在简单的三层神经网络训练中发现训练速度越来越慢，开始很快，后面血慢，请问这是什么问题？

def reader_creator(train_data):
    def reader():

        for d in train_data:
            yield d[:-1], d[-1:]
    return reader
print(reader_creator(train_data=f_data))

train_reader = fluid.io.batch(
    fluid.io.shuffle(
        reader_creator(f_data), buf_size=500),
        batch_size=BATCH_SIZE)
#print(train_reader)
test_reader = fluid.io.batch(
    fluid.io.shuffle(
        reader_creator(f_test_data), buf_size=500),
        batch_size=BATCH_SIZE)



Rs_Ra_T_H = fluid.data(name='x', shape=[None, 3], dtype='float32') # 定义输入的形状和数据类型
C = fluid.data(name='y', shape=[None, 1], dtype='float32') # 定义输出的形状和数据类型
hidden1 = fluid.layers.fc(input=Rs_Ra_T_H, size=3, act='softmax') # 连接输入和输出的全连接层
hidden2 = fluid.layers.fc(input=hidden1,size=7,act='tanh')
prediction=fluid.layers.fc(input=hidden2,size=1)

cost = fluid.layers.square_error_cost(input=prediction, label=C) # 利用标签数据和输出的预测数据估计方差
avg_loss = fluid.layers.mean(cost) # 对方差求均值，得到平均损失
FLAGS_eager_delete_tensor_gb=0.0

main_program = fluid.default_main_program() # 获取默认/全局主函数
startup_program = fluid.default_startup_program() # 获取默认/全局启动程序





#克隆main_program得到test_program
#有些operator在训练和测试之间的操作是不同的，例如batch_norm，使用参数for_test来区分该程序是用来训练还是用来测试
#该api不会删除任何操作符,请在backward和optimization之前使用
test_program = main_program.clone(for_test=True)

sgd_optimizer = fluid.optimizer.SGD(learning_rate=0.001)
sgd_optimizer.minimize(avg_loss)


use_cuda = True
place = fluid.CUDAPlace(0) if use_cuda else fluid.CPUPlace() # 指明executor的执行场所

###executor可以接受传入的program，并根据feed map(输入映射表)和fetch list(结果获取表)向program中添加数据输入算子和结果获取算子。使用close()关闭该executor，调用run(...)执行program。
exe = fluid.Executor(place)


num_epochs = 100

def train_test(executor, program, reader, feeder, fetch_list):
    accumulated =[0]
    count = 0
    for data_test in reader():
        outs = executor.run(program=program,
                            feed=feeder.feed(data_test),
                            fetch_list=fetch_list)
        accumulated = [x_c[0] + x_c[1][0] for x_c in zip(accumulated, outs)] # 累加测试过程中的损失值
        count += 1 # 累加测试集中的样本数量
    return [x_d / count for x_d in accumulated] # 计算平均损失



params_dirname = "Ann.inference.model"
feeder = fluid.DataFeeder(place=place, feed_list=[Rs_Ra_T_H, C])
exe.run(startup_program)
train_prompt = "train cost"
test_prompt = "test cost"
from paddle.utils.plot import Ploter
plot_prompt = Ploter(train_prompt, test_prompt)
step = 0

exe_test = fluid.Executor(place)

for pass_id in range(num_epochs):
    print("pass_id="+str(pass_id))
    for data_train in train_reader():
        avg_loss_value, = exe.run(main_program,
                                  feed=feeder.feed(data_train),
                                  fetch_list=[avg_loss])
        if step % 10 == 0: # 每10个批次记录并输出一下训练损失
            plot_prompt.append(train_prompt, step, avg_loss_value[0])
            plot_prompt.plot()
            print("%s, Step %d, Cost %f" %
	                  (train_prompt, step, avg_loss_value[0]))
        if step % 100 == 0:  # 每100批次记录并输出一下测试损失
            test_metics = train_test(executor=exe_test,
                                     program=test_program,
                                     reader=test_reader,
                                     fetch_list=[avg_loss],
                                     feeder=feeder)
            plot_prompt.append(test_prompt, step, test_metics[0])
            plot_prompt.plot()
            print("%s, Step %d, Cost %f" %
	                  (test_prompt, step, test_metics[0]))
            if test_metics[0] < 0.01: # 如果准确率达到要求，则停止训练
                break

        step += 1

        if math.isnan(float(avg_loss_value[0])):
            sys.exit("got NaN loss, training failed.")

        #保存训练参数到之前给定的路径中
        if params_dirname is not None:
            fluid.io.save_inference_model(params_dirname, ['x'], [prediction], exe)