當前位置：首頁 > 编程资源 > 编程问答 >内容正文

编程问答

深入Bert实战(Pytorch)----fine-Tuning 2

發布時間：2025/3/21 编程问答 33 豆豆

生活随笔收集整理的這篇文章主要介紹了深入Bert实战(Pytorch)----fine-Tuning 2 小編覺得挺不錯的,現在分享給大家,幫大家做個參考.

?

深入Bert實戰(Pytorch)----fine-Tuning 2

https://www.bilibili.com/video/BV1K5411t7MD?p=5
https://www.youtube.com/channel/UCoRX98PLOsaN8PtekB9kWrw/videos
深入BERT實戰(PyTorch) by ChrisMcCormickAI
這是ChrisMcCormickAI在油管bert，8集系列第三篇fine-Tuning的pytorch的講解的代碼，在油管視頻下有cloab地址，如果不能翻墻的可以留下郵箱我全部看完整理后發給你。但是在fine-tuning最好還是在cloab上運行

文章目錄

深入Bert實戰(Pytorch)----fine-Tuning 2
4. Train Our Classification Model
- 4.1. BertForSequenceClassification
- 4.2. Optimizer & Learning Rate Scheduler
- 4.3. 循環訓練
5. 在測試集上的表現
- 5.1. 數據準備
- 5.2. 在測試集上評估
總結
附錄
- A1. Saving & Loading Fine-Tuned Model
Revision History

4. Train Our Classification Model

4.1. BertForSequenceClassification

對于這個任務，我們首先要修改預訓練的BERT模型以給出分類輸出，然后在自己的數據集上繼續訓練模型，直到整個模型(端到端的模型)非常適合自己的任務。

值得慶幸的是，huggingface pytorch實現包含一組為各種NLP任務設計的接口。盡管這些接口都建立在訓練好的BERT模型之上，但每個接口都有不同的頂層和輸出類型，以適應它們特定的NLP任務。

這里是目前提供的fine-tuning列表

BertModel
BertForPreTraining
BertForMaskedLM
BertForNextSentencePrediction
BertForSequenceClassification?- The one we’ll use.
BertForTokenClassification
BertForQuestionAnswering

這里是transformer的文檔here.

我們使用BertForSequenceClassification。這是普通的BERT模型，上面添加了一個用于分類的線性層，我們將使用它作為句子分類器。當我們輸入數據時，整個預訓練的BERT模型和額外的未訓練的分類層是同時在這個任務上進行訓練

好的，現在加載BERT！這里有幾種不同的預訓練模型，"bert-base-uncased"版本，僅有小寫字母(“uncased”)相比于是較小的(“base” vs “large”)。

預訓練的文檔在from_pretrainedhere?定義了其它參數?here

from transformers import BertForSequenceClassification, AdamW, BertConfig# Load BertForSequenceClassification, the pretrained BERT model with a single # linear classification layer on top. # 加載BertForSequenceClassification，預訓練的模型+頂層單層線性分類層 model = BertForSequenceClassification.from_pretrained("bert-base-uncased", # Use the 12-layer BERT model, with an uncased vocab.num_labels = 2, # The number of output labels--2 for binary classification.# You can increase this for multi-class tasks.# 2分類問題，可以增加為多分類問題output_attentions = False, # Whether the model returns attentions weights.output_hidden_states = False, # Whether the model returns all hidden-states. )# Tell pytorch to run this model on the GPU. model.cuda()

出于好奇，我們可以在這里按名稱瀏覽所有的模型參數。
在下面的單元格中，我打印出了以下權重的名稱和尺寸:

這里作者打印了所有層，總共有201層，也打印了權重和大小

The embedding layer.

The first of the twelve transformers.

The output layer.

