create_model.py

#!/usr/bin/env python3
"""
中文情感分析模型訓練腳本
這個腳本展示如何創建一個可推理的 Hugging Face 模型
"""

import torch
import pandas as pd
from transformers import (
    BertTokenizer, BertForSequenceClassification,
    TrainingArguments, Trainer, pipeline
)
from torch.utils.data import Dataset
import numpy as np
from sklearn.metrics import accuracy_score

class SentimentDataset(Dataset):
    def __init__(self, texts, labels, tokenizer, max_length=128):
        self.texts = texts
        self.labels = labels
        self.tokenizer = tokenizer
        self.max_length = max_length

    def __len__(self):
        return len(self.texts)

    def __getitem__(self, idx):
        text = str(self.texts[idx])
        label = self.labels[idx]
        
        encoding = self.tokenizer(
            text,
            truncation=True,
            padding='max_length',
            max_length=self.max_length,
            return_tensors='pt'
        )
        
        return {
            'input_ids': encoding['input_ids'].flatten(),
            'attention_mask': encoding['attention_mask'].flatten(),
            'labels': torch.tensor(label, dtype=torch.long)
        }

def create_demo_model():
    """創建一個演示用的情感分析模型"""
    
    # 使用預訓練的中文 BERT
    model_name = "bert-base-chinese"
    tokenizer = BertTokenizer.from_pretrained(model_name)
    model = BertForSequenceClassification.from_pretrained(
        model_name, 
        num_labels=2,
        id2label={0: "NEGATIVE", 1: "POSITIVE"},
        label2id={"NEGATIVE": 0, "POSITIVE": 1}
    )
    
    # 創建一些示例數據
    texts = [
        "這個產品真的很棒！我非常滿意。",
        "質量很差，完全不推薦。",
        "服務態度很好，值得信賴。",
        "價格太貴了，性價比不高。",
        "非常棒的體驗，會再次購買。",
        "完全浪費錢，後悔購買。"
    ]
    labels = [1, 0, 1, 0, 1, 0]  # 1: POSITIVE, 0: NEGATIVE
    
    # 準備數據集
    dataset = SentimentDataset(texts, labels, tokenizer)
    
    # 訓練參數（這裡只做演示，實際訓練需要更多數據）
    training_args = TrainingArguments(
        output_dir='./results',
        num_train_epochs=1,
        per_device_train_batch_size=2,
        per_device_eval_batch_size=2,
        warmup_steps=10,
        weight_decay=0.01,
        logging_dir='./logs',
        save_strategy="no",  # 不保存中間檢查點
    )
    
    # 創建訓練器
    trainer = Trainer(
        model=model,
        args=training_args,
        train_dataset=dataset,
        tokenizer=tokenizer,
    )
    
    # 簡單訓練（演示用）
    trainer.train()
    
    return model, tokenizer

def save_model_for_huggingface(model, tokenizer, save_directory):
    """保存模型為 Hugging Face 格式"""
    
    # 保存模型和 tokenizer
    model.save_pretrained(save_directory)
    tokenizer.save_pretrained(save_directory)
    
    print(f"模型已保存到: {save_directory}")
    print("包含的檔案:")
    import os
    for file in os.listdir(save_directory):
        print(f"  - {file}")

def test_inference(model_directory):
    """測試模型推理功能"""
    
    # 創建推理 pipeline
    classifier = pipeline(
        "text-classification",
        model=model_directory,
        tokenizer=model_directory,
        return_all_scores=True
    )
    
    # 測試樣本
    test_texts = [
        "這個手機真的很好用！",
        "服務態度太差了。",
        "質量不錯，值得推薦。"
    ]
    
    print("\\n=== 模型推理測試 ===")
    for text in test_texts:
        result = classifier(text)
        print(f"文本: {text}")
        print(f"結果: {result}")
        print("-" * 50)

if __name__ == "__main__":
    print("開始創建中文情感分析模型...")
    
    # 創建並訓練模型
    model, tokenizer = create_demo_model()
    
    # 保存模型
    save_directory = "./my-sentiment-model"
    save_model_for_huggingface(model, tokenizer, save_directory)
    
    # 測試推理
    test_inference(save_directory)
    
    print("\\n✅ 模型創建完成！")
    print("現在你可以:")
    print("1. 將模型檔案推送到 Hugging Face")
    print("2. 讓其他人使用 transformers 載入你的模型")
    print("3. 使用 Inference API 進行線上推理")