Created by: hyo009

线上多线程共用一个模型预测在成功预测完一些样本后会报错。 单机测试capi，在每个线程中初始化单独的模型可以顺利预测，但只初始化一个模型，通过struct传入thread共用会出现预测出一些结果后报错，不知原因。所有本地使用的样本都一样，输入格式应该没问题

enter thread1
this is thread1, thread id is 3907217152
enter thread2
this is thread2, thread id is 3915609856
thread1 model predict successfully!
thread2 model predict successfully!
thread1 model predict successfully!
thread2 model predict successfully!
thread1 model predict successfully!
thread2 model predict successfully!
thread1 model predict successfully!
thread2 model predict successfully!
thread1 model predict successfully!
thread2 model predict successfully!
thread1 model predict successfully!
thread2 model predict successfully!
thread1 model predict successfully!
thread2 model predict successfully!
thread1 model predict successfully!
thread2 model predict successfully!
thread1 model predict successfully!
thread2 model predict successfully!
thread1 model predict successfully!
F0412 10:25:28.660399   477 SequencePoolLayer.cpp:59] Check failed: starts->getData()[newBatchSize_] == input.getBatchSize() (32745 vs. 43) 
*** Check failure stack trace: ***
thread1 model predict successfully!
thread1 model predict successfully!
    @     0x7fe9ed83e76d  google::LogMessage::Fail()
    @     0x7fe9ed84221c  google::LogMessage::SendToLog()
thread1 model predict successfully!
    @     0x7fe9ed83e293  google::LogMessage::Flush()
    @     0x7fe9ed84372e  google::LogMessageFatal::~LogMessageFatal()
thread1 model predict successfully!
    @     0x7fe9ed88c138  paddle::SequencePoolLayer::forward()
    @     0x7fe9ed8ddb7d  paddle::MaxLayer::forward()
    @     0x7fe9ed9a24bd  paddle::NeuralNetwork::forward()
    @     0x7fe9ed83a676  paddle_gradient_machine_forward
thread1 model predict successfully!
    @           0x4031b3  TrafficModel::traffic_predict()
thread1 model predict successfully!
    @           0x4041d0  TrafficModel::predict()
thread1 model predict successfully!
    @           0x405ac0  thread2()
    @     0x7fe9ec9b71c3  start_thread
    @     0x7fe9eccb412d  __clone
    @              (nil)  (unknown)
Aborted