# Get all of the model's parameters as a list of tuples. params = list(model.named_parameters())print('The BERT model has {:} different named parameters.\n'.format(len(params)))print('==== Embedding Layer ====\n')for p in params[0:5]:print("{:<55} {:>12}".format(p[0], str(tuple(p[1].size()))))print('\n==== First Transformer ====\n')for p in params[5:21]:print("{:<55} {:>12}".format(p[0], str(tuple(p[1].size()))))print('\n==== Output Layer ====\n')for p in params[-4:]:print("{:<55} {:>12}".format(p[0], str(tuple(p[1].size()))))

The BERT model has 201 different named parameters.==== Embedding Layer ====bert.embeddings.word_embeddings.weight (30522, 768) bert.embeddings.position_embeddings.weight (512, 768) bert.embeddings.token_type_embeddings.weight (2, 768) bert.embeddings.LayerNorm.weight (768,) bert.embeddings.LayerNorm.bias (768,)==== First Transformer ====bert.encoder.layer.0.attention.self.query.weight (768, 768) bert.encoder.layer.0.attention.self.query.bias (768,) bert.encoder.layer.0.attention.self.key.weight (768, 768) bert.encoder.layer.0.attention.self.key.bias (768,) bert.encoder.layer.0.attention.self.value.weight (768, 768) bert.encoder.layer.0.attention.self.value.bias (768,) bert.encoder.layer.0.attention.output.dense.weight (768, 768) bert.encoder.layer.0.attention.output.dense.bias (768,) bert.encoder.layer.0.attention.output.LayerNorm.weight (768,) bert.encoder.layer.0.attention.output.LayerNorm.bias (768,) bert.encoder.layer.0.intermediate.dense.weight (3072, 768) bert.encoder.layer.0.intermediate.dense.bias (3072,) bert.encoder.layer.0.output.dense.weight (768, 3072) bert.encoder.layer.0.output.dense.bias (768,) bert.encoder.layer.0.output.LayerNorm.weight (768,) bert.encoder.layer.0.output.LayerNorm.bias (768,)==== Output Layer ====bert.pooler.dense.weight (768, 768) bert.pooler.dense.bias (768,) classifier.weight (2, 768) classifier.bias (2,)

4.2. Optimizer & Learning Rate Scheduler

現在我們已經加載了模型，我們需要從存儲的模型中獲取訓練超參數。

為了進行微調，作者建議從以下值中進行選擇。(從論文的注釋?BERT paper):

Batch size:?16, 32

Learning rate (Adam):?5e-5, 3e-5, 2e-5
Number of epochs:?2, 3, 4

作者選擇的參數是：

Batch size: 32 (set when creating our DataLoaders)
Learning rate: 2e-5
Epochs: 4 (we’ll see that this is probably too many…)

參數eps = 1e-8?是"a very small number to prevent any division by zero in the implementation"(from?here)

您可以在run_glue.py中找到創建AdamW優化器的方法here.

# Note: AdamW is a class from the huggingface library (as opposed to pytorch) # AdamW是huggingface實現的類 # I believe the 'W' stands for 'Weight Decay fix" optimizer = AdamW(model.parameters(),lr = 2e-5, # args.learning_rate - default is 5e-5, our notebook had 2e-5eps = 1e-8 # args.adam_epsilon - default is 1e-8.)

from transformers import get_linear_schedule_with_warmup# Number of training epochs. The BERT authors recommend between 2 and 4. # We chose to run for 4, but we'll see later that this may be over-fitting the # training data. epochs = 4# Total number of training steps is [number of batches] x [number of epochs]. # (Note that this is not the same as the number of training samples). total_steps = len(train_dataloader) * epochs # 總共4 * 241批# Create the learning rate scheduler. scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps = 0, # Default value in run_glue.pynum_training_steps = total_steps)

4.3. 循環訓練

下面是我們的訓練循環。有很多事情要做，但從根本上來說，對于循環中的每一個過程，我們都有一個training階段和一個validation階段。

Thank you to?Stas Bekman?for contributing the insights and code for using validation loss to detect over-fitting!

Training:

打開我們的數據 inputs 和 labels
加載數據到GPU上
清除之前計算的梯度。
- 在pytorch中，除非顯式清除梯度，否則梯度默認累積(對于rnn之類的東西很有用)。
Forward pass(通過網絡輸入數據)
Backward pass 反向傳播
告訴網絡使用optimizer.step()更新參數
監控進度，跟蹤變量

Evalution:

同訓練過程一樣，打開inputs 和 labels
加載數據到GPU上
Forward pass(通過網絡輸入數據)
計算我們驗證數據的損失，監控進度，跟蹤變量

Pytorch向我們隱藏了所有詳細的計算，但是我們已經對代碼進行了注釋，指出了每一行上發生的上述步驟。

定義一個計算精度的輔助函數。

import numpy as np# Function to calculate the accuracy of our predictions vs labels # 這個函數來計算預測值和labels的準確度 def flat_accuracy(preds, labels):pred_flat = np.argmax(preds, axis=1).flatten() # 取出最大值對應的索引labels_flat = labels.flatten()return np.sum(pred_flat == labels_flat) / len(labels_flat)

