returns>The status of the operation.</returns>
+        /// <exception cref="ArgumentNullException"></exception>
+        public OperationStatus GetSpecialTokensMask(IEnumerable<int> tokenIds0, Span<int> buffer, out int written, IEnumerable<int>? tokenIds1 = null, bool alreadyHasSpecialTokens = false)
+        {
+            if (tokenIds0 is null)
+            {
+                throw new ArgumentNullException(nameof(tokenIds0));
+            }
+
+            int capacity = alreadyHasSpecialTokens ?
+                        tokenIds0.Count() + (tokenIds1?.Count() ?? 0) :
+                        tokenIds0.Count() + 2 + (tokenIds1 is null ? 0 : tokenIds1.Count() + 1);    // Add 2 for [CLS] and [SEP] tokens. Add 1 for [SEP] token if tokenIds1 is not null.
+
+            written = 0;
+            if (buffer.Length < capacity)
+            {
+                return OperationStatus.DestinationTooSmall;
+            }
+
+            if (!alreadyHasSpecialTokens)
+            {
+                buffer[written++] = 1; // CLS
+                foreach (int id in tokenIds0)
+                {
+                    buffer[written++] = 0;
+                }
+                buffer[written++] = 1; // SEP
+
+                if (tokenIds1 is not null)
+                {
+                    foreach (int id in tokenIds1)
+                    {
+                        buffer[written++] = 0;
+                    }
+                    buffer[written++] = 1; // SEP
+                }
+
+                return OperationStatus.Done;
+            }
+
+            foreach (int id in tokenIds0)
+            {
+                buffer[written++] = id == ClsTokenId || id == SepTokenId || id == PadTokenId || id == MaskTokenId || id == UnknownTokenId ? 1 : 0;
+            }
+
+            if (tokenIds1 is not null)
+            {
+                foreach (int id in tokenIds1)
+                {
+                    buffer[written++] = id == ClsTokenId || id == SepTokenId || id == PadTokenId || id == MaskTokenId || id == UnknownTokenId ? 1 : 0;
+                }
+            }
+
+            return OperationStatus.Done;
+        }
+
+        /// <summary>
+        /// Create a mask from the two sequences passed to be used in a sequence-pair classification task. A BERT sequence pair mask has the following format:
+        ///         0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1
+        ///         | first sequence    | second sequence |
+        /// If <paramref name="tokenIds1"/> is null, this method only returns the first portion of the type ids (0s).
+        /// </summary>
+        /// <param name="tokenIds0">List of token IDs for the first sequence.</param>
+        /// <param name="tokenIds1">Optional list of token IDs for the second sequence.</param>
+        /// <returns>List of token type IDs according to the given sequence(s).</returns>
+        /// <exception cref="ArgumentNullException">When <paramref name="tokenIds0"/> is null.</exception>
+        public IReadOnlyList<int> CreateTokenTypeIdsFromSequences(IEnumerable<int> tokenIds0, IEnumerable<int>? tokenIds1 = null)
+        {
+            if (tokenIds0 is null)
+            {
+                throw new ArgumentNullException(nameof(tokenIds0));
+            }
+
+            // Add 2 for [CLS] and [SEP] tokens. Add 1 for [SEP] token if tokenIds1 is not null.
+            int capacity = tokenIds0.Count() + 2 + (tokenIds1 is null ? 0 : tokenIds1.Count() + 1);
+
+            List<int> typeIds = new List<int>(capacity);
+            for (int i = 0; i < tokenIds0.Count() + 2; i++) // Add 2 for [CLS] and [SEP] tokens.
+            {
+                typeIds.Add(0);
+            }
+
+            if (tokenIds1 is not null)
+            {
+                for (int i = 0; i < tokenIds1.Count() + 1; i++) // Add 1 for [SEP] token.
+                {
+                    typeIds.Add(1);
+                }
+            }
+
+            return typeIds;
+        }
+
+        public OperationStatus CreateTokenTypeIdsFromSequences(IEnumerable<int> tokenIds0, Span<int> buffer, out int written, IEnumerable<int>? tokenIds1 = null)
+        {
+            if (tokenIds0 is null)
+            {
+                throw new ArgumentNullException(nameof(tokenIds0));
+            }
+
+            written = 0;
+
+            // Add 2 for [CLS] and [SEP] tokens. Add 1 for [SEP] token if tokenIds1 is not null.
+            int capacity = tokenIds0.Count() + 2 + (tokenIds1 is null ? 0 : tokenIds1.Count() + 1);
+            if (buffer.Length < capacity)
+            {
+                return OperationStatus.DestinationTooSmall;
+            }
+
+            for (int i = 0; i < tokenIds0.Count() + 2; i++) // Add 2 for [CLS] and [SEP] tokens.
+            {
+                buffer[written++] = 0;
+            }
+
+            if (tokenIds1 is not null)
+            {
+                for (int i = 0; i < tokenIds1.Count() + 1; i++) // Add 1 for [SEP] token.
+                {
+                    buffer[written++] = 1;
+                }
+            }
+
+            return OperationStatus.Done;
+        }
+
+        /// <summary>
+        /// Create a new instance of the <see cref="BertTokenizer"/> class.
+        /// </summary>
+        /// <param name="vocabFilePath">The path to the vocabulary file.</param>
+        /// <param name="doLowerCase">A value indicating whether the tokenizer should lowercase the input text.</param>
+        /// <param name="doBasicTokenization">A value indicating whether the tokenizer should do basic tokenization. Like clean text, normalize it, lowercasing, etc.</param>
+        /// <param name="splitOnSpecialTokens">A value indicating whether the tokenizer should split on special tokens.</param>
+        /// <param name="unknownToken">The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this token instead.</param>
+        /// <param name="sepToken">The separator token, which is used when building a sequence from multiple sequences, e.g. two sequences for sequence classification or for a text and a question for question answering. It is also used as the last token of a sequence built with special tokens.</param>
+        /// <param name="padToken">The token used for padding, for example when batching sequences of different lengths.</param>
+        /// <param name="clsToken">The classifier token which is used when doing sequence classification (classification of the whole sequence instead of per-token classification). It is the first token of the sequence when built with special tokens.</param>
+        /// <param name="maskToken">The token used for masking values. This is the token used when training this model with masked language modeling. This is the token which the model will try to predict.</param>
+        /// <param name="tokenizeChineseChars">A value indicating whether the tokenizer should split the Chinese characters into tokens.</param>
+        /// <param name="stripAccents">A value indicating whether the tokenizer should strip accents characters.</param>
+        /// <returns></returns>
+        /// <exception cref="ArgumentNullException"></exception>
+        public static BertTokenizer Create(
+                    string vocabFilePath,
+                    bool doLowerCase = true,
+                    bool doBasicTokenization = true,
+                    bool splitOnSpecialTokens = true,
+                    string unknownToken = "[UNK]",
+                    string sepToken = "[SEP]",
+                    string padToken = "[PAD]",
+                    string clsToken = "[CLS]",
+                    string maskToken = "[MASK]",
+                    bool tokenizeChineseChars = true,
+                    bool stripAccents = false) =>
+            Create(
+                string.IsNullOrEmpty(vocabFilePath) ? throw new ArgumentNullException(nameof(vocabFilePath)) : File.OpenRead(vocabFilePath),
+                doLowerCase, doBasicTokenization, splitOnSpecialTokens, unknownToken, sepToken, padToken, clsToken, maskToken, tokenizeChineseChars, stripAccents, disposeStream: true);
+
+        /// <summary>
+        /// Create a new instance of the <see cref="BertTokenizer"/> class.
+        /// </summary>
+        /// <param name="vocabStream">The stream containing the vocabulary file.</param>
+        /// <param name="doLowerCase">A value indicating whether the tokenizer should lowercase the input text.</param>
+        /// <param name="doBasicTokenization">A value indicating whether the tokenizer should do basic tokenization. Like clean text, normalize it, lowercasing, etc.</param>
+        /// <param name="splitOnSpecialTokens">A value indicating whether the tokenizer should split on special tokens.</param>
+        /// <param name="unknownToken">The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this token instead.</param>
+        /// <param name="sepToken">The separator token, which is used when building a sequence from multiple sequences, e.g. two sequences for sequence classification or for a text and a question for question answering. It is also used as the last token of a sequence built with special tokens.</param>
+        /// <param name="padToken">The token used for padding, for example when batching sequences of different lengths.</param>
+        /// <param name="clsToken">The classifier token which is used when doing sequence classification (classification of the whole sequence instead of per-token classification). It is the first token of the sequence when built with special tokens.</param>
+        /// <param name="maskToken">The token used for masking values. This is the token used when training this model with masked language modeling. This is the token which the model will try to predict.</param>
+        /// <param name="tokenizeChineseChars">A value indicating whether the tokenizer should split the Chinese characters into tokens.</param>
+        /// <param name="stripAccents">A value indicating whether the tokenizer should strip accents characters.</param>
+        /// <returns></returns>
+        /// <exception cref="ArgumentNullException"></exception>
+        public static BertTokenizer Create(
+                    Stream vocabStream,
+                    bool doLowerCase = true,
+                    bool doBasicTokenization = true,
+                    bool splitOnSpecialTokens = true,
+                    string unknownToken = "[UNK]",
+                    string sepToken = "[SEP]",
+                    string padToken = "[PAD]",
+                    string clsToken = "[CLS]",
+                    string maskToken = "[MASK]",
+                    bool tokenizeChineseChars = true,
+                    bool stripAccents = false) =>
+            Create(vocabStream, doLowerCase, doBasicTokenization, splitOnSpecialTokens, unknownToken, sepToken, padToken, clsToken, maskToken, tokenizeChineseChars, stripAccents, disposeStream: false);
+
+        /// <summary>
+        /// Create a new instance of the <see cref="BertTokenizer"/> class asynchronously.
+        /// </summary>
+        /// <param name="vocabStream">The stream containing the vocabulary file.</param>
+        /// <param name="doLowerCase">A value indicating whether the tokenizer should lowercase the input text.</param>
+        /// <param name="doBasicTokenization">A value indicating whether the tokenizer should do basic tokenization. Like clean text, normalize it, lowercasing, etc.</param>
+        /// <param name="splitOnSpecialTokens">A value indicating whether the tokenizer should split on special tokens.</param>
+        /// <param name="unknownToken">The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this token instead.</param>
+        /// <param name="sepToken">The separator token, which is used when building a sequence from multiple sequences, e.g. two sequences for sequence classification or for a text and a question for question answering. It is also used as the last token of a sequence built with special tokens.</param>
+        /// <param name="padToken">The token used for padding, for example when batching sequences of different lengths.</param>
+        /// <param name="clsToken">The classifier token which is used when doing sequence classification (classification of the whole sequence instead of per-token classification). It is the first token of the sequence when built with special tokens.</param>
+        /// <param name="maskToken">The token used for masking values. This is the token used when training this model with masked language modeling. This is the token which the model will try to predict.</param>
+        /// <param name="tokenizeChineseChars">A value indicating whether the tokenizer should split the Chinese characters into tokens.</param>
+        /// <param name="stripAccents">A value indicating whether the tokenizer should strip accents characters.</param>
+        /// <returns></returns>
+        /// <exception cref="ArgumentNullException"></exception>
+        public static async Task<BertTokenizer> CreateAsync(
+                    Stream vocabStream,
+                    bool doLowerCase = true,
+                    bool doBasicTokenization = true,
+                    bool splitOnSpecialTokens = true,
+                    string unknownToken = "[UNK]",
+                    string sepToken = "[SEP]",
+                    string padToken = "[PAD]",
+                    string clsToken = "[CLS]",
+                    string maskToken = "[MASK]",
+                    bool tokenizeChineseChars = true,
+                    bool stripAccents = false)
+        {
+            if (vocabStream is null)
+            {
+                throw new ArgumentNullException(nameof(vocabStream));
+            }
+
+            (Dictionary<StringSpanOrdinalKey, int> vocab, Dictionary<int, string> vocabReverse) = await LoadVocabAsync(vocabStream, useAsync: true).ConfigureAwait(false);
+
+            return Create(vocab, vocabReverse, doLowerCase, doBasicTokenization, splitOnSpecialTokens, unknownToken, sepToken, padToken, clsToken, maskToken, tokenizeChineseChars, stripAccents);
+        }
+
+        private static BertTokenizer Create(
+                            Stream vocabStream,
+                            bool doLowerCase,
+                            bool doBasicTokenization,
+                            bool splitOnSpecialTokens,
+                            string unknownToken,
+                            string sepToken,
+                            string padToken,
+                            string clsToken,
+                            string maskToken,
+                            bool tokenizeChineseChars,
+                            bool stripAccents,
+                            bool disposeStream)
+        {
+            if (vocabStream is null)
+            {
+                throw new ArgumentNullException(nameof(vocabStream));
+            }
+
+            try
+            {
+                (Dictionary<StringSpanOrdinalKey, int> vocab, Dictionary<int, string> vocabReverse) = LoadVocabAsync(vocabStream, useAsync: false).GetAwaiter().GetResult();
+
+                return Create(vocab, vocabReverse, doLowerCase, doBasicTokenization, splitOnSpecialTokens, unknownToken, sepToken, padToken, clsToken, maskToken, tokenizeChineseChars, stripAccents);
+            }
+            finally
+            {
+                if (disposeStream)
+                {
+                    vocabStream.Dispose();
+                }
+            }
+        }
+
+        private static BertTokenizer Create(
+                    Dictionary<StringSpanOrdinalKey, int> vocab,
+                    Dictionary<int, string> vocabReverse,
+                    bool doLowerCase,
+                    bool doBasicTokenization,
+                    bool splitOnSpecialTokens,
+                    string unknownToken,
+                    string sepToken,
+                    string padToken,
+                    string clsToken,
+                    string maskToken,
+                    bool tokenizeChineseChars,
+                    bool stripAccents)
+        {
+            Normalizer? normalizer = doBasicTokenization ? new BertNormalizer(doLowerCase, tokenizeChineseChars, stripAccents) : null;
+
+            Dictionary<string, int>? specialTokens = new();
+            bool lowerCase = doBasicTokenization && doLowerCase && splitOnSpecialTokens;
+
+            AddSpecialToken(vocab, specialTokens, unknownToken, lowerCase);
+            AddSpecialToken(vocab, specialTokens, sepToken, lowerCase);
+            AddSpecialToken(vocab, specialTokens, padToken, lowerCase);
+            AddSpecialToken(vocab, specialTokens, clsToken, lowerCase);
+            AddSpecialToken(vocab, specialTokens, maskToken, lowerCase);
+
+            PreTokenizer? preTokenizer = doBasicTokenization ?
+                                            PreTokenizer.CreateWhiteSpaceOrPunctuationPreTokenizer(splitOnSpecialTokens ? specialTokens : null) :
+                                            PreTokenizer.CreateWhiteSpacePreTokenizer();
+
+            return new BertTokenizer(vocab, vocabReverse, preTokenizer, normalizer, specialTokens, doLowerCase, doBasicTokenization,
+                                    splitOnSpecialTokens, unknownToken, sepToken, padToken, clsToken, maskToken, tokenizeChineseChars, stripAccents);
+        }
+
+        private static void AddSpecialToken(Dictionary<StringSpanOrdinalKey, int> vocab, Dictionary<string, int> specialTokens, string token, bool lowerCase)
+        {
+            if (token is null || !vocab.TryGetValue(new StringSpanOrdinalKey(token), out int id))
+            {
+                throw new ArgumentException($"The special token '{token}' is not in the vocabulary.");
+            }
+
+            string normalizedToken = token;
+            if (lowerCase)
+            {
+                // Lowercase the special tokens to have the pre-tokenization can find them as we lowercase the input text.
+                // we don't even need to do case-insensitive comparisons as we are lowercasing the input text.
+                normalizedToken = token.ToLowerInvariant();
+
+                // Add lowercased special tokens to the vocab if they are not already there.
+                // This will allow matching during the encoding process.
+                vocab[new StringSpanOrdinalKey(normalizedToken)] = id;
+            }
+
+            specialTokens[normalizedToken] = id;
+        }
+    }
+}

