pip install transformers torch datasets

from transformers import pipeline
# 使用情感分析流水線
classifier = pipeline("sentiment-analysis")
result = classifier("I love programming with Transformers!")
print(result)  # [{'label': 'POSITIVE', 'score': 0.9998}]

from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
text = "Hello, world!"
encoded = tokenizer(text, 
                    padding=True, 
                    truncation=True, 
                    return_tensors="pt")  # 返回PyTorch張量
print(encoded)
# {'input_ids': tensor([[101, 7592, 1010, 2088, 999, 102]]), 
#  'attention_mask': tensor([[1, 1, 1, 1, 1, 1]])}

from transformers import AutoModel
model = AutoModel.from_pretrained("bert-base-uncased")
outputs = model(**encoded)  # 前向傳播
last_hidden_states = outputs.last_hidden_state

from transformers import BertForSequenceClassification, Trainer, TrainingArguments
from datasets import load_dataset
# 加載數(shù)據(jù)集
dataset = load_dataset("imdb")
tokenized_datasets = dataset.map(
    lambda x: tokenizer(x["text"], padding=True, truncation=True),
    batched=True
)
# 定義模型
model = BertForSequenceClassification.from_pretrained("bert-base-uncased", num_labels=2)
# 訓(xùn)練參數(shù)配置
training_args = TrainingArguments(
    output_dir="./results",
    num_train_epochs=3,
    per_device_train_batch_size=8,
    evaluation_strategy="epoch"
)
# 訓(xùn)練器配置
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_datasets["train"],
    eval_dataset=tokenized_datasets["test"]
)
# 開(kāi)始訓(xùn)練
trainer.train()

model.save_pretrained("./my_model")
tokenizer.save_pretrained("./my_model")
# 加載自定義模型
new_model = AutoModel.from_pretrained("./my_model")

from transformers import BertModel, BertTokenizer
import torch
model = BertModel.from_pretrained("bert-base-uncased", output_attentions=True)
inputs = tokenizer("The cat sat on the mat", return_tensors="pt")
outputs = model(**inputs)
# 提取第0層的注意力權(quán)重
attention = outputs.attentions[0][0]
print(attention.shape)  # [num_heads, seq_len, seq_len]

from transformers import TrainingArguments
training_args = TrainingArguments(
    fp16=True,  # 啟用混合精度
    ...
)

from transformers import pipeline
# 加載NER流水線
ner_pipeline = pipeline("ner", model="dslim/bert-base-NER")
text = "Apple was founded by Steve Jobs in Cupertino."
results = ner_pipeline(text)
# 結(jié)果可視化
for entity in results:
    print(f"{entity['word']} -> {entity['entity']} (confidence: {entity['score']:.2f})")

from transformers import TrainingArguments
training_args = TrainingArguments(
    output_dir="./results",
    learning_rate=2e-5,
    weight_decay=0.01,
    warmup_steps=500,          # 學(xué)習(xí)率預(yù)熱步數(shù)
    gradient_accumulation_steps=2,  # 梯度累積（節(jié)省顯存）
    gradient_clipping=1.0,     # 梯度裁剪閾值
    ...
)

import torch
from transformers import BertForSequenceClassification
class CustomModel(BertForSequenceClassification):
    def __init__(self, config):
        super().__init__(config)
    def forward(self, input_ids, attention_mask, labels=None):
        outputs = super().forward(input_ids, attention_mask)
        logits = outputs.logits
        if labels is not None:
            loss_fct = torch.nn.CrossEntropyLoss(weight=torch.tensor([1.0, 2.0]))  # 類(lèi)別權(quán)重
            loss = loss_fct(logits.view(-1, 2), labels.view(-1))
            return {"loss": loss, "logits": logits}
        return outputs

from transformers import GPT2LMHeadModel, GPT2Tokenizer
tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
model = GPT2LMHeadModel.from_pretrained("gpt2")
prompt = "In a world where AI dominates,"
input_ids = tokenizer.encode(prompt, return_tensors="pt")
# 生成文本（配置生成參數(shù)）
output = model.generate(
    input_ids, 
    max_length=100, 
    temperature=0.7,        # 控制隨機(jī)性（低值更確定）
    top_k=50,               # 限制候選詞數(shù)量
    num_return_sequences=3  # 生成3個(gè)不同結(jié)果
)
for seq in output:
    print(tokenizer.decode(seq, skip_special_tokens=True))

