如何在MxNet中恢复具有优化器状态的训练

import mxnet as mx
from mxnet import gluon

# 定义模型
net = gluon.nn.Sequential()
# 添加网络层
# ...

# 创建优化器
optimizer = mx.optimizer.SGD(learning_rate=0.01)
trainer = gluon.Trainer(net.collect_params(), optimizer)

# 训练并保存模型参数和优化器状态
# ...

# 加载之前保存的模型参数和优化器状态
net.load_parameters('model.params')
trainer.load_states('trainer.states')

# 创建新的优化器，并与已加载的模型参数关联
new_optimizer = mx.optimizer.SGD(learning_rate=0.01)
new_trainer = gluon.Trainer(net.collect_params(), new_optimizer)

# 继续训练
# ...
for epoch in range(num_epochs):
    for data, label in train_data:
        with mx.autograd.record():
            output = net(data)
            loss = loss_func(output, label)
        loss.backward()
        new_trainer.step(batch_size)

# 保存新的模型参数和优化器状态
net.save_parameters('new_model.params')
new_trainer.save_states('new_trainer.states')