格式化函數時間

import time import datetimedef format_time(elapsed):'''Takes a time in seconds and returns a string hh:mm:ss'''# Round to the nearest second. 四舍五入elapsed_rounded = int(round((elapsed)))# Format as hh:mm:ssreturn str(datetime.timedelta(seconds=elapsed_rounded))

現在開始訓練，這里要修改一部分代碼，作者給的代碼有個地方要做修改,參考run_glue.py

import random import numpy as np# This training code is based on the `run_glue.py` script here: # https://github.com/huggingface/transformers/blob/5bfcd0485ece086ebcbed2d008813037968a9e58/examples/run_glue.py#L128# Set the seed value all over the place to make this reproducible. 保證可重復性 seed_val = 42random.seed(seed_val) np.random.seed(seed_val) torch.manual_seed(seed_val) torch.cuda.manual_seed_all(seed_val)# We'll store a number of quantities（保存如） such as training and validation loss, # validation accuracy, and timings.(訓練loss, 驗證loss, 驗證準確率，訓練時間) training_stats = []# Measure the total training time for the whole run. 總訓練時間 total_t0 = time.time()# For each epoch... for epoch_i in range(0, epochs):# ========================================# Training# ========================================# 對訓練集進行一次完整的測試。print("")print('======== Epoch {:} / {:} ========'.format(epoch_i + 1, epochs))print('Training...')# Measure how long the training epoch takes.t0 = time.time()# Reset the total loss for this epoch.total_train_loss = 0# Put the model into training mode. Don't be mislead--the call to # `train` just changes the *mode*, it doesn't *perform* the training.# 這里并不是執行的訓練，而是，實例化啟用 BatchNormalization 和 Dropout# `dropout` and `batchnorm` layers behave differently during training# vs. test (source: https://stackoverflow.com/questions/51433378/what-does-model-train-do-in-pytorch)model.train()# For each batch of training data...for step, batch in enumerate(train_dataloader): # 共241個batches# Progress update every 40 batches. 40步打印一次if step % 40 == 0 and not step == 0:# Calculate elapsed time in minutes.elapsed = format_time(time.time() - t0)# Report progress.print(' Batch {:>5,} of {:>5,}. Elapsed: {:}.'.format(step, len(train_dataloader), elapsed))# 例: Batch 40 of 241. Elapsed: 0:00:08.# `batch` contains three pytorch tensors:# [0]: input ids # [1]: attention masks# [2]: labels # 第一步的打開數據, 第二步將數據放到GPU `to`方法b_input_ids = batch[0].to(device)b_input_mask = batch[1].to(device)b_labels = batch[2].to(device)# 在執行 backward pass 之前，始終清除任何先前計算的梯度。# PyTorch不會自動這樣做，因為累積梯度“在訓練rnn時很方便”。# (source: https://stackoverflow.com/questions/48001598/why-do-we-need-to-call-zero-grad-in-pytorch)model.zero_grad() # 第三步，梯度清零 # 執行 forward pass (在此訓練批次上對模型進行評估).# The documentation for this `model` function is here: # https://huggingface.co/transformers/v2.2.0/model_doc/bert.html#transformers.BertForSequenceClassification# 它根據給定的參數和設置的標志返回不同數量的形參。# it returns the loss (because we provided labels) and the "logits"--the model outputs prior to activation.# 返回loss和"logits"--激活之前的模型輸出。 model = BertForSequenceClassificationoutput = model(b_input_ids, token_type_ids=None, attention_mask=b_input_mask, labels=b_labels)# 將所有批次的訓練損失累積起來，這樣我們就可以在最后計算平均損失。 # `loss` 是一個單個值的tensor; the `.item()` 函數將它轉為一個python numberloss, logits = output[:2]total_train_loss += loss.item()# 執行反向傳播計算精度.loss.backward()# Clip the norm of the gradients to 1.0.# 梯度裁剪，防止梯度爆炸torch.nn.utils.clip_grad_norm_(model.parameters(), 1.0)# Update parameters and take a step using the computed gradient.# 更新參數，計算梯度# 優化器規定“update rule”——參數如何根據梯度、學習速率等進行修改。optimizer.step()# 更新學習率scheduler.step()# 計算平均lossavg_train_loss = total_train_loss / len(train_dataloader) # 訓練時間training_time = format_time(time.time() - t0)# 打印結果print("")print(" Average training loss: {0:.2f}".format(avg_train_loss))print(" Training epcoh took: {:}".format(training_time))# ========================================# Validation# ========================================# 在驗證集查看print("")print("Running Validation...")t0 = time.time()# 將模型置于評估模式不使用BatchNormalization()和Dropout()model.eval()# 跟蹤變量total_eval_accuracy = 0total_eval_loss = 0nb_eval_steps = 0# 在每個epoch上評估for batch in validation_dataloader:# `batch` contains three pytorch tensors:# [0]: input ids # [1]: attention masks# [2]: labels b_input_ids = batch[0].to(device)b_input_mask = batch[1].to(device)b_labels = batch[2].to(device)# Tell pytorch not to bother with constructing the compute graph during# the forward pass, since this is only needed for backprop (training).with torch.no_grad(): # Forward pass, calculate logit predictions.# token_type_ids is the same as the "segment ids", which # differentiates sentence 1 and 2 in 2-sentence tasks.# The documentation for this `model` function is here: # https://huggingface.co/transformers/v2.2.0/model_doc/bert.html#transformers.BertForSequenceClassification# Get the "logits" output by the model. The "logits" are the output# values prior to applying an activation function like the softmax.(loss, logits) = model(b_input_ids, token_type_ids=None, attention_mask=b_input_mask,labels=b_labels)# 計算驗證損失loss, logits = output[:2]total_eval_loss += loss.item()# Move logits and labels to CPUlogits = logits.detach().cpu().numpy()label_ids = b_labels.to('cpu').numpy()# Calculate the accuracy for this batch of test sentences, and# accumulate it over all batches.total_eval_accuracy += flat_accuracy(logits, label_ids)# 返回驗證結果avg_val_accuracy = total_eval_accuracy / len(validation_dataloader)print(" Accuracy: {0:.2f}".format(avg_val_accuracy))# 計算平均復雜度avg_val_loss = total_eval_loss / len(validation_dataloader)# 時間validation_time = format_time(time.time() - t0)print(" Validation Loss: {0:.2f}".format(avg_val_loss))print(" Validation took: {:}".format(validation_time))# 記錄這個epoch的所有統計數據。方便后面可視化training_stats.append({'epoch': epoch_i + 1,'Training Loss': avg_train_loss,'Valid. Loss': avg_val_loss,'Valid. Accur.': avg_val_accuracy,'Training Time': training_time,'Validation Time': validation_time})print("") print("Training complete!")print("Total training took {:} (h:mm:ss)".format(format_time(time.time()-total_t0)))

