AIAS1/2_nlp_sdks/tokenizer_sdk

huggingface Tokenizer

https://github.com/deepjavalibrary/djl/blob/master/extensions/tokenizers/README.md

1. pom 配置

    <dependency>
        <groupId>ai.djl.huggingface</groupId>
        <artifactId>tokenizers</artifactId>
        <version>0.19.0</version>
    </dependency>

2. 例子代码

private static final HuggingFaceTokenizer tokenizer;

# 声明
static {
    try {
        tokenizer =
                HuggingFaceTokenizer.builder()
                        .optManager(manager)
                        .optPadding(true)
                        .optPadToMaxLength()
                        .optMaxLength(MAX_LENGTH)
                        .optTruncation(true)
                        .optTokenizerName("openai/clip-vit-large-patch14")
                        .build();
        // sentence-transformers/msmarco-distilbert-dot-v5
        // openai/clip-vit-large-patch14
        // https://huggingface.co/sentence-transformers/msmarco-distilbert-dot-v5
        // https://huggingface.co/runwayml/stable-diffusion-v1-5/blob/main/tokenizer/tokenizer_config.json
    } catch (IOException e) {
        throw new RuntimeException(e);
    }
}

# 使用
List<String> tokens = tokenizer.tokenize(prompt);

下面方法已废弃，仅用于学习参考，请使用上面的库及用法

Transformer的常用Tokenizer系列 - Java实现

这个sdk里包含了用于自然语言处理的tokenizer（分词器）。 切词输出的token序列，兼容huggingface（一个python实现的知名NLP库）。 java实现的Tokenizer有助于在java环境部署NLP模型。

包含的tokenizer如下：

• SimpleTokenizer
• BertTokenizer
• WordpieceTokenizer
• BertFullTokenizer
• ClipBPETokenizer
• GPT2Tokenizer（Todo）

Python transformer 常用tokenizer:

运行例子 - TokenizerExample

• 训练的例子代码：

    @Test
    public void testSimpleTokenizer() {
        logger.info("testSimpleTokenizer:");
        String input = "This tokenizer generates tokens for input sentence";
        String delimiter = " ";
        SimpleTokenizer tokenizer = new SimpleTokenizer(delimiter);
        List<String> tokens = tokenizer.tokenize(input);
        logger.info("Tokens: {}", tokens);

    }

    @Test
    public void testBertTokenizer() {
        logger.info("testBertTokenizer:");
        String question = "When did Radio International start broadcasting?";
        String paragraph = "Radio International was a general entertainment Channel. Which operated between December 1983 and April 2001";

        BertTokenizer tokenizer = new BertTokenizer();
        BertToken bertToken = tokenizer.encode(question, paragraph);
        logger.info("Tokens: {}", bertToken.getTokens());
        logger.info("TokenTypes: {}", bertToken.getTokenTypes());
        logger.info("AttentionMask: {}", bertToken.getAttentionMask());
    }

    @Test
    public void testWordpieceTokenizer() throws IOException {
        logger.info("testWordpieceTokenizer:");
        //replace vocab.txt with your vocabulary
        Path filePath = Paths.get("src/test/resources/vocab.txt");
        SimpleVocabulary vocabulary =
                SimpleVocabulary.builder()
                        .optMinFrequency(1)
                        .addFromTextFile(filePath)
                        .optUnknownToken("[UNK]")
                        .build();
        WordpieceTokenizer wordpieceTokenizer = new WordpieceTokenizer(vocabulary, "[UNK]", 200);
        String input = "This tokenizer generates tokens for input sentence";
        List<String> tokens = wordpieceTokenizer.tokenize(input);
        logger.info("Tokens: {}", tokens);
    }
    
    @Test
    public void testBertFullTokenizer() throws IOException {
        logger.info("testBertFullTokenizer:");
        //replace vocab.txt with your vocabulary
        Path filePath = Paths.get("src/test/resources/vocab.txt");
        SimpleVocabulary vocabulary =
                SimpleVocabulary.builder()
                        .optMinFrequency(1)
                        .addFromTextFile(filePath)
                        .optUnknownToken("[UNK]")
                        .build();
        //subword tokenizer
        BertFullTokenizer tokenizer = new BertFullTokenizer(vocabulary, true);

        String input = "This tokenizer generates tokens for input sentence";
        List<String> tokens = tokenizer.tokenize(input);
        logger.info("Tokens: {}", tokens);

    }

    @Test
    public void testClipBPETokenizer() throws IOException, ModelException {
        logger.info("testClipBPETokenizer:");
        ClipBPETokenizer tokenizer = new ClipBPETokenizer("src/test/resources/bpe_simple_vocab_16e6.txt");
        List<Integer> tokens = tokenizer.encode("we will test a diagram");
        logger.info("Tokens: {}", tokens);
    }

• 运行成功后，命令行应该看到下面的信息:

#BlazingText 测试文本：
[INFO ] - testSimpleTokenizer:
[INFO ] - Tokens: [This, tokenizer, generates, tokens, for, input, sentence]

[INFO ] - testBertTokenizer:
[INFO ] - Tokens: [[CLS], When, did, Radio, International, start, broadcasting, ?, [SEP], Radio, International, was, a, general, entertainment, Channel, ., Which, operated, between, December, 1983, and, April, 2001, [SEP]]
[INFO ] - TokenTypes: [0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
[INFO ] - AttentionMask: [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]

[INFO ] - testWordpieceTokenizer:
[INFO ] - Tokens: [[UNK], token, ##izer, generates, token, ##s, for, input, sentence]

[INFO ] - testBertFullTokenizer:
[INFO ] - Tokens: [this, token, ##izer, generates, token, ##s, for, input, sentence]

[INFO ] - testClipBPETokenizer:
[INFO ] - Tokens: [649, 751, 1628, 320, 22697]

官网：

官网链接

Git地址：

Github链接
Gitee链接