src/Microsoft.ML.Tokenizers/Model/WordPieceTokenizer.cs

@@ -0,0 +1,858 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+using System;
+using System.Buffers;
+using System.Collections.Generic;
+using System.Diagnostics;
+using System.Globalization;
+using System.IO;
+using System.Linq;
+using System.Text;
+using System.Threading;
+using System.Threading.Tasks;
+
+namespace Microsoft.ML.Tokenizers
+{
+    /// <summary>
+    /// Represent the WordPiece tokenizer.
+    /// </summary>
+    /// <remarks>
+    /// The WordPiece tokenizer is a sub-word tokenizer that is used in BERT and other transformer models.
+    /// The implementation is based on the Hugging Face WordPiece tokenizer https://huggingface.co/docs/tokenizers/api/models#tokenizers.models.WordPiece.
+    /// </remarks>
+    public partial class WordPieceTokenizer : Tokenizer
+    {
+        private readonly PreTokenizer? _preTokenizer;
+        private readonly Normalizer? _normalizer;
+        private readonly Dictionary<StringSpanOrdinalKey, int> _vocab;
+        private readonly Dictionary<int, string> _vocabReverse;
+
+        internal const string DefaultContinuingSubwordPrefix = "##";
+        internal const int DefaultMaxInputCharsPerWord = 100;
+
+        internal WordPieceTokenizer(
+                    Dictionary<StringSpanOrdinalKey, int> vocab,
+                    Dictionary<int, string> vocabReverse,
+                    PreTokenizer? preTokenizer,
+                    Normalizer? normalizer,
+                    IReadOnlyDictionary<string, int>? specialTokens,
+                    string unknownToken,
+                    string continuingSubwordPrefix = DefaultContinuingSubwordPrefix,
+                    int maxInputCharsPerWord = DefaultMaxInputCharsPerWord)
+        {
+            Debug.Assert(vocab is not null);
+            Debug.Assert(vocabReverse is not null);
+            _vocab = vocab!;
+            _vocabReverse = vocabReverse!;
+            SpecialTokens = specialTokens;
+            SpecialTokensReverse = specialTokens is not null ? specialTokens.ToDictionary(kvp => kvp.Value, kvp => kvp.Key) : null;
+
+            if (unknownToken is null)
+            {
+                throw new ArgumentNullException(nameof(unknownToken));
+            }
+
+            if (continuingSubwordPrefix is null)
+            {
+                throw new ArgumentNullException(nameof(continuingSubwordPrefix));
+            }
+
+            if (maxInputCharsPerWord <= 0)
+            {
+                throw new ArgumentOutOfRangeException(nameof(maxInputCharsPerWord), "The maximum number of characters per word must be greater than zero.");
+            }
+
+            if (!vocab!.TryGetValue(unknownToken, out int id))
+            {
+                throw new ArgumentException($"The unknown token '{unknownToken}' is not in the vocabulary.");
+            }
+
+            UnknownToken = unknownToken;
+            UnknownTokenId = id;
+            ContinuingSubwordPrefix = continuingSubwordPrefix;
+            MaxInputCharsPerWord = maxInputCharsPerWord;
+
+            _preTokenizer = preTokenizer ?? PreTokenizer.CreateWhiteSpacePreTokenizer(specialTokens);
+            _normalizer = normalizer;
+        }
+
+        /// <summary>
+        /// Gets the unknown token ID.
+        /// A token that is not in the vocabulary cannot be converted to an ID and is set to be this token instead.
+        /// </summary>
+        public int UnknownTokenId { get; }
+
+        /// <summary>
+        /// Gets the prefix to use for sub-words that are not the first part of a word.
+        /// </summary>
+        public string ContinuingSubwordPrefix { get; }
+
+        /// <summary>
+        /// Gets the maximum number of characters to authorize in a single word.
+        /// </summary>
+        public int MaxInputCharsPerWord { get; }
+
+        internal static async ValueTask<(Dictionary<StringSpanOrdinalKey, int>, Dictionary<int, string>)> LoadVocabAsync(Stream vocabStream, bool useAsync, CancellationToken cancellationToken = default)
+        {
+            if (vocabStream is null)
+            {
+                throw new ArgumentNullException(nameof(vocabStream));
+            }
+
+            Dictionary<StringSpanOrdinalKey, int> vocab = new Dictionary<StringSpanOrdinalKey, int>();
+            Dictionary<int, string> vocabReverse = new Dictionary<int, string>();
+
+            StreamReader reader = new StreamReader(vocabStream);
+            string? line = useAsync ? await Helpers.ReadLineAsync(reader, cancellationToken).ConfigureAwait(false) : reader.ReadLine();
+            int lineNumber = 0;
+
+            while (line is not null)
+            {
+                if (line.Length != 0)
+                {
+                    vocab.Add(new StringSpanOrdinalKey(line), lineNumber);
+                    vocabReverse.Add(lineNumber, line);
+                }
+
+                lineNumber++;
+                line = useAsync ? await Helpers.ReadLineAsync(reader, cancellationToken).ConfigureAwait(false) : reader.ReadLine();
+            }
+
+            return (vocab, vocabReverse);
+        }
+
+        /// <summary>
+        /// Create a new instance of the <see cref="WordPieceTokenizer"/> class.
+        /// </summary>
+        /// <param name="vocabFilePath">The path to the WordPiece vocab file.</param>
+        /// <param name="preTokenizer">The PreTokenizer to use.</param>
+        /// <param name="normalizer">The Normalizer to use.</param>
+        /// <param name="specialTokens">The dictionary containing the special tokens and their corresponding ids.</param>
+        /// <param name="unknownToken">The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this token instead.</param>
+        /// <param name="continuingSubwordPrefix">The prefix to use for sub-words that are not the first part of a word.</param>
+        /// <param name="maxInputCharsPerWord">The maximum number of characters to authorize in a single word.</param>
+        /// <returns>A new instance of the <see cref="WordPieceTokenizer"/> class.</returns>
+        /// <remarks>
+        /// If the <paramref name="preTokenizer"/> is null, the whitespace pre-tokenizer will be used.
+        /// </remarks>
+        public static WordPieceTokenizer Create(
+                        string vocabFilePath,
+                        PreTokenizer? preTokenizer = null,
+                        Normalizer? normalizer = null,
+                        IReadOnlyDictionary<string, int>? specialTokens = null,
+                        string unknownToken = "[UNK]",
+                        string continuingSubwordPrefix = DefaultContinuingSubwordPrefix,
+                        int maxInputCharsPerWord = DefaultMaxInputCharsPerWord) =>
+            Create(string.IsNullOrEmpty(vocabFilePath) ? throw new ArgumentNullException(nameof(vocabFilePath)) : File.OpenRead(vocabFilePath), preTokenizer, normalizer, specialTokens, unknownToken, continuingSubwordPrefix, maxInputCharsPerWord, disposeStream: true);
+
+        /// <summary>
+        /// Create a new instance of the <see cref="WordPieceTokenizer"/> class.
+        /// </summary>
+        /// <param name="vocabStream">The path to the WordPiece vocab file.</param>
+        /// <param name="preTokenizer">The PreTokenizer to use.</param>
+        /// <param name="normalizer">The Normalizer to use.</param>
+        /// <param name="specialTokens">The dictionary containing the special tokens and their corresponding ids.</param>
+        /// <param name="unknownToken">The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this token instead.</param>
+        /// <param name="continuingSubwordPrefix">The prefix to use for sub-words that are not the first part of a word.</param>
+        /// <param name="maxInputCharsPerWord">The maximum number of characters to authorize in a single word.</param>
+        /// <returns>A new instance of the <see cref="WordPieceTokenizer"/> class.</returns>
+        /// <remarks>
+        /// If the <paramref name="preTokenizer"/> is null, the whitespace pre-tokenizer will be used.
+        /// </remarks>
+        public static WordPieceTokenizer Create(
+                            Stream vocabStream,
+                            PreTokenizer? preTokenizer = null,
+                            Normalizer? normalizer = null,
+                            IReadOnlyDictionary<string, int>? specialTokens = null,
+                            string unknownToken = "[UNK]",
+                            string continuingSubwordPrefix = DefaultContinuingSubwordPrefix,
+                            int maxInputCharsPerWord = DefaultMaxInputCharsPerWord) => Create(vocabStream, preTokenizer, normalizer, specialTokens, unknownToken, continuingSubwordPrefix, maxInputCharsPerWord, disposeStream: false);
+
+        private static WordPieceTokenizer Create(
+                            Stream vocabStream,
+                            PreTokenizer? preTokenizer,
+                            Normalizer? normalizer,
+                            IReadOnlyDictionary<string, int>? specialTokens,
+                            string unknownToken,
+                            string continuingSubwordPrefix,
+                            int maxInputCharsPerWord,
+                            bool disposeStream)
+        {
+            if (vocabStream is null)
+            {
+                throw new ArgumentNullException(nameof(vocabStream));
+            }
+
+            try
+            {
+                (Dictionary<StringSpanOrdinalKey, int> vocab, Dictionary<int, string> vocabReverse) = LoadVocabAsync(vocabStream, useAsync: false).GetAwaiter().GetResult();
+
+                return new WordPieceTokenizer(vocab, vocabReverse, preTokenizer, normalizer, specialTokens, unknownToken, continuingSubwordPrefix, maxInputCharsPerWord);
+            }
+            finally
+            {
+                if (disposeStream)
+                {
+                    vocabStream.Dispose();
+                }
+            }
+        }
+
+        /// <summary>
+        /// Create a new instance of the <see cref="WordPieceTokenizer"/> class asynchronously.
+        /// </summary>
+        /// <param name="vocabFilePath">The path to the WordPiece vocab file.</param>
+        /// <param name="preTokenizer">The PreTokenizer to use.</param>
+        /// <param name="normalizer">The Normalizer to use.</param>
+        /// <param name="specialTokens">The dictionary containing the special tokens and their corresponding ids.</param>
+        /// <param name="unknownToken">The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this token instead.</param>
+        /// <param name="continuingSubwordPrefix">The prefix to use for sub-words that are not the first part of a word.</param>
+        /// <param name="maxInputCharsPerWord">The maximum number of characters to authorize in a single word.</param>
+        /// <param name="cancellationToken">The cancellation token.</param>
+        /// <returns>A new instance of the <see cref="WordPieceTokenizer"/> class.</returns>
+        /// <remarks>
+        /// If the <paramref name="preTokenizer"/> is null, the whitespace pre-tokenizer will be used.
+        /// </remarks>
+        public static async Task<WordPieceTokenizer> CreateAsync(
+                                string vocabFilePath,
+                                PreTokenizer? preTokenizer = null,
+                                Normalizer? normalizer = null,
+                                IReadOnlyDictionary<string, int>? specialTokens = null,
+                                string unknownToken = "[UNK]",
+                                string continuingSubwordPrefix = DefaultContinuingSubwordPrefix,
+                                int maxInputCharsPerWord = DefaultMaxInputCharsPerWord,
+                                CancellationToken cancellationToken = default) =>
+            await CreateAsync(
+                    string.IsNullOrEmpty(vocabFilePath) ? throw new ArgumentNullException(nameof(vocabFilePath)) : File.OpenRead(vocabFilePath),
+                    preTokenizer,
+                    normalizer,
+                    specialTokens,
+                    unknownToken,
+                    continuingSubwordPrefix,
+                    maxInputCharsPerWord,
+                    cancellationToken,
+                    disposeStream: true);
+
+        /// <summary>
+        /// Create a new instance of the <see cref="WordPieceTokenizer"/> class asynchronously.
+        /// </summary>
+        /// <param name="vocabStream">The path to the WordPiece vocab file.</param>
+        /// <param name="preTokenizer">The PreTokenizer to use.</param>
+        /// <param name="normalizer">The Normalizer to use.</param>
+        /// <param name="specialTokens">The dictionary containing the special tokens and their corresponding ids.</param>
+        /// <param name="unknownToken">The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this token instead.</param>
+        /// <param name="continuingSubwordPrefix">The prefix to use for sub-words that are not the first part of a word.</param>
+        /// <param name="maxInputCharsPerWord">The maximum number of characters to authorize in a single word.</param>
+        /// <param name="cancellationToken">The cancellation token.</param>
+        /// <returns>A new instance of the <see cref="WordPieceTokenizer"/> class.</returns>
+        /// <remarks>
+        /// If the <paramref name="preTokenizer"/> is null, the whitespace pre-tokenizer will be used.
+        /// </remarks>
+        public static async Task<WordPieceTokenizer> CreateAsync(
+                                Stream vocabStream,
+                                PreTokenizer? preTokenizer = null,
+                                Normalizer? normalizer = null,
+                                IReadOnlyDictionary<string, int>? specialTokens = null,
+                                string unknownToken = "[UNK]",
+                                string continuingSubwordPrefix = DefaultContinuingSubwordPrefix,
+                                int maxInputCharsPerWord = DefaultMaxInputCharsPerWord,
+                                CancellationToken cancellationToken = default) =>
+            await CreateAsync(vocabStream, preTokenizer, normalizer, specialTokens, unknownToken, continuingSubwordPrefix, maxInputCharsPerWord, cancellationToken, disposeStream: false);
+
+        private static async Task<WordPieceTokenizer> CreateAsync(
+                                Stream vocabStream,
+                                PreTokenizer? preTokenizer,
+                                Normalizer? normalizer,
+                                IReadOnlyDictionary<string, int>? specialTokens,
+                                string unknownToken,
+                                string continuingSubwordPrefix,
+                                int maxInputCharsPerWord,
+                                CancellationToken cancellationToken,
+                                bool disposeStream)
+        {
+            if (vocabStream is null)
+            {
+                throw new ArgumentNullException(nameof(vocabStream));
+            }
+
+            try
+            {
+                (Dictionary<StringSpanOrdinalKey, int> vocab, Dictionary<int, string> vocabReverse) = await LoadVocabAsync(vocabStream, useAsync: true, cancellationToken);
+
+                return new WordPieceTokenizer(vocab, vocabReverse, preTokenizer, normalizer, specialTokens, unknownToken, continuingSubwordPrefix, maxInputCharsPerWord);
+            }
+            finally
+            {
+                if (disposeStream)
+                {
+                    vocabStream.Dispose();
+                }
+            }
+        }
+
+        /// <summary>
+        /// Gets the PreTokenizer used by the Tokenizer.
+        /// </summary>
+        public override PreTokenizer? PreTokenizer => _preTokenizer;
+
+        /// <summary>
+        /// Gets the Normalizer in use by the Tokenizer.
+        /// </summary>
+        public override Normalizer? Normalizer => _normalizer;
+
+        /// <summary>
+        /// Gets the unknown token.
+        /// A token that is not in the vocabulary cannot be converted to an ID and is set to be this token instead.
+        /// </summary>
+        public string UnknownToken { get; }
+
+        /// <summary>
+        /// Gets the special tokens and their corresponding ids.
+        /// </summary>
+        public IReadOnlyDictionary<string, int>? SpecialTokens { get; }
+
+        /// <summary>
+        /// Gets the Ids to tokens mapping for special tokens.
+        /// </summary>
+        internal IReadOnlyDictionary<int, string>? SpecialTokensReverse { get; }
+
+        /// <summary>
+        /// Encodes input text to a list of <see cref="EncodedToken" />s.
+        /// </summary>
+        /// <param name="text">The text to encode.</param>
+        /// <param name="textSpan">The span of the text to encode which will be used if the <paramref name="text"/> is <see langword="null"/>.</param>
+        /// <param name="settings">The settings used to encode the text.</param>
+        protected override EncodeResults<EncodedToken> EncodeToTokens(string? text, ReadOnlySpan<char> textSpan, EncodeSettings settings)
+        {
+            if (string.IsNullOrEmpty(text) && textSpan.IsEmpty)
+            {
+                return new EncodeResults<EncodedToken> { NormalizedText = null, Tokens = [], CharsConsumed = 0 };
+            }
+
+            IEnumerable<(int Offset, int Length)>? splits = InitializeForEncoding(
+                                                                text,
+                                                                textSpan,
+                                                                settings.ConsiderPreTokenization,
+                                                                settings.ConsiderNormalization,
+                                                                _normalizer,
+                                                                _preTokenizer,
+                                                                out string? normalizedString,
+                                                                out ReadOnlySpan<char> textSpanToEncode,
+                                                                out int charsConsumed);
+
+            List<EncodedToken> tokens = new();
+
+            if (splits is not null)
+            {
+                foreach ((int Offset, int Length) split in splits)
+                {
+                    EncodeToTokens(textSpanToEncode.Slice(split.Offset, split.Length), tokens, split.Offset);
+                }
+            }
+            else
+            {
+                EncodeToTokens(textSpanToEncode, tokens, 0);
+            }
+
+            return new EncodeResults<EncodedToken> { NormalizedText = normalizedString, Tokens = tokens, CharsConsumed = charsConsumed };
+        }
+
+        /// <summary>
+        /// Encode text to a list of tokens.
+        /// </summary>
+        /// <param name="text">The text to encode.</param>
+        /// <param name="tokens">The list of tokens to populate.</param>
+        /// <param name="offset">The offset to start encoding from.</param>
+        private void EncodeToTokens(ReadOnlySpan<char> text, List<EncodedToken> tokens, int offset)
+        {
+            Debug.Assert(!text.IsEmpty);
+
+            if (text.Length > MaxInputCharsPerWord)
+            {
+                tokens.Add(new EncodedToken(UnknownTokenId, UnknownToken, new Range(offset, offset + text.Length)));
+                return;
+            }
+
+            int maxLength = MaxInputCharsPerWord + ContinuingSubwordPrefix.Length;
+            char[]? arrayPool = maxLength <= 250 ? null : ArrayPool<char>.Shared.Rent(maxLength);
+            Span<char> buffer = arrayPool is null ? stackalloc char[maxLength] : arrayPool;
+            ContinuingSubwordPrefix.AsSpan().CopyTo(buffer);
+
+            int initialTokensCount = tokens.Count;
+            int textLength = text.Length;
+            bool isBad = false;
+
+            int start = 0;
+
+            while (start < textLength)
+            {
+                int end = textLength;
+                EncodedToken curToken = default;
+
+                while (start < end)
+                {
+                    scoped ReadOnlySpan<char> subStr = text.Slice(start, end - start);
+
+                    if (start > 0)
+                    {
+                        subStr.CopyTo(buffer.Slice(ContinuingSubwordPrefix.Length));
+                        subStr = buffer.Slice(0, ContinuingSubwordPrefix.Length + subStr.Length);
+                    }
+
+                    if (_vocab.TryGetValue(subStr, out int id))
+                    {
+                        Debug.Assert(_vocabReverse.ContainsKey(id));