from transformers import pipeline
qa_pipeline = pipeline("question-answering", model="deepset/roberta-base-squad2")
context = """
Hugging Face is a company based in New York City. 
Its Transformers library is widely used in NLP.
"""
question = "Where is Hugging Face located?"
result = qa_pipeline(question=question, context=context)
print(f"Answer: {result['answer']} (score: {result['score']:.2f})")
# Answer: New York City (score: 0.92)

from transformers import BertModel, AutoTokenizer
import torch
model = BertModel.from_pretrained("bert-base-uncased")
quantized_model = torch.quantization.quantize_dynamic(
    model, 
    {torch.nn.Linear},   # 量化所有線性層
    dtype=torch.qint8
)
# 量化后推理速度提升2-4倍，模型體積減少約75%

from transformers import BertTokenizer, BertForSequenceClassification
from torch.onnx import export
model = BertForSequenceClassification.from_pretrained("bert-base-uncased")
tokenizer = BertTokenizer.from_pretrained("bert-base-uncased")
# 示例輸入
dummy_input = tokenizer("This is a test", return_tensors="pt")
# 導(dǎo)出為ONNX
export(
    model,
    (dummy_input["input_ids"], dummy_input["attention_mask"]),
    "model.onnx",
    opset_version=13,
    input_names=["input_ids", "attention_mask"],
    output_names=["logits"],
    dynamic_axes={"input_ids": {0: "batch"}, "attention_mask": {0: "batch"}}
)

import torch
# 在訓(xùn)練循環(huán)中插入顯存監(jiān)控
print(f"Allocated: {torch.cuda.memory_allocated() / 1e9:.2f} GB")
print(f"Cached: {torch.cuda.memory_reserved() / 1e9:.2f} GB")

from torch.profiler import profile, record_function, ProfilerActivity
with profile(activities=[ProfilerActivity.CUDA], record_shapes=True) as prof:
    outputs = model(**inputs)
print(prof.key_averages().table(sort_by="cuda_time_total", row_limit=10))

from transformers import MBartForConditionalGeneration, MBart50TokenizerFast
model = MBartForConditionalGeneration.from_pretrained("facebook/mbart-large-50-many-to-many-mmt")
tokenizer = MBart50TokenizerFast.from_pretrained("facebook/mbart-large-50-many-to-many-mmt")
# 中文轉(zhuǎn)英文
tokenizer.src_lang = "zh_CN"
text = "歡迎使用Transformers庫(kù)"
encoded = tokenizer(text, return_tensors="pt")
generated_tokens = model.generate(**encoded, forced_bos_token_id=tokenizer.lang_code_to_id["en_XX"])
print(tokenizer.batch_decode(generated_tokens, skip_special_tokens=True))
# ['Welcome to the Transformers library']

from PIL import Image
from transformers import CLIPProcessor, CLIPModel
model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32")
processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
image = Image.open("cat.jpg")
text = ["a photo of a cat", "a photo of a dog"]
inputs = processor(text=text, images=image, return_tensors="pt", padding=True)
outputs = model(**inputs)
# 計(jì)算圖文相似度
logits_per_image = outputs.logits_per_image
probs = logits_per_image.softmax(dim=1)  # 概率分布

import torch.nn as nn
class TransformerBlock(nn.Module):
    def __init__(self, d_model=512, nhead=8):
        super().__init__()
        self.attention = nn.MultiheadAttention(d_model, nhead)
        self.linear = nn.Linear(d_model, d_model)
        self.norm = nn.LayerNorm(d_model)
    def forward(self, x):
        attn_output, _ = self.attention(x, x, x)
        x = x + attn_output
        x = self.norm(x)
        x = x + self.linear(x)
        return x

from transformers import BertTokenizer
tokenizer = BertTokenizer.from_pretrained("bert-base-chinese")
# 手動(dòng)添加特殊詞匯
tokenizer.add_tokens(["【特殊詞】"])
# 調(diào)整模型嵌入層
model.resize_token_embeddings(len(tokenizer)) 

from transformers import LlamaForCausalLM, LlamaTokenizer, TrainingArguments
# 加載模型和分詞器（需申請(qǐng)權(quán)限）
model = LlamaForCausalLM.from_pretrained("decapoda-research/llama-7b-hf")
tokenizer = LlamaTokenizer.from_pretrained("decapoda-research/llama-7b-hf")
# 低秩適配（LoRA）微調(diào)
from peft import get_peft_model, LoraConfig
lora_config = LoraConfig(
    r=8,  # 低秩維度
    lora_alpha=32,
    target_modules=["q_proj", "v_proj"],  # 僅微調(diào)部分模塊
    lora_dropout=0.05,
    bias="none"
)
model = get_peft_model(model, lora_config)
model.print_trainable_parameters()  # 顯示可訓(xùn)練參數(shù)占比（通常 <1%）
# 繼續(xù)配置訓(xùn)練參數(shù)...

