resnet50能用cpu跑吗_tutorial下的cifar_resnet50 改成cpu训练报错

| name | about | labels |

| ---------- | ------------------------------------- | -------- |

| Bug Report | Use tutorials_code for reporting a bug | bug |

## Environment

- **CPU**:

- **Software Environment**:

-- MindSpore version :mindspore/mindspore-cpu:0.1.0-alpha

-- Python 3.7.5

-- OS platform: MacOS Docker19.03.8

## Describe the current behavior

tutorials_code中的示例：[cifar_resnet50.py](https://gitee.com/mindspore/docs/blob/master/tutorials/tutorial_code/resnet/cifar_resnet50.py) 是在Ascend上训练的，由于我安装的是cpu版本的me，所以将这行代码：

```

context.set_context(mode=context.GRAPH_MODE, device_target="Ascend")

```

改成：

```

context.set_context(mode=context.GRAPH_MODE, device_target="CPU")

```

会报如下错误：

```

RuntimeError: mindspore/ccsrc/device/cpu/cpu_session.cc:115 BuildKernel] Operator[InitDataSetQueue] is not support.

```

我发现，CPU环境不支持dataset_sink_mode=True，但是train的时候是默认dataset_sink_mode=True的，于是我在这里加上了：

```

model.train(epoch_size, dataset, callbacks=[ckpoint_cb, loss_cb],dataset_sink_mode=False)

```

解决了上面出现的问题，但是运行时会自动被kill：

```

[WARNING] ME(2987:139919646736512,MainProcess):2020-03-30-10:13:22.847.454 [mindspore/dataset/engine/datasets.py:823] Repeat is located before batch, data from two epochs can be batched together.

Killed

```

我查了下docker容器里的使用情况，空间也足够

```

root@b5a013bc4b55:~/docs/tutorials/tutorial_code/resnet# free -h

total used free shared buff/cache available

Mem: 1.9G 123M 1.7G 4K 91M 1.7G

Swap: 1.0G 341M 682M

```

请问这个应该如何改呢？我在跑lenet.py的示例时，也是将device_target改成了CPU，可以正常运行，不知道为什么在这里就会报错 T_T

>ps:在处理数据时，这里有两行代码写反了，建议先用batch再用repeat

```

cifar_ds = cifar_ds.shuffle(buffer_size=10)

cifar_ds =cifar_ds.batch(batch_size=args_opt.batch_size,drop_remainder=True)

cifar_ds = cifar_ds.repeat(repeat_num))

```