+                        curToken = new EncodedToken(id, _vocabReverse[id], new Range(offset + start, offset + end));
+                        break;
+                    }
+
+                    end -= 1;
+                }
+
+                if (curToken.Value is null)
+                {
+                    isBad = true;
+                    break;
+                }
+
+                tokens.Add(curToken);
+                start = end;
+            }
+
+            if (isBad)
+            {
+                // remove previously added tokens and add the unknown token
+                tokens.RemoveRange(initialTokensCount, tokens.Count - initialTokensCount);
+                tokens.Add(new EncodedToken(UnknownTokenId, UnknownToken, new Range(offset, offset + textLength)));
+            }
+
+            if (arrayPool is not null)
+            {
+                ArrayPool<char>.Shared.Return(arrayPool);
+            }
+        }
+
+        /// <summary>
+        /// Encodes input text to token Ids.
+        /// </summary>
+        /// <param name="text">The text to encode.</param>
+        /// <param name="textSpan">The span of the text to encode which will be used if the <paramref name="text"/> is <see langword="null"/>.</param>
+        /// <param name="settings">The settings used to encode the text.</param>
+        /// <returns>The encoded results containing the list of encoded Ids.</returns>
+        protected override EncodeResults<int> EncodeToIds(string? text, ReadOnlySpan<char> textSpan, EncodeSettings settings)
+        {
+            int maxTokenCount = settings.MaxTokenCount;
+            if (maxTokenCount <= 0)
+            {
+                throw new ArgumentOutOfRangeException(nameof(settings.MaxTokenCount), "The maximum number of tokens must be greater than zero.");
+            }
+
+            if (string.IsNullOrEmpty(text) && textSpan.IsEmpty)
+            {
+                return new EncodeResults<int> { NormalizedText = null, Tokens = [], CharsConsumed = 0 };
+            }
+
+            IEnumerable<(int Offset, int Length)>? splits = InitializeForEncoding(
+                                                                text,
+                                                                textSpan,
+                                                                settings.ConsiderPreTokenization,
+                                                                settings.ConsiderNormalization,
+                                                                _normalizer,
+                                                                _preTokenizer,
+                                                                out string? normalizedString,
+                                                                out ReadOnlySpan<char> textSpanToEncode,
+                                                                out int charsConsumed);
+
+            List<int> ids = new();
+
+            if (splits is not null)
+            {
+                charsConsumed = 0;
+                foreach ((int Offset, int Length) split in splits)
+                {
+                    EncodeToIds(textSpanToEncode.Slice(split.Offset, split.Length), ids, out int length, maxTokenCount - ids.Count);
+
+                    if (length < split.Length || ids.Count >= maxTokenCount)
+                    {
+                        break;
+                    }
+
+                    charsConsumed = split.Offset + length;
+                }
+            }
+            else
+            {
+                EncodeToIds(textSpanToEncode, ids, out charsConsumed);
+            }
+
+            return new EncodeResults<int> { NormalizedText = normalizedString, Tokens = ids, CharsConsumed = charsConsumed };
+        }
+
+        /// <summary>
+        /// Encode text to a list of Ids.
+        /// </summary>
+        /// <param name="text">The text to encode.</param>
+        /// <param name="accumulatedIds">The list of accumulated Ids.</param>
+        /// <param name="charsConsumed">The length of the text that encompasses the maximum encoded tokens.</param>
+        /// <param name="maxTokenCount">The maximum number of tokens to encode.</param>
+        /// <returns>The number of tokens that the input text will be encoded to.</returns>
+        private int EncodeToIds(ReadOnlySpan<char> text, List<int>? accumulatedIds, out int charsConsumed, int maxTokenCount = int.MaxValue)
+        {
+            Debug.Assert(maxTokenCount > 0);
+
+            if (text.IsEmpty)
+            {
+                charsConsumed = 0;
+                return 0;
+            }
+
+            if (text.Length > MaxInputCharsPerWord)
+            {
+                accumulatedIds?.Add(UnknownTokenId);
+                charsConsumed = text.Length;
+                return 1;
+            }
+
+            int maxLength = MaxInputCharsPerWord + ContinuingSubwordPrefix.Length;
+            char[]? arrayPool = maxLength <= 250 ? null : ArrayPool<char>.Shared.Rent(maxLength);
+            Span<char> buffer = arrayPool is null ? stackalloc char[maxLength] : arrayPool;
+            ContinuingSubwordPrefix.AsSpan().CopyTo(buffer);
+
+            int addedIds = 0;
+            int textLength = text.Length;
+            bool isBad = false;
+
+            int start = 0;
+
+            while (start < textLength)
+            {
+                int end = textLength;
+                int curId = 0;
+                bool found = false;
+
+                while (start < end)
+                {
+                    scoped ReadOnlySpan<char> subStr = text.Slice(start, end - start);
+
+                    if (start > 0)
+                    {
+                        subStr.CopyTo(buffer.Slice(ContinuingSubwordPrefix.Length));
+                        subStr = buffer.Slice(0, ContinuingSubwordPrefix.Length + subStr.Length);
+                    }
+
+                    if (_vocab.TryGetValue(subStr, out curId))
+                    {
+                        found = true;
+                        break;
+                    }
+
+                    end -= 1;
+                }
+
+                if (!found)
+                {
+                    isBad = true;
+                    break;
+                }
+
+                accumulatedIds?.Add(curId);
+                addedIds++;
+                start = end;
+            }
+
+            charsConsumed = textLength;
+            if (addedIds > maxTokenCount)
+            {
+                // not enough space to hold added ids. Remove previously added ids
+                accumulatedIds?.RemoveRange(accumulatedIds.Count - addedIds, addedIds);
+                addedIds = 0;
+                charsConsumed = 0;
+            }
+            else if (isBad)
+            {
+                // remove previously added ids and add the unknown token id
+                accumulatedIds?.RemoveRange(accumulatedIds.Count - addedIds, addedIds);
+                accumulatedIds?.Add(UnknownTokenId);
+                addedIds = 1;
+            }
+
+            if (arrayPool is not null)
+            {
+                ArrayPool<char>.Shared.Return(arrayPool);
+            }
+
+            return addedIds;
+        }
+
+        /// <summary>
+        /// Get the number of tokens that the input text will be encoded to.
+        /// </summary>
+        /// <param name="text">The text to encode.</param>
+        /// <param name="textSpan">The span of the text to encode which will be used if the <paramref name="text"/> is <see langword="null"/>.</param>
+        /// <param name="settings">The settings used to encode the text.</param>
+        /// <returns>The number of token Ids that the input text will be encoded to.</returns>
+        protected override int CountTokens(string? text, ReadOnlySpan<char> textSpan, EncodeSettings settings)
+        {
+            int maxTokenCount = settings.MaxTokenCount;
+            if (maxTokenCount <= 0)
+            {
+                throw new ArgumentOutOfRangeException(nameof(settings.MaxTokenCount), "The maximum number of tokens must be greater than zero.");
+            }
+
+            if (string.IsNullOrEmpty(text) && textSpan.IsEmpty)
+            {
+                return 0;
+            }
+
+            IEnumerable<(int Offset, int Length)>? splits = InitializeForEncoding(
+                                                                text,
+                                                                textSpan,
+                                                                settings.ConsiderPreTokenization,
+                                                                settings.ConsiderNormalization,
+                                                                _normalizer,
+                                                                _preTokenizer,
+                                                                out string? normalizedString,
+                                                                out ReadOnlySpan<char> textSpanToEncode,
+                                                                out int charsConsumed);
+
+            int count = 0;
+            if (splits is not null)
+            {
+                foreach ((int Offset, int Length) split in splits)
+                {
+                    count += EncodeToIds(textSpanToEncode.Slice(split.Offset, split.Length), accumulatedIds: null, out int length, maxTokenCount - count);
+
+                    if (length < split.Length || count >= maxTokenCount)
+                    {
+                        break;
+                    }
+                }
+            }
+            else
+            {
+                count = EncodeToIds(textSpanToEncode, accumulatedIds: null, out charsConsumed, maxTokenCount);
+            }
+
+            return count;
+        }
+
+        /// <summary>
+        /// Find the index of the maximum encoding capacity without surpassing the token limit.
+        /// </summary>
+        /// <param name="text">The text to encode.</param>
+        /// <param name="textSpan">The span of the text to encode which will be used if the <paramref name="text"/> is <see langword="null"/>.</param>
+        /// <param name="settings">The settings used to encode the text.</param>
+        /// <param name="fromEnd">Indicate whether to find the index from the end of the text.</param>
+        /// <param name="normalizedString">If the tokenizer's normalization is enabled or <paramRef name="settings" /> has <see cref="EncodeSettings.ConsiderNormalization"/> is <see langword="false"/>, this will be set to <paramRef name="text" /> in its normalized form; otherwise, this value will be set to <see langword="null"/>.</param>
+        /// <param name="tokenCount">The token count can be generated which should be smaller than the maximum token count.</param>
+        /// <returns>
+        /// The index of the maximum encoding capacity within the processed text without surpassing the token limit.
+        /// If <paramRef name="fromEnd" /> is <see langword="false"/>, it represents the index immediately following the last character to be included. In cases where no tokens fit, the result will be 0; conversely,
+        /// if all tokens fit, the result will be length of the input text or the <paramref name="normalizedString"/> if the normalization is enabled.
+        /// If <paramRef name="fromEnd" /> is <see langword="true"/>, it represents the index of the first character to be included. In cases where no tokens fit, the result will be the text length; conversely,
+        /// if all tokens fit, the result will be zero.
+        /// </returns>
+        protected override int GetIndexByTokenCount(string? text, ReadOnlySpan<char> textSpan, EncodeSettings settings, bool fromEnd, out string? normalizedString, out int tokenCount)
+        {
+            if (settings.MaxTokenCount <= 0)
+            {
+                throw new ArgumentOutOfRangeException(nameof(settings.MaxTokenCount), "The max token count must be greater than 0.");
+            }
+
+            if (string.IsNullOrEmpty(text) && textSpan.IsEmpty)
+            {
+                normalizedString = null;
+                tokenCount = 0;
+                return 0;
+            }
+
+            IEnumerable<(int Offset, int Length)>? splits = InitializeForEncoding(
+                                                                text,
+                                                                textSpan,
+                                                                settings.ConsiderNormalization,
+                                                                settings.ConsiderNormalization,
+                                                                _normalizer,
+                                                                _preTokenizer,
+                                                                out normalizedString,
+                                                                out ReadOnlySpan<char> textSpanToEncode,
+                                                                out _);
+
+            int charsConsumed;
+
+            if (splits is null)
+            {
+                tokenCount = EncodeToIds(textSpanToEncode, accumulatedIds: null, out charsConsumed, settings.MaxTokenCount);
+                if (charsConsumed != textSpanToEncode.Length)
+                {
+                    tokenCount = 0;
+                    return fromEnd ? textSpanToEncode.Length : 0;
+                }
+
+                return fromEnd ? 0 : textSpanToEncode.Length;
+            }
+
+            if (fromEnd)
+            {
+                splits = splits.Reverse();
+            }
+
+            tokenCount = 0;
+            foreach ((int Offset, int Length) split in splits)
+            {
+                int count = EncodeToIds(textSpanToEncode.Slice(split.Offset, split.Length), accumulatedIds: null, out charsConsumed, settings.MaxTokenCount - tokenCount);
+                if (charsConsumed != split.Length)
+                {
+                    return fromEnd ? split.Offset + split.Length : split.Offset;
+                }
+
+                tokenCount += count;
+
+                if (count >= settings.MaxTokenCount)
+                {
+                    return fromEnd ? split.Offset : split.Offset + split.Length;
+                }
+            }
+
+            return fromEnd ? 0 : textSpanToEncode.Length;
+        }
+
+        /// <summary>
+        /// Decode the given ids, back to a String.
+        /// </summary>
+        /// <param name="ids">The list of ids that we want to decode.</param>
+        /// <returns>The decoded string.</returns>
+        public override string Decode(IEnumerable<int> ids) => Decode(ids, skipSpecialTokens: false);
+
+        /// <summary>
+        /// Decode the given ids, back to a String.
+        /// </summary>
+        /// <param name="ids">The list of ids that we want to decode.</param>
+        /// <param name="skipSpecialTokens">Indicate whether to skip the special tokens during the decoding.</param>
+        /// <returns>The decoded string.</returns>
+        public string Decode(IEnumerable<int> ids, bool skipSpecialTokens)
+        {
+            ValueStringBuilder sb = new ValueStringBuilder();
+            bool first = true;
+            bool ignoreSpecialTokens = skipSpecialTokens && SpecialTokensReverse is not null;
+
+            foreach (int id in ids)
+            {
+                if (ignoreSpecialTokens && SpecialTokensReverse!.TryGetValue(id, out _))
+                {
+                    continue;
+                }
+
+                if (_vocabReverse.TryGetValue(id, out string? token))
+                {
+                    if (token.StartsWith(ContinuingSubwordPrefix))
+                    {
+                        sb.Append(token.AsSpan().Slice(ContinuingSubwordPrefix.Length));
+                    }
+                    else
+                    {
+                        if (!first && token[0] is not ('.' or ',' or '!' or '?' or '\''))
+                        {
+                            sb.Append(' ');
+                        }
+
+                        sb.Append(token);
+                    }
+                }
+
+                first = false;
+            }
+
+            return sb.ToString();
+        }
+
+        /// <summary>
+        /// Decode the given ids back to text and store the result in the <paramref name="destination"/> span.
+        /// </summary>
+        /// <param name="ids">The list of ids that we want to decode.</param>
+        /// <param name="destination">The span to store the decoded text.</param>
+        /// <param name="idsConsumed">The number of ids consumed during the decoding.</param>
+        /// <param name="charsWritten">The number of characters written to the destination span.</param>
+        /// <returns>The operation status indicates whether all IDs were successfully decoded or if the <paramref name="destination"/> is too small to contain the entire decoded result.</returns>
+        public override OperationStatus Decode(IEnumerable<int> ids, Span<char> destination, out int idsConsumed, out int charsWritten) =>
+            Decode(ids, destination, skipSpecialTokens: false, out idsConsumed, out charsWritten);
+
+        /// <summary>
+        /// Decode the given ids back to text and store the result in the <paramref name="destination"/> span.
+        /// </summary>
+        /// <param name="ids">The list of ids that we want to decode.</param>
+        /// <param name="destination">The span to store the decoded text.</param>
+        /// <param name="skipSpecialTokens">Indicate whether to skip the special tokens during the decoding.</param>
+        /// <param name="idsConsumed">The number of ids consumed during the decoding.</param>
+        /// <param name="charsWritten">The number of characters written to the destination span.</param>
+        /// <returns>The operation status indicates whether all IDs were successfully decoded or if the <paramref name="destination"/> is too small to contain the entire decoded result.</returns>
+        public OperationStatus Decode(IEnumerable<int> ids, Span<char> destination, bool skipSpecialTokens, out int idsConsumed, out int charsWritten)
+        {
+            charsWritten = 0;
+            idsConsumed = 0;
+            Span<char> buffer = destination;
+
+            bool first = true;
+            bool ignoreSpecialTokens = SpecialTokensReverse is not null && skipSpecialTokens;
+
+            foreach (int id in ids)
+            {
+                if (ignoreSpecialTokens && SpecialTokensReverse!.TryGetValue(id, out _))
+                {
+                    continue;
+                }
+
+                if (_vocabReverse.TryGetValue(id, out string? token))
+                {
+                    if (token.StartsWith(ContinuingSubwordPrefix, StringComparison.Ordinal))
+                    {
+                        if (token.Length - ContinuingSubwordPrefix.Length > buffer.Length)
+                        {
+                            return OperationStatus.DestinationTooSmall;
+                        }
+                        token.AsSpan().Slice(ContinuingSubwordPrefix.Length).CopyTo(buffer);
+                        buffer = buffer.Slice(token.Length - ContinuingSubwordPrefix.Length);
+                        charsWritten += token.Length - ContinuingSubwordPrefix.Length;
+                    }
+                    else
+                    {
+                        if (!first)
+                        {
+                            if (token.Length + 1 > buffer.Length)
+                            {
+                                return OperationStatus.DestinationTooSmall;
+                            }
+
+                            buffer[0] = ' ';
+                            token.AsSpan().CopyTo(buffer.Slice(1));
+                            buffer = buffer.Slice(token.Length + 1);
+                            charsWritten += token.Length + 1;
+                        }
+                        else
+                        {
+                            if (token.Length > buffer.Length)
+                            {
+                                return OperationStatus.DestinationTooSmall;
+                            }
+
+                            token.AsSpan().CopyTo(buffer);
+                            buffer = buffer.Slice(token.Length);
+                            charsWritten += token.Length;
+                        }
+                    }
+
+                    first = false;
+
+                    idsConsumed++;
+                }
+                else
+                {
+                    return OperationStatus.InvalidData;
+                }
+            }
+
+            return OperationStatus.Done;
+        }
+    }
+}