# 使用 TRL 庫(kù)進(jìn)行 RLHF 訓(xùn)練
from trl import PPOTrainer, AutoModelForCausalLMWithValueHead
model = AutoModelForCausalLMWithValueHead.from_pretrained("gpt2")
ppo_trainer = PPOTrainer(
    model=model,
    config=training_args,
    dataset=dataset,
    tokenizer=tokenizer
)
# 定義獎(jiǎng)勵(lì)模型
for epoch in range(3):
    for batch in ppo_trainer.dataloader:
        # 生成響應(yīng)
        response_tensors = model.generate(batch["input_ids"])
        # 計(jì)算獎(jiǎng)勵(lì)（需自定義獎(jiǎng)勵(lì)函數(shù)）
        rewards = calculate_rewards(response_tensors, batch)
        # PPO 優(yōu)化步驟
        ppo_trainer.step(
            response_tensors,
            rewards,
            batch["attention_mask"]
        )

from transformers import TrainingArguments
# 配置分布式訓(xùn)練
training_args = TrainingArguments(
    per_device_train_batch_size=4,
    gradient_accumulation_steps=8,
    fp16=True,
    tpu_num_cores=8,  # 使用TPU時(shí)指定核心數(shù)
    dataloader_num_workers=4,
    deepspeed="./configs/deepspeed_config.json"  # 使用DeepSpeed優(yōu)化
)
# DeepSpeed 配置文件示例（ds_config.json）：
{
  "fp16": {
    "enabled": true
  },
  "optimizer": {
    "type": "AdamW",
    "params": {
      "lr": 3e-5
    }
  },
  "zero_optimization": {
    "stage": 3  # 啟用ZeRO-3優(yōu)化
  }
}

from fastapi import FastAPI
from pydantic import BaseModel
from transformers import pipeline
app = FastAPI()
generator = pipeline("text-generation", model="gpt2")
class Request(BaseModel):
    text: str
    max_length: int = 100
@app.post("/generate")
async def generate_text(request: Request):
    result = generator(request.text, max_length=request.max_length)
    return {"generated_text": result[0]["generated_text"]}
# 啟動(dòng)服務(wù)：uvicorn main:app --port 8000

from transformers import AutoTokenizer, AutoModelForQuestionAnswering
tokenizer = AutoTokenizer.from_pretrained("bert-large-uncased-whole-word-masking-finetuned-squad")
model = AutoModelForQuestionAnswering.from_pretrained("bert-large-uncased-whole-word-masking-finetuned-squad")
def process_long_text(context, question, max_length=384, stride=128):
    # 分塊處理長(zhǎng)文本
    inputs = tokenizer(
        question,
        context,
        max_length=max_length,
        truncation="only_second",
        stride=stride,
        return_overflowing_tokens=True,
        return_offsets_mapping=True
    )
    # 對(duì)各塊推理并合并結(jié)果
    best_score = 0
    best_answer = ""
    for i in range(len(inputs["input_ids"])):
        outputs = model(**{k: torch.tensor([v[i]]) for k, v in inputs.items()})
        answer_start = torch.argmax(outputs.start_logits)
        answer_end = torch.argmax(outputs.end_logits) + 1
        score = (outputs.start_logits[answer_start] + outputs.end_logits[answer_end-1]).item()
        if score > best_score:
            best_score = score
            best_answer = tokenizer.decode(inputs["input_ids"][i][answer_start:answer_end])
    return best_answer

