app.py

from gradio import Interface
import gradio as gr
import aranizer
from aranizer import aranizer_bpe50k, aranizer_bpe64k, aranizer_bpe86k, aranizer_sp32k, aranizer_sp50k, aranizer_sp64k, aranizer_sp86k
from transformers import AutoTokenizer

# Load additional tokenizers from transformers
gpt_13b_tokenizer = AutoTokenizer.from_pretrained("FreedomIntelligence/AceGPT-13B")
gpt_7b_tokenizer = AutoTokenizer.from_pretrained("FreedomIntelligence/AceGPT-7B")
jais_13b_tokenizer = AutoTokenizer.from_pretrained("inception-mbzuai/jais-13b")

# List of available tokenizers and a dictionary to load them
tokenizer_options = [
    "aranizer_bpe50k", "aranizer_bpe64k", "aranizer_bpe86k",
    "aranizer_sp32k", "aranizer_sp50k", "aranizer_sp64k", "aranizer_sp86k",
    "FreedomIntelligence/AceGPT-13B",  # Previously added GPT tokenizer
    "FreedomIntelligence/AceGPT-7B",   # Another previously added GPT tokenizer
    "inception-mbzuai/jais-13b"        # Adding the new tokenizer to the options
]

tokenizers = {
    "aranizer_bpe50k": aranizer_bpe50k.get_tokenizer,
    "aranizer_bpe64k": aranizer_bpe64k.get_tokenizer,
    "aranizer_bpe86k": aranizer_bpe86k.get_tokenizer,
    "aranizer_sp32k": aranizer_sp32k.get_tokenizer,
    "aranizer_sp50k": aranizer_sp50k.get_tokenizer,
    "aranizer_sp64k": aranizer_sp64k.get_tokenizer,
    "aranizer_sp86k": aranizer_sp86k.get_tokenizer,
    "FreedomIntelligence/AceGPT-13B": lambda: gpt_13b_tokenizer,
    "FreedomIntelligence/AceGPT-7B": lambda: gpt_7b_tokenizer,
    "inception-mbzuai/jais-13b": lambda: jais_13b_tokenizer  # Adding the new Jais tokenizer
}

def compare_tokenizers(tokenizer_name, text):
    # Handle the transformer tokenizers separately due to API differences
    if tokenizer_name in ["FreedomIntelligence/AceGPT-13B", "FreedomIntelligence/AceGPT-7B", "inception-mbzuai/jais-13b"]:
        tokenizer = tokenizers[tokenizer_name]()
        tokens = tokenizer.tokenize(text)
        encoded_output = tokenizer.encode(text, add_special_tokens=True, return_tensors="pt")
        decoded_text = tokenizer.decode(encoded_output[0], skip_special_tokens=True)
    else:
        # AraNizer tokenizers
        tokenizer = tokenizers[tokenizer_name]()
        tokens = tokenizer.tokenize(text)
        encoded_output = tokenizer.encode(text, add_special_tokens=True)
        decoded_text = tokenizer.decode(encoded_output)
    
    # Prepare the results to be displayed
    results = [(tokenizer_name, tokens, encoded_output, decoded_text)]
    return results

# Define the Gradio interface components with a dropdown for model selection
inputs_component = [
    gr.Dropdown(choices=tokenizer_options, label="Select Tokenizer"),
    gr.Textbox(lines=2, placeholder="Enter text here...", label="Input Text")
]

outputs_component = gr.Dataframe(
    headers=["Tokenizer", "Tokens", "Encoded Output", "Decoded Text"], 
    label="Results",
    type="pandas"
)

# Setting up the interface
iface = Interface(
    fn=compare_tokenizers, 
    inputs=inputs_component, 
    outputs=outputs_component, 
    title="Tokenizer Comparison",
    live=True
)

# Launching the Gradio app
iface.launch()