src/Microsoft.ML.Tokenizers/Normalizer/BertNormalizer.cs

@@ -0,0 +1,200 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+using System;
+using System.Buffers;
+using System.Diagnostics;
+using System.Globalization;
+using System.Runtime.CompilerServices;
+using System.Runtime.InteropServices;
+using System.Text;
+
+namespace Microsoft.ML.Tokenizers
+{
+    /// <summary>
+    /// Normalizer that performs the Bert model normalization.
+    /// </summary>
+    internal sealed class BertNormalizer : Normalizer
+    {
+        private readonly bool _doLowerCase;
+        private readonly bool _tokenizeChineseChars;
+        private readonly bool _stripAccents;
+
+        /// <summary>
+        /// Normalize the input string.
+        /// </summary>
+        /// <param name="original">The input string to normalize.</param>
+        /// <returns>The normalized string.</returns>
+        public override string Normalize(string original)
+        {
+            if (string.IsNullOrEmpty(original))
+            {
+                return string.Empty;
+            }
+
+            if (_stripAccents)
+            {
+                original = original.Normalize(NormalizationForm.FormD);
+            }
+
+            Span<char> casingBuffer = stackalloc char[10];
+            char[] buffer = ArrayPool<char>.Shared.Rent(original.Length);
+            int index = 0;
+
+            for (int i = 0; i < original.Length; i++)
+            {
+                char c = original[i];
+
+                if (c == '\u0000' || c == '\uFFFD')
+                {
+                    continue;
+                }
+
+                int inc = 0;
+                int codePoint = (int)c;
+                if (char.IsHighSurrogate(c) && i + 1 < original.Length && char.IsLowSurrogate(original[i + 1]))
+                {
+                    codePoint = char.ConvertToUtf32(c, original[i + 1]);
+                    inc = 1;
+                }
+
+                UnicodeCategory category = CharUnicodeInfo.GetUnicodeCategory(original, i);
+
+                if (category == UnicodeCategory.Control)
+                {
+                    i += inc;
+                    continue;
+                }
+
+                if (category == UnicodeCategory.SpaceSeparator)
+                {
+                    InsertChar(ref buffer, ref index, ' ');
+                    i += inc;
+                    continue;
+                }
+
+                if (_stripAccents && category is UnicodeCategory.NonSpacingMark or UnicodeCategory.SpacingCombiningMark)
+                {
+                    i += inc;
+                    continue;
+                }
+
+                if (_doLowerCase && category == UnicodeCategory.UppercaseLetter)
+                {
+                    int length = original.AsSpan().Slice(i, inc + 1).ToLowerInvariant(casingBuffer);
+                    Debug.Assert(length > 0);
+
+                    InsertSpan(ref buffer, ref index, casingBuffer.Slice(0, length));
+
+                    i += inc;
+                    continue;
+                }
+
+                if (_tokenizeChineseChars && IsChineseChar(codePoint))
+                {
+                    InsertChar(ref buffer, ref index, ' ');
+                    InsertChar(ref buffer, ref index, c);
+                    if (inc > 0)
+                    {
+                        InsertChar(ref buffer, ref index, original[i + 1]);
+                    }
+                    InsertChar(ref buffer, ref index, ' ');
+
+                    i += inc;
+                    continue;
+                }
+
+                InsertChar(ref buffer, ref index, c);
+                if (inc > 0)
+                {
+                    InsertChar(ref buffer, ref index, original[i + 1]);
+                }
+                i += inc;
+            }
+
+            string result = index == 0 ? string.Empty : new string(buffer, 0, index).Normalize(NormalizationForm.FormC);
+            ArrayPool<char>.Shared.Return(buffer);
+            return result;
+        }
+
+        /// <summary>
+        /// Normalize the input character span.
+        /// </summary>
+        /// <param name="original">The input character span to normalize.</param>
+        /// <returns>The normalized string.</returns>
+        public override string Normalize(ReadOnlySpan<char> original)
+        {
+            if (original.IsEmpty)
+            {
+                return string.Empty;
+            }
+
+            return Normalize(original.ToString());
+        }
+
+        /// <summary>
+        /// Initializes a new instance of the <see cref="BertNormalizer"/> class.
+        /// </summary>
+        /// <param name="doLowerCase">Whether to lowercase the input.</param>
+        /// <param name="tokenizeChineseChars">Whether to tokenize Chinese characters.</param>
+        /// <param name="stripAccents">Whether to strip accents from the input.</param>
+        public BertNormalizer(bool doLowerCase, bool tokenizeChineseChars, bool stripAccents)
+        {
+            _doLowerCase = doLowerCase;
+            _tokenizeChineseChars = tokenizeChineseChars;
+            _stripAccents = stripAccents;
+        }
+
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        private static void InsertChar(ref char[] buffer, ref int index, char c)
+        {
+            if (index >= buffer.Length)
+            {
+                Helpers.ArrayPoolGrow(ref buffer, index + 40);
+            }
+
+            buffer[index++] = c;
+        }
+
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        private static void InsertSpan(ref char[] buffer, ref int index, Span<char> chars)
+        {
+            if (index + buffer.Length >= buffer.Length)
+            {
+                Helpers.ArrayPoolGrow(ref buffer, index + buffer.Length + 10);
+            }
+
+            chars.CopyTo(buffer.AsSpan(index));
+            index += chars.Length;
+        }
+
+        /// <summary>
+        /// Checks whether CP is the codepoint of a CJK character.
+        /// This defines a "chinese character" as anything in the CJK Unicode block:
+        ///   https://en.wikipedia.org/wiki/CJK_Unified_Ideographs_(Unicode_block)
+        /// </summary>
+        /// <param name="codePoint">The codepoint to check.</param>
+        /// <remarks>
+        /// The CJK Unicode block is NOT all Japanese and Korean characters,
+        /// despite its name. The modern Korean Hangul alphabet is a different block,
+        /// as is Japanese Hiragana and Katakana. Those alphabets are used to write
+        /// space-separated words, so they are not treated specially and handled
+        /// like the all of the other languages.
+        /// </remarks>
+        /// <returns>True if the codepoint is a CJK character, false otherwise.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        private static bool IsChineseChar(int codePoint)
+        {
+            return (codePoint > 0x3400) && // Quick check to exit early if the codepoint is outside of the CJK range
+               (((uint)(codePoint - 0x3400) <= (uint)(0x4DBF - 0x3400)) ||
+                ((uint)(codePoint - 0xF900) <= (uint)(0xFAFF - 0xF900)) ||
+                ((uint)(codePoint - 0x4E00) <= (uint)(0x9FFF - 0x4E00)) ||
+                ((uint)(codePoint - 0x20000) <= (uint)(0x2A6DF - 0x20000)) ||
+                ((uint)(codePoint - 0x2A700) <= (uint)(0x2B73F - 0x2A700)) ||
+                ((uint)(codePoint - 0x2B740) <= (uint)(0x2B81F - 0x2B740)) ||
+                ((uint)(codePoint - 0x2B820) <= (uint)(0x2CEAF - 0x2B820)) ||
+                ((uint)(codePoint - 0x2F800) <= (uint)(0x2FA1F - 0x2F800)));
+        }
+    }
+}

src/Microsoft.ML.Tokenizers/PreTokenizer/PreTokenizer.cs

@@ -40,8 +40,61 @@ namespace Microsoft.ML.Tokenizers
             }
         }
 
-        private const string WhiteSpacePattern = /*lang=regex*/ @"\w+|[^\w\s]+";
+        private const string WhiteSpaceOrPunctuationPattern = @"\w+|[\p{P}]";
+        private static PreTokenizer? _whiteSpaceOrPunctuationPreTokenizer;
+#if NET7_0_OR_GREATER
+        [GeneratedRegex(WhiteSpaceOrPunctuationPattern)]
+        private static partial Regex WhiteSpaceOrPunctuationRegex();
+#else
+        private static Regex WhiteSpaceOrPunctuationRegex() => new Regex(WhiteSpaceOrPunctuationPattern, RegexOptions.Compiled);
+#endif
+
+        /// <summary>
+        /// Create a new instance of the <see cref="PreTokenizer"/> class which split the text at the whitespace or punctuation characters.
+        /// </summary>
+        /// <param name="specialTokensEncoder">The dictionary containing the special tokens and their corresponding ids.</param>
+        /// <returns>The pre-tokenizer that splits the text at the whitespace or punctuation characters.</returns>
+        public static PreTokenizer CreateWhiteSpaceOrPunctuationPreTokenizer(IReadOnlyDictionary<string, int>? specialTokensEncoder = null)
+        {
+            if (specialTokensEncoder is null)
+            {
+                // return a singleton instance of the WhiteSpace pre-tokenizer
+                return _whiteSpaceOrPunctuationPreTokenizer ??= new RegexPreTokenizer(WhiteSpaceOrPunctuationRegex(), null);
+            }
+
+            return new RegexPreTokenizer(WhiteSpaceOrPunctuationRegex(), specialTokensEncoder);
+        }
+
+        private const string WordOrNonWordPattern = /*lang=regex*/ @"\w+|[^\w\s]+";
+        private static PreTokenizer? _wordOrNonWordPreTokenizer;
+
+#if NET7_0_OR_GREATER
+        [GeneratedRegex(WordOrNonWordPattern)]
+        private static partial Regex WordOrNonWordRegex();
+#else
+        private static Regex WordOrNonWordRegex() => new Regex(WordOrNonWordPattern, RegexOptions.Compiled);
+#endif
+
+        /// <summary>
+        /// Create a new instance of the <see cref="PreTokenizer"/> class which split the text at the word or non-word boundary.
+        /// The word is a set of alphabet, numeric, and underscore characters.
+        /// </summary>
+        /// <param name="specialTokensEncoder">The dictionary containing the special tokens and their corresponding ids.</param>
+        /// <returns>The pre-tokenizer that splits the text at the word boundary.</returns>
+        public static PreTokenizer CreateWordOrNonWordPreTokenizer(IReadOnlyDictionary<string, int>? specialTokensEncoder = null)
+        {
+            if (specialTokensEncoder is null)
+            {
+                // return a singleton instance of the WhiteSpace pre-tokenizer
+                return _wordOrNonWordPreTokenizer ??= new RegexPreTokenizer(WordOrNonWordRegex(), null);
+            }
+
+            return new RegexPreTokenizer(WordOrNonWordRegex(), specialTokensEncoder);
+        }
+
+        private const string WhiteSpacePattern = @"\S+";
         private static PreTokenizer? _whiteSpacePreTokenizer;
+
 #if NET7_0_OR_GREATER
         [GeneratedRegex(WhiteSpacePattern)]
         private static partial Regex WhiteSpaceRegex();
@@ -50,12 +103,11 @@ namespace Microsoft.ML.Tokenizers
 #endif
 
         /// <summary>
-        /// Create a new instance of the <see cref="PreTokenizer"/> class which split the text at the word boundary.
-        /// The word is a set of alphabet, numeric, and underscore characters.
+        /// Create a new instance of the <see cref="PreTokenizer"/> class which split the text at the white spaces.
         /// </summary>
         /// <param name="specialTokensEncoder">The dictionary containing the special tokens and their corresponding ids.</param>
-        /// <returns>The pre-tokenizer that splits the text at the word boundary.</returns>
-        public static PreTokenizer CreateWhiteSpace(IReadOnlyDictionary<string, int>? specialTokensEncoder = null)
+        /// <returns>The pre-tokenizer that splits the text at the white spaces.</returns>
+        public static PreTokenizer CreateWhiteSpacePreTokenizer(IReadOnlyDictionary<string, int>? specialTokensEncoder = null)
         {
             if (specialTokensEncoder is null)
             {