# 使用 XLM-RoBERTa 進(jìn)行跨語(yǔ)言遷移
from transformers import XLMRobertaTokenizer, XLMRobertaForSequenceClassification
tokenizer = XLMRobertaTokenizer.from_pretrained("xlm-roberta-base")
model = XLMRobertaForSequenceClassification.from_pretrained("xlm-roberta-base")
# 通過(guò)少量樣本微調(diào)（代碼與BERT訓(xùn)練類(lèi)似）

from captum.attr import LayerIntegratedGradients
from transformers import BertForSequenceClassification
model = BertForSequenceClassification.from_pretrained("bert-base-uncased")
def forward_func(input_ids, attention_mask):
    return model(input_ids, attention_mask).logits
lig = LayerIntegratedGradients(forward_func, model.bert.embeddings)
# 計(jì)算輸入詞重要性
attributions, delta = lig.attribute(
    inputs=input_ids,
    baselines=tokenizer.pad_token_id * torch.ones_like(input_ids),
    additional_forward_args=attention_mask,
    return_convergence_delta=True
)
# 可視化結(jié)果
import matplotlib.pyplot as plt
plt.bar(range(len(attributions[0])), attributions[0].detach().numpy())
plt.xticks(ticks=range(len(tokens)), labels=tokens, rotation=90)
plt.show()

import spacy
from spacy_transformers import TransformersLanguage, TransformersWordPiecer
# 創(chuàng)建spacy管道
nlp = TransformersLanguage(trf_name="bert-base-uncased")
# 自定義組件
@spacy.registry.architectures("CustomClassifier.v1")
def create_classifier(transformer, tok2vec, n_classes):
    return TransformersTextCategorizer(transformer, tok2vec, n_classes)
# 在spacy中直接使用Transformer模型
doc = nlp("This is a text to analyze.")
print(doc._.trf_last_hidden_state.shape)  # [seq_len, hidden_dim]

import gradio as gr
from transformers import pipeline
ner_pipeline = pipeline("ner")
def extract_entities(text):
    results = ner_pipeline(text)
    return {"text": text, "entities": [
        {"entity": res["entity"], "start": res["start"], "end": res["end"]}
        for res in results
    ]}
gr.Interface(
    fn=extract_entities,
    inputs=gr.Textbox(lines=5),
    outputs=gr.HighlightedText()
).launch()

# 使用 nn_pruning 進(jìn)行結(jié)構(gòu)化剪枝
from transformers import BertForSequenceClassification
from nn_pruning import ModelPruning
model = BertForSequenceClassification.from_pretrained("bert-base-uncased")
pruner = ModelPruning(
    model,
    target_sparsity=0.5,  # 剪枝50%的注意力頭
    pattern="block_sparse"  # 結(jié)構(gòu)化剪枝模式
)
# 執(zhí)行剪枝并微調(diào)
pruned_model = pruner.prune()
pruned_model.save_pretrained("./pruned_bert")
# 知識(shí)蒸餾（教師→學(xué)生模型）
from transformers import DistilBertForSequenceClassification, DistilBertTokenizer
teacher = BertForSequenceClassification.from_pretrained("bert-base-uncased")
student = DistilBertForSequenceClassification.from_pretrained("distilbert-base-uncased")
# 使用蒸餾訓(xùn)練器
from transformers import DistillationTrainingArguments, DistillationTrainer
training_args = DistillationTrainingArguments(
    output_dir="./distilled",
    temperature=2.0,  # 軟化概率分布
    alpha_ce=0.5,     # 交叉熵?fù)p失權(quán)重
    alpha_mse=0.5     # 隱藏層MSE損失權(quán)重
)
trainer = DistillationTrainer(
    teacher=teacher,
    student=student,
    args=training_args,
    train_dataset=tokenized_datasets["train"],
    tokenizer=tokenizer
)
trainer.train()

# 轉(zhuǎn)換模型為T(mén)ensorRT引擎
trtexec --onnx=model.onnx --saveEngine=model.trt --fp16

# Python 調(diào)用TensorRT引擎
import tensorrt as trt
import pycuda.driver as cuda
runtime = trt.Runtime(trt.Logger(trt.Logger.WARNING))
with open("model.trt", "rb") as f:
    engine = runtime.deserialize_cuda_engine(f.read())
context = engine.create_execution_context()
# 綁定輸入輸出緩沖區(qū)進(jìn)行推理

from transformers import AutoTokenizer, AutoModelForTokenClassification
tokenizer = AutoTokenizer.from_pretrained("dmis-lab/biobert-v1.1")
model = AutoModelForTokenClassification.from_pretrained("dmis-lab/biobert-v1.1")
text = "The patient exhibited EGFR mutations and responded to osimertinib."
inputs = tokenizer(text, return_tensors="pt")
outputs = model(**inputs).logits
# 提取基因?qū)嶓w
predictions = torch.argmax(outputs, dim=2)
print([tokenizer.decode([token]) for token in inputs.input_ids[0]])
print(predictions.tolist())  # BIO標(biāo)注結(jié)果

