How can take advantage of multiprocessing and multithreading in Deep learning using Keras?

I'd assume that most frameworks like keras/tensorflow/... automatically use all CPU cores but in practice it seems they are not. I just could find few sources which can lead us to use whole capacity of CPU during Deep learning process. I found an article which is written about usage of

from multiprocessing import Pool 
import psutil
import ray 

in another hand, based on this answer for using a keras model in multiple processes there is no track of above-mentioned libraries. Is there the more elegant way to take advantage of Multiprocessing for Keras since it's very popular for implementation.

• For instance , how can modify following simple RNN implementation to achieve at least 50% capacity of CPU during learning process?

• Should I use 2nd model as multitasking like LSTM which I comment bellow? I mean can we simultaneously manage to run multi-models by using more capacity of CPU?

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

from keras.layers.normalization import BatchNormalization
from sklearn.preprocessing import MinMaxScaler
from sklearn.metrics import mean_squared_error
from sklearn.metrics import accuracy_score
from keras.layers import Dense
from keras.layers import Dropout
from keras.layers import LSTM,SimpleRNN
from keras.models import Sequential
from keras.optimizers import Adam, RMSprop

df = pd.read_csv("D:\Train.csv", header=None)

index = [i for i in list(range(1440)) if i%3==2]

Y_train= df[index]
df = df.values

#making history by using look-back to prediction next
def create_dataset(dataset,data_train,look_back=1):
    dataX,dataY = [],[]
    print("Len:",len(dataset)-look_back-1)
    for i in range(len(dataset)-look_back-1):
        a = dataset[i:(i+look_back), :]
        dataX.append(a)
        dataY.append(data_train[i + look_back,  :])
    return np.array(dataX), np.array(dataY)

Y_train=np.array(Y_train)
df=np.array(df)

look_back = 10
trainX,trainY = create_dataset(df,Y_train, look_back=look_back)

#Split data into train & test
trainX, testX, trainY, testY = train_test_split(trainX,trainY, test_size=0.2 , shuffle=False)



model_RNN = Sequential()
model_RNN.add(SimpleRNN(units=1440, input_shape=(trainX.shape[1], trainX.shape[2])))
model_RNN.add(Dense(960))
model_RNN.add(BatchNormalization())
model_RNN.add(Activation('tanh'))
# Compile model
model_RNN.compile(loss='mean_squared_error', optimizer='adam')
callbacks = [
    EarlyStopping(patience=10, verbose=1),
    ReduceLROnPlateau(factor=0.1, patience=3, min_lr=0.00001, verbose=1)]
# Fit the model
hist_RNN=model_RNN.fit(trainX, trainY, epochs =50, batch_size =20,validation_data=(testX,testY),verbose=1, callbacks=callbacks)


#predict

Y_train=np.array(trainY)
Y_test=np.array(testX)

Y_RNN_Train_pred=model_RNN.predict(trainX)
Y_RNN_Test_pred=model_RNN.predict(testX)

train_MSE=mean_squared_error(trainY, Y_RNN_Train_pred)
test_MSE=mean_squared_error(testY, Y_RNN_Test_pred)

# create and fit the Simple LSTM model as 2nd model for multi-tasking

#model_LSTM = Sequential()
#model_LSTM.add(LSTM(units = 1440, input_shape=(trainX.shape[1], trainX.shape[2])))
#model_LSTM.add(Dense(units = 960))
#model_LSTM.add(BatchNormalization())
#model_LSTM.add(Activation('tanh'))
#model_LSTM.compile(loss='mean_squared_error', optimizer='adam')
#hist_LSTM=model_LSTM.fit(trainX, trainY, epochs =50, batch_size =20,validation_data=(testX,testY),verbose=1, callbacks=callbacks)

#Y_train=np.array(trainY)
#Y_test=np.array(testX)

#Y_LSTM_Train_pred=model_LSTM.predict(trainX)
#Y_LSTM_Test_pred=model_LSTM.predict(testX)

#train_MSE=mean_squared_error(trainY, Y_LSTM_Train_pred)
#test_MSE=mean_squared_error(testY, Y_LSTM_Test_pred)

Note I don't access to CUDA just I access powerful server without VGA. My aim is to take advantage of multiprocessing and multithreading for use maximum capacity of CPU instead of 30% it means just one core while I have Quad-core! Any advice would be greatly appreciated. I have uploaded a formatted csv dataset.

from How can take advantage of multiprocessing and multithreading in Deep learning using Keras?