test/Microsoft.ML.Tokenizers.Tests/BertTokenizerTests.cs

@@ -0,0 +1,513 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+using System;
+using System.Buffers;
+using System.Collections.Generic;
+using System.Diagnostics.Tracing;
+using System.IO;
+using System.Threading.Tasks;
+using Xunit;
+
+namespace Microsoft.ML.Tokenizers.Tests
+{
+    public class BertTokenizerTests
+    {
+        [Fact]
+        public void TestWithLowerCasing()
+        {
+            //                     Ids: 0        1        2        3        4       5    6    7      8        9      10      11     12
+            string[] vocabTokens = ["[PAD]", "[UNK]", "[CLS]", "[SEP]", "[MASK]", "!", ",", "?", "hello", "world", "how", "are", "you"];
+
+            string vocabFile = WordPieceTests.CreateVocabFile(vocabTokens);
+
+            try
+            {
+                using Stream vocabStream = File.OpenRead(vocabFile);
+                BertTokenizer[] bertTokenizers = [BertTokenizer.Create(vocabFile), BertTokenizer.Create(vocabStream)];
+
+                foreach (var tokenizer in bertTokenizers)
+                {
+                    Assert.NotNull(tokenizer.PreTokenizer);
+                    Assert.Equal("[UNK]", tokenizer.UnknownToken);
+                    Assert.Equal(1, tokenizer.UnknownTokenId);
+                    Assert.NotNull(tokenizer.Normalizer);
+                    Assert.NotNull(tokenizer.PreTokenizer);
+
+                    string text = "Hello, How are you?";
+                    var tokens = tokenizer.EncodeToTokens(text, out string? normalizedText);
+                    Assert.Equal("hello, how are you?", normalizedText);
+
+                    Assert.Equal(
+                        [
+                            new EncodedToken(8, "hello", new Range(0, 5)),
+                            new EncodedToken(6, ",", new Range(5, 6)),
+                            new EncodedToken(10, "how", new Range(7, 10)),
+                            new EncodedToken(11, "are", new Range(11, 14)),
+                            new EncodedToken(12, "you", new Range(15, 18)),
+                            new EncodedToken(7, "?", new Range(18, 19))
+                        ],
+                        tokens);
+
+                    var ids = tokenizer.EncodeToIds(text);
+                    Assert.Equal([tokenizer.ClsTokenId, 8, 6, 10, 11, 12, 7, tokenizer.SepTokenId], ids);
+
+                    Assert.Equal("[CLS] hello, how are you? [SEP]", tokenizer.Decode(ids));
+                    Assert.Equal("hello, how are you?", tokenizer.Decode(ids, skipSpecialTokens: true));
+
+                    tokens = tokenizer.EncodeToTokens(tokenizer.Decode(ids), out normalizedText);
+                    Assert.Equal("[cls] hello, how are you? [sep]", normalizedText);
+                    Assert.Equal(
+                        [
+                            new EncodedToken(2, "[CLS]", new Range(0, 5)),
+                            new EncodedToken(8, "hello", new Range(6, 11)),
+                            new EncodedToken(6, ",", new Range(11, 12)),
+                            new EncodedToken(10, "how", new Range(13, 16)),
+                            new EncodedToken(11, "are", new Range(17, 20)),
+                            new EncodedToken(12, "you", new Range(21, 24)),
+                            new EncodedToken(7, "?", new Range(24, 25)),
+                            new EncodedToken(3, "[SEP]", new Range(26, 31))
+                        ],
+                        tokens);
+
+                    ids = tokenizer.EncodeToIds(normalizedText!);
+                    Assert.Equal([tokenizer.ClsTokenId, tokenizer.ClsTokenId, 8, 6, 10, 11, 12, 7, tokenizer.SepTokenId, tokenizer.SepTokenId], ids);
+                }
+            }
+            finally
+            {
+                File.Delete(vocabFile);
+            }
+        }
+
+        [Fact]
+        public void TestWithNoLowerCasing()
+        {
+            //                   Ids: 0        1        2        3        4       5    6    7      8        9      10      11     12
+            string[] vocabTokens = ["[PAD]", "[UNK]", "[CLS]", "[SEP]", "[MASK]", "!", ",", "?", "hello", "world", "how", "are", "you"];
+
+            string vocabFile = WordPieceTests.CreateVocabFile(vocabTokens);
+
+            try
+            {
+                using Stream vocabStream = File.OpenRead(vocabFile);
+                BertTokenizer[] bertTokenizers = [BertTokenizer.Create(vocabFile, doLowerCase: false), BertTokenizer.Create(vocabStream, doLowerCase: false)];
+
+                foreach (var tokenizer in bertTokenizers)
+                {
+                    Assert.NotNull(tokenizer.PreTokenizer);
+                    Assert.Equal("[UNK]", tokenizer.UnknownToken);
+                    Assert.Equal(1, tokenizer.UnknownTokenId);
+                    Assert.NotNull(tokenizer.Normalizer);
+                    Assert.NotNull(tokenizer.PreTokenizer);
+
+                    string text = "Hello, How are you?";
+                    var tokens = tokenizer.EncodeToTokens(text, out string? normalizedText);
+                    Assert.Equal("Hello, How are you?", normalizedText);
+
+                    Assert.Equal(
+                        [
+                            new EncodedToken(1, "[UNK]", new Range(0, 5)),
+                            new EncodedToken(6, ",", new Range(5, 6)),
+                            new EncodedToken(1, "[UNK]", new Range(7, 10)),
+                            new EncodedToken(11, "are", new Range(11, 14)),
+                            new EncodedToken(12, "you", new Range(15, 18)),
+                            new EncodedToken(7, "?", new Range(18, 19))
+                        ],
+                        tokens);
+
+                    var ids = tokenizer.EncodeToIds(text);
+                    Assert.Equal([tokenizer.ClsTokenId, 1, 6, 1, 11, 12, 7, tokenizer.SepTokenId], ids);
+
+                    Assert.Equal("[CLS] [UNK], [UNK] are you? [SEP]", tokenizer.Decode(ids));
+                    Assert.Equal(", are you?", tokenizer.Decode(ids, skipSpecialTokens: true));
+                }
+            }
+            finally
+            {
+                File.Delete(vocabFile);
+            }
+        }
+
+        [Fact]
+        public async Task TestWithAccentMarks()
+        {
+            //                  Ids:   0        1        2        3        4       5    6    7      8       9      10      11     12       13       14           15          16         17        18       19
+            string[] vocabTokens = ["[PAD]", "[UNK]", "[CLS]", "[SEP]", "[MASK]", "!", ",", "?", "Café", "cafe", "café", "Über", "über", "uber", "Ångström", "ångström", "angstrom", "Résumé", "résumé", "resume",
+            //                  Ids:  20     21        22         23
+                                    "Cafe", "Uber", "Angstrom", "Resume"];
+            string vocabFile = WordPieceTests.CreateVocabFile(vocabTokens);
+
+            try
+            {
+                using Stream vocabStream = File.OpenRead(vocabFile);
+                BertTokenizer bertTokenizer = await BertTokenizer.CreateAsync(vocabStream); // lowercasing and no accent stripping
+
+                string text = "Café Über Ångström Résumé!";
+                var tokens = bertTokenizer.EncodeToTokens(text, out string? normalizedText);
+                Assert.Equal(
+                    [
+                        new EncodedToken(10, "café", new Range(0, 4)),
+                        new EncodedToken(12, "über", new Range(5, 9)),
+                        new EncodedToken(15, "ångström", new Range(10, 18)),
+                        new EncodedToken(18, "résumé", new Range(19, 25)),
+                        new EncodedToken(5, "!", new Range(25, 26)),
+                    ],
+                    tokens);
+
+                Assert.Equal("café über ångström résumé!", normalizedText);
+
+                vocabStream.Position = 0;
+                bertTokenizer = await BertTokenizer.CreateAsync(vocabStream, doLowerCase: false); // no lowercasing and no accent stripping
+                tokens = bertTokenizer.EncodeToTokens(text, out normalizedText);
+                Assert.Equal(
+                    [
+                        new EncodedToken(8, "Café", new Range(0, 4)),
+                        new EncodedToken(11, "Über", new Range(5, 9)),
+                        new EncodedToken(14, "Ångström", new Range(10, 18)),
+                        new EncodedToken(17, "Résumé", new Range(19, 25)),
+                        new EncodedToken(5, "!", new Range(25, 26)),
+                    ],
+                    tokens);
+
+                Assert.Equal("Café Über Ångström Résumé!", normalizedText);
+
+                vocabStream.Position = 0;
+                bertTokenizer = await BertTokenizer.CreateAsync(vocabStream, stripAccents: true); // lowercasing and accent stripping
+                tokens = bertTokenizer.EncodeToTokens(text, out normalizedText);
+                Assert.Equal("cafe uber angstrom resume!", normalizedText);
+                Assert.Equal(
+                    [
+                        new EncodedToken(9, "cafe", new Range(0, 4)),
+                        new EncodedToken(13, "uber", new Range(5, 9)),
+                        new EncodedToken(16, "angstrom", new Range(10, 18)),
+                        new EncodedToken(19, "resume", new Range(19, 25)),
+                        new EncodedToken(5, "!", new Range(25, 26)),
+                    ],
+                    tokens);
+
+                vocabStream.Position = 0;
+                bertTokenizer = await BertTokenizer.CreateAsync(vocabStream, doLowerCase: false, stripAccents: true); // no lowercasing and accent stripping
+                tokens = bertTokenizer.EncodeToTokens(text, out normalizedText);
+                Assert.Equal("Cafe Uber Angstrom Resume!", normalizedText);
+                Assert.Equal(
+                    [
+                        new EncodedToken(20, "Cafe", new Range(0, 4)),
+                        new EncodedToken(21, "Uber", new Range(5, 9)),
+                        new EncodedToken(22, "Angstrom", new Range(10, 18)),
+                        new EncodedToken(23, "Resume", new Range(19, 25)),
+                        new EncodedToken(5, "!", new Range(25, 26)),
+                    ],
+                    tokens);
+            }
+            finally
+            {
+                File.Delete(vocabFile);
+            }
+        }
+
+        [Fact]
+        public async Task TestChineseCharacters()
+        {
+            //                 Ids:    0        1        2        3        4       5     6       7      8     9    10    11    12
+            string[] vocabTokens = ["[PAD]", "[UNK]", "[CLS]", "[SEP]", "[MASK]", "!", "##驷", "##驸", "受", "叟", "叢", "驷", "驸"];
+            string vocabFile = WordPieceTests.CreateVocabFile(vocabTokens);
+
+            try
+            {
+                using Stream vocabStream = File.OpenRead(vocabFile);
+                BertTokenizer bertTokenizer = await BertTokenizer.CreateAsync(vocabStream); // tokenize Chinese characters
+                string text = "叟驷 叢驸!";
+
+                var tokens = bertTokenizer.EncodeToTokens(text, out string? normalizedText);
+                Assert.Equal(" 叟  驷   叢  驸 !", normalizedText);
+                Assert.Equal(
+                    [
+                        new EncodedToken(9, "叟", new Range(1, 2)),
+                        new EncodedToken(11, "驷", new Range(4, 5)),
+                        new EncodedToken(10, "叢", new Range(8, 9)),
+                        new EncodedToken(12, "驸", new Range(11, 12)),
+                        new EncodedToken(5, "!", new Range(13, 14))
+                    ],
+                    tokens);
+                IReadOnlyList<int> ids = bertTokenizer.EncodeToIds(text);
+                Assert.Equal("[CLS] 叟 驷 叢 驸! [SEP]", bertTokenizer.Decode(bertTokenizer.EncodeToIds(text)));
+                Assert.Equal("叟 驷 叢 驸!", bertTokenizer.Decode(bertTokenizer.EncodeToIds(text), skipSpecialTokens: true));
+
+                vocabStream.Position = 0;
+                bertTokenizer = await BertTokenizer.CreateAsync(vocabStream, tokenizeChineseChars: false); // do not tokenize Chinese characters
+                tokens = bertTokenizer.EncodeToTokens(text, out normalizedText);
+                Assert.Equal("叟驷 叢驸!", normalizedText);
+
+                Assert.Equal(
+                    [
+                        new EncodedToken(9, "叟", new Range(0, 1)),
+                        new EncodedToken(6, "##驷", new Range(1, 2)),
+                        new EncodedToken(10, "叢", new Range(3, 4)),
+                        new EncodedToken(7, "##驸", new Range(4, 5)),
+                        new EncodedToken(5, "!", new Range(5, 6))
+                    ],
+                    tokens);
+                ids = bertTokenizer.EncodeToIds(text);
+                Assert.Equal("[CLS] 叟驷 叢驸! [SEP]", bertTokenizer.Decode(bertTokenizer.EncodeToIds(text)));
+                Assert.Equal("叟驷 叢驸!", bertTokenizer.Decode(bertTokenizer.EncodeToIds(text), skipSpecialTokens: true));
+            }
+            finally
+            {
+                File.Delete(vocabFile);
+            }
+        }
+
+        [Fact]
+        public void TestBuildInputsWithSpecialTokens()
+        {
+            //                   Ids: 0        1        2        3        4        5    6    7      8       9       10     11     12    13    14     15