讓我們來看看訓練過程的總結。

import pandas as pd# 顯示浮點數小數點后兩位。 pd.set_option('precision', 2)# 從訓練統計數據里，創建一個 DataFrame df_stats = pd.DataFrame(data=training_stats)# 用'epoch'行坐標 df_stats = df_stats.set_index('epoch')# A hack to force the column headers to wrap. #df = df.style.set_table_styles([dict(selector="th",props=[('max-width', '70px')])])# Display the table. df_stats

Training LossValid. LossValid. Accur.Training TimeValidation Time epoch

1	0.50	0.45	0.80	0:00:51
2	0.32	0.46	0.81	0:00:51
3	0.22	0.49	0.82	0:00:51
4	0.16	0.55	0.82	0:00:51

這里我跑這代碼train loss沒有下降，反而上升了，有了解這個問題的大大，麻煩請留言指教下

請注意，當訓練損失隨著時間的推移而下降時，驗證損失卻在增加!這表明我們訓練模型的時間太長了，它對訓練數據過于擬合。

(作為參考，我們使用了7,695個訓練樣本和856個驗證樣本)。

驗證損失是比精度更精確的度量，因為有了精度，我們不關心確切的輸出值，而只關心它落在閾值的哪一邊。

如果我們預測的是正確的答案，但缺乏信心，那么驗證損失將捕捉到這一點，而準確性則不會。

import matplotlib.pyplot as plt % matplotlib inlineimport seaborn as sns# Use plot styling from seaborn. sns.set(style='darkgrid')# Increase the plot size and font size. sns.set(font_scale=1.5) plt.rcParams["figure.figsize"] = (12,6)# 繪制學習曲線 plt.plot(df_stats['Training Loss'], 'b-o', label="Training") plt.plot(df_stats['Valid. Loss'], 'g-o', label="Validation")# Label the plot. plt.title("Training & Validation Loss") plt.xlabel("Epoch") plt.ylabel("Loss") plt.legend() plt.xticks([1, 2, 3, 4])plt.show()

