如何在Keras+DistilBERT的model.fit中正确加入验证数据？

我是NLP新手，正在使用DistilBERT在Kaggle的NLP入门数据集上进行二分类任务。其他流程均正常，但在model.fit中加入验证数据后，epoch输出里始终不显示val_accuracy和val_loss。以下是我的代码及当前输出，恳请各位提供解决思路。

代码 #

#import libraries
import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split
from transformers import DistilBertTokenizer, TFDistilBertForSequenceClassification
import tensorflow as tf

#import tokenizer and pre-trained model
tokenizer = DistilBertTokenizer.from_pretrained('distilbert-base-uncased', do_lower_case=True)
model = TFDistilBertForSequenceClassification.from_pretrained('distilbert-base-uncased')

#random seed
random_seed = 42

#set random seed in tensorflow
tf.random.set_seed(random_seed)

#set random seed in numpy
np.random.seed(random_seed)

#load the raw training data
df_raw_train = pd.read_csv("data/train.csv")
#make a copy of df_raw_train
df_train = df_raw_train.copy(deep=True)

#load the raw test data
df_raw_test = pd.read_csv("data/test.csv")
#make a copy of df_raw_test
df_test = df_raw_test.copy(deep=True)

#get target name
target = 'target'

#drop columns
df_train.drop(['id','keyword','location'],axis=1,inplace=True)
df_test.drop(['id','keyword','location'],axis=1,inplace=True)

#training (80%) and validation (20%) data split
df_train, df_val = train_test_split(df_train, train_size=0.8, random_state=random_seed)

#reset index
df_train, df_val = df_train.reset_index(drop=True), df_val.reset_index(drop=True)

#batch tokenize our tweet field
X_train = tokenizer.batch_encode_plus(df_train.text, pad_to_max_length=True, return_tensors="tf")
X_val = tokenizer.batch_encode_plus(df_val.text, pad_to_max_length=True, return_tensors="tf")
X_test = tokenizer.batch_encode_plus(df_test.text, pad_to_max_length=True, return_tensors="tf")

#obtain target
y_train = df_train['target'].to_numpy()
y_val = df_val['target'].to_numpy()

#optimize model
optimizer = tf.keras.optimizers.Adam(learning_rate=3e-5, epsilon=1e-08, clipnorm=1.0)
loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)
bce = tf.keras.losses.BinaryCrossentropy()
metric = tf.keras.metrics.SparseCategoricalAccuracy('accuracy')

model.compile(optimizer=optimizer, loss=loss, metrics=[metric])
model.fit(x=X_train['input_ids'], y=y_train, epochs=2, batch_size=15, verbose=2, validation_data=(X_val, y_val))

当前输出 #

Epoch 1/2
343/508 [===================>..........] - ETA: 6:30 - loss: 0.4728 - accuracy: 0.7811

• 输入格式不匹配：训练时你只传入了X_train['input_ids']，但验证数据传入的是完整的X_val字典（包含input_ids和attention_mask）。模型训练和验证的输入结构必须一致，要么都传完整的tokenizer输出，要么都只传input_ids。修改model.fit的训练输入为完整的X_train即可：

  model.fit(x=X_train, y=y_train, epochs=2, batch_size=15, verbose=2, validation_data=(X_val, y_val))
  

• 验证指标显示时机：你设置的verbose=2会在每个epoch结束后输出验证指标，当前输出是训练中的进度提示（ETA），等第一个epoch跑完后才会显示val_loss和val_accuracy，前提是输入格式已修正。
• 指标与损失匹配：你使用的SparseCategoricalAccuracy和SparseCategoricalCrossentropy搭配是合理的（标签为0/1整数），这部分无需调整，确保验证标签y_val和训练标签格式一致即可。

内容的提问来源于stack exchange，提问作者Paul