+            string[] vocabTokens = ["[PAD]", "[UNK]", "[CLS]", "[SEP]", "[MASK]", "!", ",", "?", "hello", "world", "how", "are", "you", "i", "am", "fine"];
+
+            string vocabFile = WordPieceTests.CreateVocabFile(vocabTokens);
+
+            try
+            {
+                using Stream vocabStream = File.OpenRead(vocabFile);
+                BertTokenizer bertTokenizer = BertTokenizer.Create(vocabFile);
+
+                string text1 = "Hello, How are you?";
+                string text2 = "I am fine!";
+
+                var ids1 = bertTokenizer.EncodeToIds(text1);
+                Assert.Equal([bertTokenizer.ClsTokenId, 8, 6, 10, 11, 12, 7, bertTokenizer.SepTokenId], ids1);
+
+                var ids2 = bertTokenizer.EncodeToIds(text2);
+                Assert.Equal([bertTokenizer.ClsTokenId, 13, 14, 15, 5, bertTokenizer.SepTokenId], ids2);
+
+                Assert.Equal(
+                    [bertTokenizer.ClsTokenId, bertTokenizer.ClsTokenId, 8, 6, 10, 11, 12, 7, bertTokenizer.SepTokenId, bertTokenizer.SepTokenId],
+                    bertTokenizer.BuildInputsWithSpecialTokens(ids1));
+
+                Span<int> ids1Span = stackalloc int[1];
+                OperationStatus status = bertTokenizer.BuildInputsWithSpecialTokens(ids1, ids1Span, out int written);
+                Assert.Equal(OperationStatus.DestinationTooSmall, status);
+                Assert.Equal(0, written);
+
+                ids1Span = stackalloc int[ids1.Count + 2];
+                status = bertTokenizer.BuildInputsWithSpecialTokens(ids1, ids1Span, out written);
+                Assert.Equal(OperationStatus.Done, status);
+                Assert.Equal(ids1.Count + 2, written);
+                Assert.Equal(new int[] { bertTokenizer.ClsTokenId, bertTokenizer.ClsTokenId, 8, 6, 10, 11, 12, 7, bertTokenizer.SepTokenId, bertTokenizer.SepTokenId }, ids1Span.ToArray());
+
+                Assert.Equal(
+                    [bertTokenizer.ClsTokenId, bertTokenizer.ClsTokenId, 8, 6, 10, 11, 12, 7, bertTokenizer.SepTokenId, bertTokenizer.SepTokenId, bertTokenizer.ClsTokenId, 13, 14, 15, 5, bertTokenizer.SepTokenId, bertTokenizer.SepTokenId],
+                    bertTokenizer.BuildInputsWithSpecialTokens(ids1, ids2));
+
+                ids1Span = stackalloc int[1];
+                status = bertTokenizer.BuildInputsWithSpecialTokens(ids1, ids1Span, out written, ids2);
+                Assert.Equal(OperationStatus.DestinationTooSmall, status);
+                Assert.Equal(0, written);
+
+                ids1Span = stackalloc int[ids1.Count + ids2.Count + 3];
+                status = bertTokenizer.BuildInputsWithSpecialTokens(ids1, ids1Span, out written, ids2);
+                Assert.Equal(OperationStatus.Done, status);
+                Assert.Equal(ids1Span.Length, written);
+                Assert.Equal(
+                        new int[] { bertTokenizer.ClsTokenId, bertTokenizer.ClsTokenId, 8, 6, 10, 11, 12, 7, bertTokenizer.SepTokenId, bertTokenizer.SepTokenId, bertTokenizer.ClsTokenId, 13, 14, 15, 5, bertTokenizer.SepTokenId, bertTokenizer.SepTokenId },
+                        ids1Span.ToArray());
+
+                ids1 = bertTokenizer.EncodeToIds(text1, addSpecialTokens: false);
+                Assert.Equal([8, 6, 10, 11, 12, 7], ids1);
+
+                ids2 = bertTokenizer.EncodeToIds(text2, addSpecialTokens: false);
+                Assert.Equal([13, 14, 15, 5], ids2);
+
+                Assert.Equal(
+                    [bertTokenizer.ClsTokenId, 8, 6, 10, 11, 12, 7, bertTokenizer.SepTokenId],
+                    bertTokenizer.BuildInputsWithSpecialTokens(ids1));
+
+                ids1Span = stackalloc int[1];
+                status = bertTokenizer.BuildInputsWithSpecialTokens(ids1, ids1Span, out written);
+                Assert.Equal(OperationStatus.DestinationTooSmall, status);
+                Assert.Equal(0, written);
+
+                ids1Span = stackalloc int[ids1.Count + 2];
+                status = bertTokenizer.BuildInputsWithSpecialTokens(ids1, ids1Span, out written);
+                Assert.Equal(OperationStatus.Done, status);
+                Assert.Equal(ids1Span.Length, written);
+                Assert.Equal(
+                        new int[] { bertTokenizer.ClsTokenId, 8, 6, 10, 11, 12, 7, bertTokenizer.SepTokenId },
+                        ids1Span.ToArray());
+
+                Assert.Equal(
+                    [bertTokenizer.ClsTokenId, 8, 6, 10, 11, 12, 7, bertTokenizer.SepTokenId, 13, 14, 15, 5, bertTokenizer.SepTokenId],
+                    bertTokenizer.BuildInputsWithSpecialTokens(ids1, ids2));
+
+                ids1Span = stackalloc int[1];
+                status = bertTokenizer.BuildInputsWithSpecialTokens(ids1, ids1Span, out written, ids2);
+                Assert.Equal(OperationStatus.DestinationTooSmall, status);
+                Assert.Equal(0, written);
+
+                ids1Span = stackalloc int[ids1.Count + ids2.Count + 3];
+                status = bertTokenizer.BuildInputsWithSpecialTokens(ids1, ids1Span, out written, ids2);
+                Assert.Equal(OperationStatus.Done, status);
+                Assert.Equal(ids1Span.Length, written);
+                Assert.Equal(
+                        new int[] { bertTokenizer.ClsTokenId, 8, 6, 10, 11, 12, 7, bertTokenizer.SepTokenId, 13, 14, 15, 5, bertTokenizer.SepTokenId },
+                        ids1Span.ToArray());
+            }
+            finally
+            {
+                File.Delete(vocabFile);
+            }
+        }
+
+        [Fact]
+        public void TestGetSpecialTokensMask()
+        {
+            //                   Ids: 0        1        2        3        4        5    6    7      8       9       10     11     12    13    14     15
+            string[] vocabTokens = ["[PAD]", "[UNK]", "[CLS]", "[SEP]", "[MASK]", "!", ",", "?", "hello", "world", "how", "are", "you", "i", "am", "fine"];
+
+            string vocabFile = WordPieceTests.CreateVocabFile(vocabTokens);
+
+            try
+            {
+                using Stream vocabStream = File.OpenRead(vocabFile);
+                BertTokenizer bertTokenizer = BertTokenizer.Create(vocabFile);
+
+                string text1 = "Hello, How are you?";
+                string text2 = "I am fine!";
+
+                var ids1 = bertTokenizer.EncodeToIds(text1);
+                Assert.Equal([bertTokenizer.ClsTokenId, 8, 6, 10, 11, 12, 7, bertTokenizer.SepTokenId], ids1);
+
+                var ids2 = bertTokenizer.EncodeToIds(text2);
+                Assert.Equal([bertTokenizer.ClsTokenId, 13, 14, 15, 5, bertTokenizer.SepTokenId], ids2);
+
+                Assert.Equal(
+                    [1, 0, 0, 0, 0, 0, 0, 1],
+                    bertTokenizer.GetSpecialTokensMask(ids1, tokenIds1: null, alreadyHasSpecialTokens: true));
+
+                Span<int> ids1Span = stackalloc int[1];
+                OperationStatus status = bertTokenizer.GetSpecialTokensMask(ids1, ids1Span, out int written, alreadyHasSpecialTokens: true);
+                Assert.Equal(OperationStatus.DestinationTooSmall, status);
+                Assert.Equal(0, written);
+
+                ids1Span = stackalloc int[ids1.Count];
+                status = bertTokenizer.GetSpecialTokensMask(ids1, ids1Span, out written, alreadyHasSpecialTokens: true);
+                Assert.Equal(OperationStatus.Done, status);
+                Assert.Equal(ids1.Count, written);
+                Assert.Equal(new int[] { 1, 0, 0, 0, 0, 0, 0, 1 }, ids1Span.ToArray());
+
+                Assert.Equal(
+                    [1, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 1],
+                    bertTokenizer.GetSpecialTokensMask(ids1, tokenIds1: ids2, alreadyHasSpecialTokens: true));
+
+                ids1Span = stackalloc int[1];
+                status = bertTokenizer.GetSpecialTokensMask(ids1, ids1Span, out written, ids2, alreadyHasSpecialTokens: true);
+                Assert.Equal(OperationStatus.DestinationTooSmall, status);
+                Assert.Equal(0, written);
+
+                ids1Span = stackalloc int[ids1.Count + ids2.Count];
+                status = bertTokenizer.GetSpecialTokensMask(ids1, ids1Span, out written, ids2, alreadyHasSpecialTokens: true);
+                Assert.Equal(OperationStatus.Done, status);
+                Assert.Equal(ids1.Count + ids2.Count, written);
+                Assert.Equal(new int[] { 1, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 1 }, ids1Span.ToArray());
+
+                ids1 = bertTokenizer.EncodeToIds(text1, addSpecialTokens: false);
+                Assert.Equal([8, 6, 10, 11, 12, 7], ids1);
+
+                ids2 = bertTokenizer.EncodeToIds(text2, addSpecialTokens: false);
+                Assert.Equal([13, 14, 15, 5], ids2);
+                Assert.Equal(
+                    [1, 0, 0, 0, 0, 0, 0, 1],
+                    bertTokenizer.GetSpecialTokensMask(ids1, tokenIds1: null, alreadyHasSpecialTokens: false));
+
+                ids1Span = stackalloc int[1];
+                status = bertTokenizer.GetSpecialTokensMask(ids1, ids1Span, out written, alreadyHasSpecialTokens: false);
+                Assert.Equal(OperationStatus.DestinationTooSmall, status);
+                Assert.Equal(0, written);
+
+                ids1Span = stackalloc int[ids1.Count + 2];
+                status = bertTokenizer.GetSpecialTokensMask(ids1, ids1Span, out written, alreadyHasSpecialTokens: false);
+                Assert.Equal(OperationStatus.Done, status);
+                Assert.Equal(ids1.Count + 2, written);
+                Assert.Equal(new int[] { 1, 0, 0, 0, 0, 0, 0, 1 }, ids1Span.ToArray());
+
+                Assert.Equal(
+                    [1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1],
+                    bertTokenizer.GetSpecialTokensMask(ids1, tokenIds1: ids2, alreadyHasSpecialTokens: false));
+
+                ids1Span = stackalloc int[1];
+                status = bertTokenizer.GetSpecialTokensMask(ids1, ids1Span, out written, ids2, alreadyHasSpecialTokens: false);
+                Assert.Equal(OperationStatus.DestinationTooSmall, status);
+                Assert.Equal(0, written);
+
+                ids1Span = stackalloc int[ids1.Count + ids2.Count + 3];
+                status = bertTokenizer.GetSpecialTokensMask(ids1, ids1Span, out written, ids2, alreadyHasSpecialTokens: false);
+                Assert.Equal(OperationStatus.Done, status);
+                Assert.Equal(ids1.Count + ids2.Count + 3, written);
+                Assert.Equal(new int[] { 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1 }, ids1Span.ToArray());
+            }
+            finally
+            {
+                File.Delete(vocabFile);
+            }
+        }
+
+        [Fact]
+        public void TestCreateTokenTypeIdsFromSequences()
+        {
+            //                   Ids: 0        1        2        3        4        5    6    7      8       9       10     11     12    13    14     15
+            string[] vocabTokens = ["[PAD]", "[UNK]", "[CLS]", "[SEP]", "[MASK]", "!", ",", "?", "hello", "world", "how", "are", "you", "i", "am", "fine"];
+
+            string vocabFile = WordPieceTests.CreateVocabFile(vocabTokens);
+
+            try
+            {
+                using Stream vocabStream = File.OpenRead(vocabFile);
+                BertTokenizer bertTokenizer = BertTokenizer.Create(vocabFile);
+
+                string text1 = "Hello, How are you?";
+                string text2 = "I am fine!";
+
+                var ids1 = bertTokenizer.EncodeToIds(text1, addSpecialTokens: false);
+                Assert.Equal([8, 6, 10, 11, 12, 7], ids1);
+
+                var ids2 = bertTokenizer.EncodeToIds(text2, addSpecialTokens: false);
+                Assert.Equal([13, 14, 15, 5], ids2);
+
+                Assert.Equal(
+                    [0, 0, 0, 0, 0, 0, 0, 0],
+                    bertTokenizer.CreateTokenTypeIdsFromSequences(ids1));
+
+                Span<int> ids1Span = stackalloc int[1];
+                OperationStatus status = bertTokenizer.CreateTokenTypeIdsFromSequences(ids1, ids1Span, out int written);
+                Assert.Equal(OperationStatus.DestinationTooSmall, status);
+                Assert.Equal(0, written);
+
+                ids1Span = stackalloc int[ids1.Count + 2];
+                status = bertTokenizer.CreateTokenTypeIdsFromSequences(ids1, ids1Span, out written);
+                Assert.Equal(OperationStatus.Done, status);
+                Assert.Equal(ids1.Count + 2, written);
+                Assert.Equal(new int[] { 0, 0, 0, 0, 0, 0, 0, 0 }, ids1Span.ToArray());
+
+                Assert.Equal(
+                    [0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1],
+                    bertTokenizer.CreateTokenTypeIdsFromSequences(ids1, ids2));
+
+                ids1Span = stackalloc int[1];
+                status = bertTokenizer.CreateTokenTypeIdsFromSequences(ids1, ids1Span, out written, ids2);
+                Assert.Equal(OperationStatus.DestinationTooSmall, status);
+                Assert.Equal(0, written);
+
+                ids1Span = stackalloc int[ids1.Count + ids2.Count + 3];
+                status = bertTokenizer.CreateTokenTypeIdsFromSequences(ids1, ids1Span, out written, ids2);
+                Assert.Equal(OperationStatus.Done, status);
+                Assert.Equal(ids1Span.Length, written);
+                Assert.Equal(new int[] { 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1 }, ids1Span.ToArray());
+            }
+            finally
+            {
+                File.Delete(vocabFile);
+            }
+        }
+    }
+}

