app.py

import gradio as gr
import random
from transformers import AutoTokenizer
from huggingface_hub import login, logout
from markupsafe import escape
from gradio_huggingfacehub_search import HuggingfaceHubSearch
from fractions import Fraction

def random_light_color():
    """Generates a random light color with black text."""
    return f"hsl({random.randint(0, 360)}, 100%, 80%)"

def utf8_tokens(tokens):
    """Generates UTF-8 token representations with valid Unicode for each token."""
    utf8_representation = []
    for token in tokens:
        try:
            utf8_bytes = token.encode('utf-8')
            utf8_hex = " ".join([f"<0x{byte:02X}>" for byte in utf8_bytes])
            unicode_token = utf8_bytes.decode('utf-8')
            utf8_representation.append(
                f'<span style="background-color:{random_light_color()}; color: black;">{escape(unicode_token)} {utf8_hex}</span>'
            )
        except UnicodeDecodeError:
            utf8_representation.append(
                f'<span style="background-color:{random_light_color()}; color: brown;">{escape(token)} {utf8_hex}</span>'
            )
    return " ".join(utf8_representation)

def tokenize_text(tokenizer_name_1, tokenizer_name_2, text, hf_token=None):
    def tokenize_with_model(tokenizer_name):
        try:
            tokenizer = AutoTokenizer.from_pretrained(tokenizer_name, use_auth_token=hf_token)
            tokens = tokenizer.tokenize(text)
            word_count = len(text.split())
            token_count = len(tokens)
            ratio_simplified = f"{Fraction(word_count, token_count).numerator}/{Fraction(word_count, token_count).denominator}" if token_count > 0 else "N/A"
            colored_tokens = [
                f'<span style="background-color:{random_light_color()}; color: black;">{escape(token)}</span>' for token in tokens
            ]
            tokenized_text = " ".join(colored_tokens)
            utf8_representation = utf8_tokens(tokens)
            return tokenized_text, token_count, word_count, ratio_simplified, utf8_representation
        except Exception as e:
            return f"Error loading tokenizer {tokenizer_name}: {str(e)}", 0, 0, "N/A", ""
    
    if hf_token:
        login(hf_token)
    
    tokenizer_1_output = tokenize_with_model(tokenizer_name_1)
    tokenizer_2_output = tokenize_with_model(tokenizer_name_2)

    try:
        logout()
    except Exception as err:
        pass

    return (
        f"<p><strong>Tokenizer 1:</strong><br>{tokenizer_1_output[0]}</p>", 
        f"Tokenizer 1 - Total tokens: {tokenizer_1_output[1]}, Total words: {tokenizer_1_output[2]}, Word/Token ratio: {tokenizer_1_output[3]}",
        f"<p>{tokenizer_1_output[4]}</p>",
        f"<p><strong>Tokenizer 2:</strong><br>{tokenizer_2_output[0]}</p>",
        f"Tokenizer 2 - Total tokens: {tokenizer_2_output[1]}, Total words: {tokenizer_2_output[2]}, Word/Token ratio: {tokenizer_2_output[3]}",
        f"<p>{tokenizer_2_output[4]}</p>"
    )

def fill_example_text(example_text):
    """Fills the textbox with the selected example."""
    return example_text

examples = {
    "Example 1 (en)": "Hugging Face's tokenizers are really cool!",
    "Example 2 (en)": "Gradio makes building UIs so easy and intuitive.",
    "Example 3 (en)": "Machine learning models often require extensive training data.",
    "Example 4 (ta)": "விரைவு பழுப்பு நரி சோம்பேறி நாய் மீது குதிக்கிறது",
    "Example 5 (si)": "ඉක්මන් දුඹුරු නරියා කම්මැලි බල්ලා උඩින් පනියි"
}

with gr.Blocks() as demo:
    with gr.Row():
        with gr.Column():
            tokenizer_search_1 = HuggingfaceHubSearch(
                label="Search Huggingface Hub for Tokenizer 1",
                placeholder="Search for Tokenizer 1",
                search_type="model"
            )
        with gr.Column():
            tokenizer_search_2 = HuggingfaceHubSearch(
                label="Search Huggingface Hub for Tokenizer 2",
                placeholder="Search for Tokenizer 2",
                search_type="model"
            )
    
    example_dropdown = gr.Dropdown(label="Select Example", choices=list(examples.keys()), value="Example 1")
    input_text = gr.Textbox(label="Input Text", lines=5)
    
    with gr.Accordion("Hugging Face Token (Optional)", open=False):
        hf_token = gr.Textbox(label="Hugging Face Token", placeholder="Enter HF token if needed for private tokenizers")
    
    with gr.Row():
        with gr.Column():
            gr.Markdown("### Tokenizer 1 Outputs")
            tokenized_output_1 = gr.HTML(label="Tokenizer 1 - Tokenized Text")
            token_count_label_1 = gr.Label(label="Tokenizer 1 - Token Count and Word Count")
            with gr.Accordion("Tokenizer 1 - UTF-8 Decoded Text", open=False):
                utf8_output_1 = gr.HTML(label="Tokenizer 1 - UTF-8 Decoded Text")
        
        with gr.Column():
            gr.Markdown("### Tokenizer 2 Outputs")
            tokenized_output_2 = gr.HTML(label="Tokenizer 2 - Tokenized Text")
            token_count_label_2 = gr.Label(label="Tokenizer 2 - Token Count and Word Count")
            with gr.Accordion("Tokenizer 2 - UTF-8 Decoded Text", open=False):
                utf8_output_2 = gr.HTML(label="Tokenizer 2 - UTF-8 Decoded Text")
    
    example_dropdown.change(fn=lambda x: fill_example_text(examples[x]), inputs=example_dropdown, outputs=input_text)
    
    input_text.change(tokenize_text, 
                      inputs=[tokenizer_search_1, tokenizer_search_2, input_text, hf_token], 
                      outputs=[tokenized_output_1, token_count_label_1, utf8_output_1, tokenized_output_2, token_count_label_2, utf8_output_2])
    
    tokenizer_search_1.change(tokenize_text, 
                              inputs=[tokenizer_search_1, tokenizer_search_2, input_text, hf_token], 
                              outputs=[tokenized_output_1, token_count_label_1, utf8_output_1, tokenized_output_2, token_count_label_2, utf8_output_2])

    tokenizer_search_2.change(tokenize_text, 
                              inputs=[tokenizer_search_1, tokenizer_search_2, input_text, hf_token], 
                              outputs=[tokenized_output_1, token_count_label_1, utf8_output_1, tokenized_output_2, token_count_label_2, utf8_output_2])

demo.launch()