5. 在測試集上的表現

現在，我們將加載holdout數據集并準備輸入，就像我們對訓練集所做的那樣。然后，我們將使用Matthew’s correlation coefficient評估預測，因為這是更廣泛的NLP社區用于評估CoLA性能的指標。在這個指標下，+1是最好的分數，-1是最差的分數。通過這種方式，我們可以看到針對這個特定任務的先進模型的性能如何。

5.1. 數據準備

我們需要應用與訓練數據相同的所有步驟來準備測試數據集。

import pandas as pd# 加載數據 df = pd.read_csv("./cola_public/raw/out_of_domain_dev.tsv", delimiter='\t', header=None, names=['sentence_source', 'label', 'label_notes', 'sentence'])# 顯示句子數量 print('Number of test sentences: {:,}\n'.format(df.shape[0]))# 創建句子和標簽列表 sentences = df.sentence.values labels = df.label.values# Tokenize input_ids = [] attention_masks = []# For every sentence... for sent in sentences:# `encode_plus` will:# (1) Tokenize the sentence.# (2) 添加 `[CLS]` token 到開始# (3) 添加 `[SEP]` token 到結束# (4) 映射tokens 到 IDs.# (5) 填充或截斷句子到`max_length`# (6) Create attention masks for [PAD] tokens.encoded_dict = tokenizer.encode_plus(sent, # 對句子做encode.add_special_tokens = True, # Add '[CLS]' and '[SEP]'max_length = 64, # Pad & truncate all sentences.pad_to_max_length = True,return_attention_mask = True, # Construct attn. masks.return_tensors = 'pt', # Return pytorch tensors.)# 將已編碼的句子添加到列表中。 input_ids.append(encoded_dict['input_ids'])# 以及它的注意力掩碼(簡單地區分填充和非填充)。attention_masks.append(encoded_dict['attention_mask'])# Convert the lists into tensors. input_ids = torch.cat(input_ids, dim=0) attention_masks = torch.cat(attention_masks, dim=0) labels = torch.tensor(labels)# Set the batch size. batch_size = 32 # Create the DataLoader. prediction_data = TensorDataset(input_ids, attention_masks, labels) prediction_sampler = SequentialSampler(prediction_data) prediction_dataloader = DataLoader(prediction_data, sampler=prediction_sampler, batch_size=batch_size)

Number of test sentences: 516

5.2. 在測試集上評估

準備好測試集之后，我們可以應用我們的微調模型來生成測試集的預測。

# Prediction on test setprint('Predicting labels for {:,} test sentences...'.format(len(input_ids)))# 在測試模型 model.eval()# 跟蹤變量 predictions , true_labels = [], []# Predict for batch in prediction_dataloader:# Add batch to GPUbatch = tuple(t.to(device) for t in batch)# Unpack the inputs from our dataloaderb_input_ids, b_input_mask, b_labels = batch# 不讓模型計算或存儲梯度，節省內存和加速預測with torch.no_grad():# Forward pass, calculate logit predictionsoutputs = model(b_input_ids, token_type_ids=None, attention_mask=b_input_mask)logits = outputs[0]# Move logits and labels to CPUlogits = logits.detach().cpu().numpy()label_ids = b_labels.to('cpu').numpy()# Store predictions and true labelspredictions.append(logits)true_labels.append(label_ids)print(' DONE.')