# 合同條款分類(lèi)
from transformers import BertTokenizer, BertForSequenceClassification
tokenizer = BertTokenizer.from_pretrained("nlpaueb/legal-bert-base-uncased")
model = BertForSequenceClassification.from_pretrained("nlpaueb/legal-bert-base-uncased")
clause = "The Parties hereby agree to arbitrate all disputes in accordance with ICC rules."
inputs = tokenizer(clause, return_tensors="pt", truncation=True, padding=True)
outputs = model(**inputs)
predicted_class = torch.argmax(outputs.logits).item()  # 0: 仲裁條款, 1: 保密條款等

from transformers import BertForSequenceClassification
import coremltools as ct
model = BertForSequenceClassification.from_pretrained("bert-base-uncased")
tokenizer = BertTokenizer.from_pretrained("bert-base-uncased")
# 轉(zhuǎn)換模型
traced_model = torch.jit.trace(model, (input_ids, attention_mask))
mlmodel = ct.convert(
    traced_model,
    inputs=[
        ct.TensorType(name="input_ids", shape=input_ids.shape),
        ct.TensorType(name="attention_mask", shape=attention_mask.shape)
    ]
)
mlmodel.save("BertSenti.mlmodel")

from transformers import TFBertForSequenceClassification
import tensorflow as tf
model = TFBertForSequenceClassification.from_pretrained("bert-base-uncased")
# 轉(zhuǎn)換為T(mén)FLite
converter = tf.lite.TFLiteConverter.from_keras_model(model)
converter.optimizations = [tf.lite.Optimize.DEFAULT]  # 動(dòng)態(tài)范圍量化
tflite_model = converter.convert()
with open("model_quant.tflite", "wb") as f:
    f.write(tflite_model)

from transformers import pipeline
from fairness_metrics import demographic_parity
# 檢測(cè)模型偏見(jiàn)
classifier = pipeline("text-classification", model="bert-base-uncased")
protected_groups = {
    "gender": ["she", "he"],
    "race": ["African", "European"]
}
bias_scores = {}
for category, terms in protected_groups.items():
    texts = [f"{term} is qualified for this position" for term in terms]
    results = classifier(texts)
    bias_scores[category] = demographic_parity(results)

from textattack import AttackRecipe
from textattack.models.wrappers import HuggingFaceModelWrapper
model_wrapper = HuggingFaceModelWrapper(model, tokenizer)
attack = AttackRecipe.build("bae")  # BAE攻擊方法
# 生成對(duì)抗樣本
attack_args = textattack.AttackArgs(num_examples=5)
attacker = textattack.Attacker(attack, model_wrapper, attack_args)
attack_results = attacker.attack_dataset(dataset)

from transformers import LongformerModel
model = LongformerModel.from_pretrained("allenai/longformer-base-4096")
inputs = tokenizer("This is a very long document..."*1000, return_tensors="pt")
outputs = model(**inputs)  # 支持最長(zhǎng)4096 tokens

# 使用Switch Transformers
from transformers import SwitchTransformersForConditionalGeneration
model = SwitchTransformersForConditionalGeneration.from_pretrained("google/switch-base-8")
outputs = model.generate(
    input_ids,
    expert_choice_mask=True,  # 追蹤專(zhuān)家路由
)
print(outputs.expert_choices)  # 顯示每個(gè)token使用的專(zhuān)家

"""
端到端文本分類(lèi)系統(tǒng)架構(gòu)：
1. 數(shù)據(jù)采集 → 2. 清洗 → 3. 標(biāo)注 → 4. 模型訓(xùn)練 → 5. 評(píng)估 → 6. 部署 → 7. 監(jiān)控
"""
# 步驟4的增強(qiáng)訓(xùn)練流程
from transformers import TrainerCallback
class CustomCallback(TrainerCallback):
    def on_log(self, args, state, control, logs=None, **kwargs):
        # 實(shí)時(shí)記錄指標(biāo)到Prometheus
        prometheus_logger.log_metrics(logs)
# 步驟7的漂移檢測(cè)
from alibi_detect.cd import MMDDrift
detector = MMDDrift(
    X_train, 
    backend="tensorflow", 
    p_val=0.05
)
drift_preds = detector.predict(X_prod)