共计 4061 个字符，预计需要花费 11 分钟才能阅读完成。
MindSpore 常见问题次要包含数据加载与解决问题、网络构建与训练问题以及分布式并行配置问题等。MindSpore 在不同场景下，积攒了大量常见的问题案例。以下是对应的案例列表。
1、数据集加载与加强报错案例
MindSpore 数据集加载 - 调试小工具 py-spy【MindData】如何将自有数据高效的生成 MindRecord 格局数据集，并且避免爆内存。【MindData】在应用 Dataset 解决数据过程中内存占用高，怎么优化 如何解决数据集加载多过程(multiprocessing) 谬误 MindRecord-Windows 下中文门路问题 Unexpected error. Failed to open file MindRecord-Windows 下数据集报错 Invalid file, DB file can not match MindSpore 数据集加载 -GeneratorDataset 卡住、卡死 GeneratorDataset 报谬误 Unexpected error. Invalid data type GeneratorDataset 数据处理报错：The pointer[cnode] is null MindSpore 数据集报错【The data pipeline is not a tree】MindSpore 数据加强报错：Use Decode() for encoded data or ToPIL() MindSpore 自定义数据加强报错 args should be Numpy narray.Got class tuple MindSpore 数据集加载 -GeneratorDataset 性能及常见问题 MindSpore 数据集加载 -【too many open files】谬误 MindSpore 数据加强 -【Invalid with type】谬误 MindSpore 数据集格局 -【MindRecord File could not open successfully】MindSpore 数据集加载 -【has no attribute ‘child_sampler’】MindSpore 数据集加载 -【’DictIterator’ has no attribute ‘get_next’】谬误 MindSpore 数据集加载 -【IndexError: list index out of range】谬误 MindSpore 报错: Unexpected error. Inconsistent batch .. MindSpore 数据加强: 内存不足，主动退出 MindSpore 报错: Invalid data, Page size. MindSpore 报错: parse() missing 1 required positional. MindSpore 报错: Unable to data from Generator.. MindSpore 报错：GeneratorDataset’s num_workers=8, this value is … MindSpore 报错: Exception thrown from PyFunc. MindSpore 报错: Unexpected error. Invalid data.
2、网络构建与训练报错案例
MindSpore 报错: Exceed function call depth limit 1000 解决 MindSpore 报错: ‘Tensor’ has no attribute ‘data’ MindSpore 报错: ‘NoneType’ has no attribute ‘name’ Mindspore 报错: the dimension of logits must be equal to 2 MindSpore 报错：For ‘Pad’,all elements of paddings must >=0 MindSpore 报错：For primitive[BinaryCrossEntropy], the x shape must be equal to MindSpore 报错: For ‘Tensor’,the type of input_data should be one of ‘Tensor’, ‘ndarray’ … MindSpore 报错: Data type conversion of ‘Parameter’ is not supported MindSpore 报错: ScalarAdd 不反对 bool 类型 MindSpore 报错: ReduceSum 算子不反对 8 维及以上的输出 MindSpore 报错: Unsupported parameter type for python primitive.. MindSpore 报错：Primitive ScatterAdd’s bprop not defined MindSpore 报错：For ‘Conv2d’,x shape 的 C_in 除以 group 应等于 weight 的 C_in MindSpore 报错：For ‘Reshape’, the shape of ‘input_x’ is …,but product MindSpore 报错：For ‘Sub’, x.shape and y.shape are supposed to broadcast MindSpore 报错：For ‘MatMul’, the input dimensions must be equal, MindSpore 报错：Cann’t select valid kernel[Default/Pow-op] MindSpore 报错：rank in NLLLoss should be int and must be MindSpore 报错：operator[StridedSlice] input(kNumberTypeUInt16) output(kNumberTypeUInt16) is not supported. MindSpore 报错：Select GPU kernel op * fail! Incompatible data type MindSpore 报错：For ‘MirrorPad’, paddings must be a Tensor with MindSpore 报错：For primitive[TensorSummary], the v rank must be greater than 0 MindSpore 报错：For ‘CellList’, each cell should be subclass of Cell MindSpore 报错：task_fail_info or current_graph_ is nullptr MindSpore 报错：padding_idx in Embedding 超出范围的报错 MindSpore 报错：seed2 in StandardNormal should be int and >=0 MindSpore 报错：ReduceMean 不反对 8 维及其以上的输出 MindSpore 报错：output_shape shape element [2] must be positive integer or SHP_ANY MindSpore 报错：For ‘TopK’, the type of ‘x’ should be … MindSpore 报错：Minimum inputs size 0 does not match… MindSpore 报错：half_pixel_centers=True only support in Ascend MindSpore 报错：For ‘AvgPool’ 输入的 shape 每一维都要大于零
MindSpore 报错：The function construct need xx positional argument … MindSpore 报错：Operator[] input() output() is not support MindSpore 报错：sth should be initialized as a Parameter type in … MindSpore 报错：”operation does not support the type kMetaTypeNone” MindSpore 报错：When ‘…’ by using Fallback feature, an error occurred MindSpore 报错: Net parameters weight shape xxx i MindSpore GPU 设施算力有余导致计算结果谬误 MindSpore 报错：设置卡失败 SetDevice failed mindspore 报错：Total stream number xxx exceeds the limit of … MindSpore 报错：[LoadTask] Distribute Task Failed MindSpore 报错：GPU 训练提醒调配流失败[cudaStreamCreate failed] MindSpore 报错：呈现内存申请大小为 0 的谬误[The memory alloc size is 0] MindSpore 报错：Please input the correct checkpoint MindSpore 告警：”xxx parameters in the net are not MindSpore 报错：For ‘context.set_context’, package type xxx support ‘devic
MindSpore 报错：PyNative 模式下 The pointer[top_cell_] is null MindSpore 报错：Malloc device memory failed MindSpore 报错: Please input the correct checkpoint MindSpore 求导传入 sens 值报错: For ‘MatMul’, the input dimensions
3、分布式并行报错案例
报错：the value of strategy must be the power of 2, but get xx 报错：PyNative Only support STAND_ALONE, DATA_PARALLEL… 报错：Parallel mode dose not support 报错：davinci_model : load task fail, return ret 报错：connect returned Connection timed out