app.py

import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer
import gradio as gr
import matplotlib.pyplot as plt
import io
import re
import os
from datetime import datetime
import spaces

@spaces.GPU
def load_model():
    model_id = "oshizo/japanese-sexual-moderation-v2"
    tokenizer = AutoTokenizer.from_pretrained(model_id)
    model = AutoModelForSequenceClassification.from_pretrained(
        model_id,
        problem_type="regression"
    )
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    model = model.to(device)
    return tokenizer, model, device

@spaces.GPU
def analyze_text(text, tokenizer, model, device):
    with torch.no_grad():
        encoding = tokenizer([text], padding='max_length', truncation=True, max_length=64, return_tensors="pt")
        encoding = {k: v.to(device) for k, v in encoding.items()}
        score = model(**encoding).logits.item()
    return score

@spaces.GPU
def split_text(text, split_by='sentence'):
    if split_by == 'sentence':
        return [sent.strip() for sent in re.split('。|！|？', text) if sent.strip()]
    else:  # split by line
        return [line.strip() for line in text.split('\n') if line.strip()]

@spaces.GPU
def create_graph(texts, scores):
    fig, ax = plt.subplots(figsize=(12, 6))
    ax.bar(range(len(scores)), scores)
    ax.set_xlabel('テキスト番号')
    ax.set_ylabel('スコア')
    ax.set_title("分析結果")
    ax.set_xticks(range(len(scores)))
    ax.set_xticklabels(range(1, len(scores) + 1))
    plt.tight_layout()
    return fig

@spaces.GPU
def create_not_r18_text(texts, scores):
    not_r18_texts = []
    for text, score in zip(texts, scores):
        if score < 0.4:
            not_r18_texts.append(text)
        else:
            not_r18_texts.append('')  # 除外された行の位置に空行を挿入
    return '\n'.join(not_r18_texts)

tokenizer, model, device = load_model()

@spaces.GPU
def process_text(text, split_by):
    texts = split_text(text, split_by)
    scores = [analyze_text(t, tokenizer, model, device) for t in texts]
    
    graph = create_graph(texts, scores)
    not_r18_text = create_not_r18_text(texts, scores)
    
    result = {
        "texts": texts,
        "scores": scores,
    }
    
    return result, graph, not_r18_text

# Gradio インターフェースの定義
iface = gr.Interface(
    fn=process_text,
    inputs=[
        gr.Textbox(label="テキスト入力"),
        gr.Radio(["sentence", "line"], label="分割方法", value="sentence")
    ],
    outputs=[
        gr.JSON(label="分析結果"),
        gr.Plot(label="スコアグラフ"),
        gr.Textbox(label="R18判定除外テキスト")
    ],
    title="テキスト分析API",
    description="テキストを入力し、R18判定と分析を行います。"
)

# サーバーの起動
if __name__ == "__main__":
    iface.launch(server_name="0.0.0.0", server_port=7860)