test/Microsoft.ML.Tokenizers.Tests/BpeTests.cs

@@ -251,7 +251,7 @@ namespace Microsoft.ML.Tokenizers.Tests
 
             try
             {
-                BpeTokenizer bpe = BpeTokenizer.Create(vocabFile: vocabFile, mergesFile: mergesFile, preTokenizer: PreTokenizer.CreateWhiteSpace(), normalizer: null, unknownToken: unknownToken,
+                BpeTokenizer bpe = BpeTokenizer.Create(vocabFile: vocabFile, mergesFile: mergesFile, preTokenizer: PreTokenizer.CreateWordOrNonWordPreTokenizer(), normalizer: null, unknownToken: unknownToken,
                                     continuingSubwordPrefix: continuingSubwordPrefix, endOfWordSuffix: endOfWordSuffix, fuseUnknownTokens: fuseUnknownToken);
                 Tokenizer tokenizer = bpe;
                 IReadOnlyList<EncodedToken> encoding = tokenizer.EncodeToTokens(sentence, out _);
@@ -500,7 +500,7 @@ namespace Microsoft.ML.Tokenizers.Tests
             using Stream vocabStream = File.OpenRead(Path.Combine(@"Gpt-2", "vocab.json"));
             using Stream mergesStream = File.OpenRead(Path.Combine(@"Gpt-2", "merges.txt"));
 
