【NLP】32. Transformers （HuggingFace Pipelines 实战）

🤖 Transformers （HuggingFace Pipelines 实战）

本教程基于 Hugging Face 的 transformers 库，展示如何使用预训练模型完成以下任务：

• 情感分析（Sentiment Analysis）
• 文本生成（Text Generation）
• 翻译（Translation）
• 掩码填空（Masked Language Modeling）
• 零样本分类（Zero-shot Classification）
• 特定任务模型推理（Text2Text Generation, e.g., T5）
• 文本摘要（Summarization）
• 自定义分类模型载入与推理（如 BERT）

---

📦 安装依赖

!pip install datasets evaluate transformers sentencepiece -q

---

✅ 情感分析

from transformers import pipeline

classifier = pipeline("sentiment-analysis")

# 单个句子
classifier("I really enjoy learning new things every day.")

# 多个句子
classifier([
    "I really enjoy learning new things every day.",
    "I dislike rainy weather on weekends."
])

---

✍️ 文本生成（默认模型）

from transformers import pipeline

generator = pipeline("text-generation")
generator("Today we are going to explore something exciting")

---

🎯 文本生成（指定参数）

# 两句话，每句最多15词
generator("Today we are going to explore something exciting", num_return_sequences=2, max_length=15)

---

⚙️ 使用 distilgpt2 模型生成文本

from transformers import pipeline

generator = pipeline("text-generation", model="distilgpt2")
generator(
    "Today we are going to explore something exciting",
    max_length=30,
    num_return_sequences=2,
)

---

🌍 翻译（英语→德语）

from transformers import pipeline

translator = pipeline("translation_en_to_de")
translator("This is a great day to learn something new!")

---

🧩 掩码填空任务（填词）

from transformers import pipeline

unmasker = pipeline("fill-mask")
unmasker("Life is like a <mask> of chocolates.")

---

🧠 零样本分类（Zero-shot Classification）

from transformers import pipeline

classifier = pipeline("zero-shot-classification")
classifier(
    "This tutorial is about machine learning and natural language processing.",
    candidate_labels=["education", "sports", "politics"]
)

---

🔁 T5 模型（Text2Text）

from transformers import pipeline

text2text = pipeline("text2text-generation")
text2text("Translate English to French: How are you today?")

---

✂️ 文本摘要（Summarization）

from transformers import pipeline

summarizer = pipeline("summarization")
summarizer(
    "Machine learning is a field of artificial intelligence that focuses on enabling machines to learn from data..."
)

---

🧪 使用自己的模型（以 BERT 为例）

from transformers import BertTokenizer, BertForSequenceClassification
from transformers import pipeline

model_name = "nlptown/bert-base-multilingual-uncased-sentiment"
tokenizer = BertTokenizer.from_pretrained(model_name)
model = BertForSequenceClassification.from_pretrained(model_name)

pipe = pipeline("sentiment-analysis", model=model, tokenizer=tokenizer)
pipe("I had an amazing experience with this product!")