app.py

import os
import random
from collections import Counter

import gradio as gr
import polars as pl
import spaces
import torch

from metric import PerplexityCalculator

IS_DEBUG = False

os.environ["OMP_NUM_THREADS"] = "1"
os.environ["TOKENIZERS_PARALLELISM"] = "false"
PAD_TOKEN_LABEL_ID = torch.nn.CrossEntropyLoss().ignore_index
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")

df_sample_submission = pl.read_csv("data/sample_submission.csv")
text_list = df_sample_submission.get_column("text").to_list()
text_counters = [Counter(text.split()) for text in text_list]

# Model Loading
if not IS_DEBUG:
    scorer = PerplexityCalculator("google/gemma-2-9b")


@spaces.GPU()
def inference(text: str, progress=gr.Progress(track_tqdm=True)):
    score = -1

    if IS_DEBUG:
        index_text = f"[DEBUG] "
    else:
        index_text = ""
        score = scorer.get_perplexity(text)

    input_counter = Counter(text.split())
    is_match_list = [input_counter == text_counter for text_counter in text_counters]

    if any(is_match_list):
        index = is_match_list.index(True)
        index_text += f"Task #{index}"
        return score, index_text
    else:
        index_text += "No Match to All Tasks"
        gr.Warning(index_text)
        return score, index_text


def random_inference(text: str, progress=gr.Progress(track_tqdm=True)):
    if text == "":
        text = text_list[0]

    words = text.split()
    random.shuffle(words)

    random_text = " ".join(words)

    score, index_text = inference(random_text)
    return random_text, score, index_text


if __name__ == "__main__":
    theme = gr.themes.Default(
        primary_hue=gr.themes.colors.emerald,
        secondary_hue=gr.themes.colors.green,
    )

    with gr.Blocks(theme=theme) as demo:
        with gr.Column():
            title = gr.Markdown(
                "<h1 style='text-align: center; margin-bottom: 1rem'>Gemma-2-9b Perplexity Calculator</h1>"
            )

            with gr.Row():
                with gr.Column():
                    input_text = gr.Textbox(label="Text Input")

                    output_perplexity = gr.Number(label="Perplexity", render=False)
                    output_index = gr.Textbox(label="Index", render=False)

                    with gr.Row():
                        clear_button = gr.ClearButton(
                            [input_text, output_perplexity, output_index]
                        )
                        random_button = gr.Button("Randomize", variant="secondary")
                        submit_button = gr.Button("Run", variant="primary")

                with gr.Column():
                    output_perplexity.render()
                    output_index.render()

            sample_table = gr.Dataframe(
                df_sample_submission, label="Sample Submission", type="polars"
            )

        submit_button.click(
            inference, inputs=[input_text], outputs=[output_perplexity, output_index]
        )
        input_text.submit(
            inference, inputs=[input_text], outputs=[output_perplexity, output_index]
        )
        random_button.click(
            random_inference,
            inputs=[input_text],
            outputs=[input_text, output_perplexity, output_index],
        )

    demo.queue().launch()