-            var bpeTokenizer = BpeTokenizer.Create(vocabStream, mergesStream, PreTokenizer.CreateWhiteSpace(addedTokens), normalizer: null, addedTokens: addedTokens, unknownToken: "<|endoftext|>");
+            var bpeTokenizer = BpeTokenizer.Create(vocabStream, mergesStream, PreTokenizer.CreateWordOrNonWordPreTokenizer(addedTokens), normalizer: null, addedTokens: addedTokens, unknownToken: "<|endoftext|>");
 
             string input = "Hello, y'all! <issue_comment>How are you 😁 ?<|endoftext|>";
 
@@ -556,7 +556,7 @@ namespace Microsoft.ML.Tokenizers.Tests
             emptyVocabStream.Position = 0;
 
             return BpeTokenizer.Create(
-                        vocabStream: emptyVocabStream, mergesStream: null, preTokenizer: preTokenizer ?? PreTokenizer.CreateWhiteSpace(), normalizer: normalizer, unknownToken: "Ukn");
+                        vocabStream: emptyVocabStream, mergesStream: null, preTokenizer: preTokenizer ?? PreTokenizer.CreateWordOrNonWordPreTokenizer(), normalizer: normalizer, unknownToken: "Ukn");
         }
     }
 }

test/Microsoft.ML.Tokenizers.Tests/PreTokenizerTests.cs

@@ -18,18 +18,25 @@ namespace Microsoft.ML.Tokenizers.Tests
             {
                 yield return new object[]
                 {
-                    PreTokenizer.CreateWhiteSpace(),
+                    PreTokenizer.CreateWordOrNonWordPreTokenizer(),
                     "How are you doing?",
                     new (int Offset, int Length)[] { (0, 3), (4, 3), (8, 3), (12, 5), (17, 1), }
                 };
 
                 yield return new object[]
                 {
-                    PreTokenizer.CreateWhiteSpace(),
+                    PreTokenizer.CreateWordOrNonWordPreTokenizer(),
                     "I_am_Just_Fine!",
                     new (int Offset, int Length)[] { (0, 14), (14, 1) }
                 };
 
+                yield return new object[]
+                {
+                    PreTokenizer.CreateWhiteSpacePreTokenizer(),
+                    "Hello, how are you doing?!",
+                    new (int Offset, int Length)[] { (0, 6), (7, 3), (11, 3), (15, 3), (19, 7) }
+                };
+
                 yield return new object[]
                 {
                     new SpacePreTokenizer(),
@@ -61,9 +68,9 @@ namespace Microsoft.ML.Tokenizers.Tests
         }
 
         [Fact]
-        public void TestWhiteSpacePreTokenizer()
+        public void TestWordOrNonWordPreTokenizer()
         {
-            Assert.Empty(PreTokenizer.CreateWhiteSpace().PreTokenize((string)null!));
+            Assert.Empty(PreTokenizer.CreateWordOrNonWordPreTokenizer().PreTokenize((string)null!));
         }
 
         public class SpacePreTokenizer : PreTokenizer

test/Microsoft.ML.Tokenizers.Tests/WordPieceTests.cs

@@ -0,0 +1,221 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+using System;
+using System.Buffers;
+using System.Collections.Generic;
+using System.IO;
+using System.Linq;
+using Xunit;
+
+namespace Microsoft.ML.Tokenizers.Tests
+{
+    public class WordPieceTests
+    {
+        static string[] _vocabTokens = ["[UNK]", "[CLS]", "[SEP]", "want", "##want", "##ed", "wa", "un", "runn", "##ing"];
+
+        internal static string CreateVocabFile(string[] vocabTokens)
+        {
+            string vocabFile = Path.GetTempFileName();
+            File.WriteAllLines(vocabFile, vocabTokens);
+            return vocabFile;
+        }
+
+        [Fact]
+        public void TestCreation()
+        {
+            string vocabFile = CreateVocabFile(_vocabTokens);
+
+            try
+            {
+                using Stream vocabStream = File.OpenRead(vocabFile);
+                WordPieceTokenizer[] wordPieceTokenizers = [WordPieceTokenizer.Create(vocabFile), WordPieceTokenizer.Create(vocabStream)];
+
+                foreach (var tokenizer in wordPieceTokenizers)
+                {
+                    Assert.NotNull(tokenizer.PreTokenizer);
+                    Assert.Equal("[UNK]", tokenizer.UnknownToken);
+                    Assert.Equal(0, tokenizer.UnknownTokenId);
+                    Assert.Null(tokenizer.Normalizer);
+                    Assert.Equal(100, tokenizer.MaxInputCharsPerWord);
+                    Assert.Equal("##", tokenizer.ContinuingSubwordPrefix);
+                }
+            }
+            finally
+            {
+                File.Delete(vocabFile);
+            }
+        }
+
+        [Fact]
+        public void TestTokenization()
+        {
+            string vocabFile = CreateVocabFile(_vocabTokens);
+
+            try
+            {
+                WordPieceTokenizer tokenizer = WordPieceTokenizer.Create(vocabFile);
+
+                Assert.Null(tokenizer.SpecialTokens);
+
+                IReadOnlyList<EncodedToken> tokens = tokenizer.EncodeToTokens("", out _);
+                Assert.Empty(tokens);
+                Assert.Equal(0, tokenizer.CountTokens(""));
+                IReadOnlyList<int> ids = tokenizer.EncodeToIds("");
+                Assert.Empty(ids);
+                int index = tokenizer.GetIndexByTokenCount("", maxTokenCount: 10, normalizedString: out _, tokenCount: out int tokenCount);
+                Assert.Equal(0, index);
+                Assert.Equal(0, tokenCount);
+                index = tokenizer.GetIndexByTokenCountFromEnd("", maxTokenCount: 10, normalizedString: out _, tokenCount: out tokenCount);
+                Assert.Equal(0, index);
+                Assert.Equal(0, tokenCount);
+
+                string text = "unwanted running";
+                tokens = tokenizer.EncodeToTokens(text, out _);
+                Assert.Equal(
+                    [
+                        new EncodedToken(7, "un", new Range(0, 2)),
+                        new EncodedToken(4, "##want", new Range(2, 6)),
+                        new EncodedToken(5, "##ed", new Range(6, 8)),
+                        new EncodedToken(8, "runn", new Range(9, 13)),
+                        new EncodedToken(9, "##ing", new Range(13, 16))
+                    ],
+                    tokens
+                );
+
+                ids = tokenizer.EncodeToIds(text);
+                Assert.Equal([7, 4, 5, 8, 9], ids);
+
+                int[] expectedTokenCount = [0, 0, 3, 3, 5];
+                for (int i = 1; i <= 5; i++)
+                {
+                    Assert.Equal(ids.Take(expectedTokenCount[i - 1]).ToArray(), tokenizer.EncodeToIds(text, maxTokenCount: i, normalizedText: out _, out tokenCount));
+                }
+
+                Assert.Equal(text, tokenizer.Decode(ids));
+
+                Span<char> buffer = stackalloc char[text.Length];
+                for (int i = 0; i < text.Length - 1; i++)
+                {
+                    Span<char> bufferSlice = buffer.Slice(0, i);
+                    OperationStatus result = tokenizer.Decode(ids, bufferSlice, out int idsConsumed, out int charsWritten);
+                    Assert.Equal(OperationStatus.DestinationTooSmall, result);
+
+                    int j = 0;
+
+                    while (i >= tokens[j].Offset.End.Value)
+                    {
+                        j++;
+                    }
+
+                    Assert.Equal(j, idsConsumed);
+                    Assert.Equal(j == 0 ? 0 : tokens[j - 1].Offset.End.Value, charsWritten);
+                    Assert.Equal(j == 0 ? "" : text.Substring(0, tokens[j - 1].Offset.End.Value), bufferSlice.Slice(0, charsWritten).ToString());
+                }
+
+                Assert.Equal(5, tokenizer.CountTokens(text));
+
+                int[] expectedIndexes = [0, 0, 8, 9, 16];
+                expectedTokenCount = [0, 0, 3, 3, 5];
+
+                for (int i = 1; i <= 5; i++)
+                {
+                    index = tokenizer.GetIndexByTokenCount(text, maxTokenCount: i, normalizedString: out _, out tokenCount);
+                    Assert.Equal(expectedTokenCount[i - 1], tokenCount);
+                    Assert.Equal(expectedIndexes[i - 1], index);
+                }
+
+                expectedIndexes = [16, 9, 8, 8, 0];
+                expectedTokenCount = [0, 2, 2, 2, 5];
+
+                for (int i = 1; i <= 5; i++)
+                {
+                    index = tokenizer.GetIndexByTokenCountFromEnd(text, maxTokenCount: i, normalizedString: out _, out tokenCount);
+                    Assert.Equal(expectedTokenCount[i - 1], tokenCount);
+                    Assert.Equal(expectedIndexes[i - 1], index);
+                }
+            }
+            finally
+            {
+                File.Delete(vocabFile);
+            }
+        }
+
+        [Fact]
+        public void TestTokenizationWithUnknownTokens()
+        {
+            string vocabFile = CreateVocabFile(_vocabTokens);
+
+            try
+            {
+                WordPieceTokenizer tokenizer = WordPieceTokenizer.Create(vocabFile);
+
+                string text = "unwantedX running";
+
+                IReadOnlyList<EncodedToken> tokens = tokenizer.EncodeToTokens(text, out _);
+                Assert.Equal(
+                    [
+                        new EncodedToken(0, "[UNK]", new Range(0, 9)),
+                        new EncodedToken(8, "runn",  new Range(10, 14)),
+                        new EncodedToken(9, "##ing", new Range(14, 17))
+                    ],
+                    tokens
+                );
+
+                IReadOnlyList<int> ids = tokenizer.EncodeToIds(text);
+                Assert.Equal([0, 8, 9], ids);
+
+                Assert.Equal("[UNK] running", tokenizer.Decode(ids));
+            }
+            finally
+            {
+                File.Delete(vocabFile);
+            }
+        }
+
+        [Fact]
+        public void TestTokenizationWithSpecialTokens()
+        {
+            string vocabFile = CreateVocabFile(_vocabTokens);
+
+            try
+            {
+                Dictionary<string, int> specialTokens = new Dictionary<string, int>
+                {
+                    { "[UNK]", 0 }, { "[CLS]", 1 }, { "[SEP]", 2 }
+                };
+                WordPieceTokenizer tokenizer = WordPieceTokenizer.Create(vocabFile, specialTokens: specialTokens);
+
+                Assert.Equal(specialTokens, tokenizer.SpecialTokens);
+
+                string text = "[UNK] unwanted [SEP][CLS] running [CLS]";
+
+                IReadOnlyList<EncodedToken> tokens = tokenizer.EncodeToTokens(text, out _);
+                Assert.Equal(
+                    [
+                        new EncodedToken(0, "[UNK]", new Range(0, 5)),
+                        new EncodedToken(7, "un", new Range(6, 8)),
+                        new EncodedToken(4, "##want", new Range(8, 12)),
+                        new EncodedToken(5, "##ed", new Range(12, 14)),
+                        new EncodedToken(2, "[SEP]", new Range(15, 20)),
+                        new EncodedToken(1, "[CLS]", new Range(20, 25)),
+                        new EncodedToken(8, "runn", new Range(26, 30)),
+                        new EncodedToken(9, "##ing", new Range(30, 33)),
+                        new EncodedToken(1, "[CLS]", new Range(34, 39)),
+                    ],
+                    tokens
+                );
+
+                IReadOnlyList<int> ids = tokenizer.EncodeToIds(text);
+                Assert.Equal([0, 7, 4, 5, 2, 1, 8, 9, 1], ids);
+
+                Assert.Equal("[UNK] unwanted [SEP] [CLS] running [CLS]", tokenizer.Decode(ids));
+            }
+            finally
+            {
+                File.Delete(vocabFile);
+            }
+        }
+    }
+}