app.py

from transformers import AutoTokenizer
import gradio as gr


def tokenize(input_text):
    llama_tokens = len(
        llama_tokenizer(input_text, add_special_tokens=True)["input_ids"]
    )
    mistral_tokens = len(
        mistral_tokenizer(input_text, add_special_tokens=True)["input_ids"]
    )
    gpt2_tokens = len(gpt2_tokenizer(input_text, add_special_tokens=True)["input_ids"])
    gpt_neox_tokens = len(
        gpt_neox_tokenizer(input_text, add_special_tokens=True)["input_ids"]
    )
    falcon_tokens = len(
        falcon_tokenizer(input_text, add_special_tokens=True)["input_ids"]
    )
    phi2_tokens = len(phi2_tokenizer(input_text, add_special_tokens=True)["input_ids"])
    t5_tokens = len(t5_tokenizer(input_text, add_special_tokens=True)["input_ids"])

    results = {
        "LLaMa": llama_tokens,
        "Mistral": mistral_tokens,
        "GPT-2/GPT-J": gpt2_tokens,
        "GPT-NeoX": gpt_neox_tokens,
        "Falcon": falcon_tokens,
        "Phi-2": phi2_tokens,
        "T5": t5_tokens,
    }

    # Sort the results in descending order based on token length
    sorted_results = sorted(results.items(), key=lambda x: x[1], reverse=True)

    return "\n".join([f"{model}: {tokens}" for model, tokens in sorted_results])


if __name__ == "__main__":
    llama_tokenizer = AutoTokenizer.from_pretrained("TheBloke/Llama-2-7B-fp16")
    mistral_tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1")
    gpt2_tokenizer = AutoTokenizer.from_pretrained("gpt2")
    gpt_neox_tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-neox-20b")
    falcon_tokenizer = AutoTokenizer.from_pretrained("tiiuae/falcon-7b")
    phi2_tokenizer = AutoTokenizer.from_pretrained("microsoft/phi-2")
    t5_tokenizer = AutoTokenizer.from_pretrained("google/flan-t5-xxl")

    iface = gr.Interface(fn=tokenize, inputs=gr.Textbox(lines=7), outputs="text")
    iface.launch()