CoLA基準的精度是用“Matthews correlation coefficient”來測量的。(MCC)。

我們在這里使用MCC是因為類是不平衡的:

print('Positive samples: %d of %d (%.2f%%)' % (df.label.sum(), len(df.label), (df.label.sum() / len(df.label) * 100.0)))

Positive samples: 354 of 516 (68.60%)

# 計算相關系數 from sklearn.metrics import matthews_corrcoefmatthews_set = []# 使用Matthew相關系數對每個測試批進行評估 print('Calculating Matthews Corr. Coef. for each batch...')# For each input batch... for i in range(len(true_labels)):# 這個批處理的預測是一個2列的ndarray(一個列是“0”，一個列是“1”)。 # 選擇值最高的label，并將其轉換為0和1的列表。pred_labels_i = np.argmax(predictions[i], axis=1).flatten()# Calculate and store the coef for this batch. matthews = matthews_corrcoef(true_labels[i], pred_labels_i) matthews_set.append(matthews)

最終的分數將基于整個測試集，但是讓我們看一下單個批次的分數，以了解批次之間度量的可變性。

每批有32個句子，除了最后一批只有(516% 32)= 4個測試句子。

創建一個柱狀圖，顯示每批測試樣品的MCC分數。

ax = sns.barplot(x=list(range(len(matthews_set))), y=matthews_set, ci=None)plt.title('MCC Score per Batch') plt.ylabel('MCC Score (-1 to +1)') plt.xlabel('Batch #')plt.show()

現在我們將合并所有批次的結果并計算我們最終的MCC分數。

# 合并所有批次的結果。 flat_predictions = np.concatenate(predictions, axis=0)# 對于每個樣本，選擇得分較高的標簽(0或1)。 flat_predictions = np.argmax(flat_predictions, axis=1).flatten()# 將每個批次的正確標簽組合成一個單獨的列表。 flat_true_labels = np.concatenate(true_labels, axis=0)# Calculate the MCC mcc = matthews_corrcoef(flat_true_labels, flat_predictions)print('Total MCC: %.3f' % mcc)

在大約半個小時的時間里，我們沒有做任何超參數的調整(learning rate, epochs, batch size, ADAM properties屬性等)，我們就獲得了一個很好的分數。

為了使分數最大化，我們應該刪除“驗證集”(我們用來幫助確定要訓練多少個紀元)，并在整個訓練集上訓練。

庫將基準測試此處的預期精度文檔為“49.23”。

官方排行?here.

請注意(由于數據集的大小較小?)在不同的運行中，精度可能會有很大的變化。

總結

這篇文章演示了使用預先訓練好的BERT模型，不管你感興趣的是什么特定的NLP任務，你都可以使用pytorch接口，用最少的努力和訓練時間，快速有效地創建一個高質量的模型。

附錄

A1. Saving & Loading Fine-Tuned Model

(取自’ run_glue。py 'here)將模型和標記器寫入磁盤。

import os# 保存best-practices:如果您使用模型的默認名稱，您可以使用from_pretraining()重新加載它 # Saving best-practices: if you use defaults names for the model, you can reload it using from_pretrained()output_dir = './model_save/'# 如果需要，創建輸出目錄 if not os.path.exists(output_dir):os.makedirs(output_dir)print("Saving model to %s" % output_dir)# 使用`save_pretrained()`保存訓練過的模型、配置和標記器。 # 用`from_pretrained()`重新加載模型。 model_to_save = model.module if hasattr(model, 'module') else model # 注意distributed/parallel training model_to_save.save_pretrained(output_dir) tokenizer.save_pretrained(output_dir)# Good practice: 保存訓練好的模型于模型參數 # torch.save(args, os.path.join(output_dir, 'training_args.bin'))

Revision History

Version 3?-?Mar 18th, 2020?- (current)

Simplified the tokenization and input formatting (for both training and test) by leveraging the?tokenizer.encode_plus?function.
encode_plus?handles padding?and?creates the attention masks for us.
Improved explanation of attention masks.
Switched to using?torch.utils.data.random_split?for creating the training-validation split.
Added a summary table of the training statistics (validation loss, time per epoch, etc.).
Added validation loss to the learning curve plot, so we can see if we’re overfitting.
- Thank you to?Stas Bekman?for contributing this!
Displayed the per-batch MCC as a bar plot.

Version 2?-?Dec 20th, 2019?-?link

huggingface renamed their library to?transformers.
Updated the notebook to use the?transformers?library.

Version 1?-?July 22nd, 2019

Initial version.

總結

以上是生活随笔為你收集整理的深入Bert实战(Pytorch)----fine-Tuning 2的全部內容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網站內容還不錯，歡迎將生活随笔推薦給好友。

上一篇： PyTorch环境下对BERT进行Fin
下一篇：图神经网络学习一(浅谈embedding