app.py

import gradio as gr
import numpy as np
import pandas as pd
import plotly.express as px
import torch
from transformers import AutoTokenizer, AutoModelForMaskedLM


model_checkpoint = "xlm-roberta-base"

tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)
model = AutoModelForMaskedLM.from_pretrained(model_checkpoint)
mask_token = tokenizer.mask_token


def add_mask(target_word, text):
    text_mask = text.replace(target_word, mask_token)
    return text_mask


def eval_prob(target_word, text):
    text_mask = add_mask(target_word, text)

    # Get index of target_word
    target_idx = tokenizer.encode(target_word)[2]

    # Get logits
    inputs = tokenizer(text_mask, return_tensors="pt")
    token_logits = model(**inputs).logits

    # Find the location of the MASK and extract its logits
    mask_token_index = torch.where(inputs["input_ids"] == tokenizer.mask_token_id)[1]
    mask_token_logits = token_logits[0, mask_token_index, :]

    # Convert logits to softmax probability
    logits = mask_token_logits[0].tolist()
    probs = torch.nn.functional.softmax(torch.tensor([logits]), dim=1)[0]

    # Get probability of target word filling the MASK
    # result = float(probs[target_idx])

    return probs, target_idx


def plot_results(target_word, text):
    probs, target_idx = eval_prob(target_word, text)

    # Sort tokens based on probability scores
    words = [
        tokenizer.decode(idx) for idx in torch.sort(probs, descending=True).indices
    ]
    scores = torch.sort(probs, descending=True).values

    # Consolidate results in dataframe
    d = {"word": words, "score": scores}
    df = pd.DataFrame(data=d)

    # Get score rank of target word
    result_rank = words.index(target_word)
    target_col = [0] * len(scores)
    target_col[result_rank] = 1
    df["target"] = target_col

    # Plot
    fig = px.bar(
        df[:100],
        x="word",
        y="score",
        color="target",
        color_continuous_scale=px.colors.sequential.Bluered,
    )
    # fig.update(layout_coloraxis_showscale=False)
    fig.show()
    return fig


gr.Interface(
    fn=plot_results,
    inputs=[
        gr.Textbox(label="词语", placeholder="标准"),
        gr.Textbox(label="造句", placeholder="小明朗读课文时发音标准，被老师评为优秀。"),
    ],
    examples=[
        ["聪明", "小明很聪明，每年考班上第一名。"],
        ["尴尬", "小明去朋友的生日庆祝会，忘了带礼物，感到很尴尬。"],
        ["标准", "小明朗读课文时发音标准，被老师评为优秀。"],
    ],
    outputs=["plot"],
    title="Chinese Sentence Grading",
).launch()