#947 RuntimeError：E40021和E61001

你好，可以提供个简化的代码吗，从这个报错信息上看像是自动微分报的错，有可能是框架bug导致。一般如果期望GPU和NPU进行比较的时候，最好保持mindspore以及mindtorch的版本是一致的，可以快速排查是框架还是具体硬件算子的影响

我在官网上找到的GPU最新版本是mindtorch0.2.1+mindspore2.2.14（我在AutoDL上调试的），而Ascend调试任务只有mindtorch0.3版本的镜像可用。所以版本不一致，我晚点提供一个简化版的代码。

Ascend调试任务中，并且在图模式下，模型可以正常训练。但相同的环境，到训练任务中，训练就失败了。（我之前没有设置ms.set_context，用的应该是默认的配置）。报错有点区别:

----------------------------------------------------
- Ascend Error Message:
----------------------------------------------------
E40021: 2024-08-06-06:52:49.033.290 Failed to compile Op [SplitD25]. (oppath: [Compile /usr/local/Ascend/ascend-toolkit/8.0.RC1/opp/built-in/op_impl/ai_core/tbe/impl/dynamic/split_d.py failed with errormsg/stack: File "/usr/local/Ascend/ascend-toolkit/latest/python/site-packages/tbe/dsl/varshape/split_variable_shape.py", line 46, in check_params
    raise RuntimeError(err_msg__args, get_error_message(err_msg__args))
RuntimeError: ({'errCode': 'E90001', 'detailed_cause': 'split numbers error, split numbers must be greater 0 and less equal 63, now, it is 128'}, 'Compile operator failed, cause: Parameters check failed, detailed information: split numbers error, split numbers must be greater 0 and less equal 63, now, it is 128.')
], optype: [SplitD])[THREAD:2876]
        Solution: See the host log for details, and then check the Python stack where the error log is reported.
        TraceBack (most recent call last):
        Compile op[SplitD25] failed, oppath[/usr/local/Ascend/ascend-toolkit/8.0.RC1/opp/built-in/op_impl/ai_core/tbe/impl/dynamic/split_d.py], optype[SplitD], taskID[143]. Please check op's compilation error message.[FUNC:ReportBuildErrMessage][FILE:fusion_manager.cc][LINE:748][THREAD:2876]
        [SubGraphOpt][Compile][ProcFailedCompTask] Thread[281464742588640] recompile single op[SplitD25] failed[FUNC:ProcessAllFailedCompileTasks][FILE:tbe_op_store_adapter.cc][LINE:962][THREAD:2876]
        [SubGraphOpt][Compile][ParalCompOp] Thread[281464742588640] process fail task failed[FUNC:ParallelCompileOp][FILE:tbe_op_store_adapter.cc][LINE:1010][THREAD:2876]
        [SubGraphOpt][Compile][CompOpOnly] CompileOp failed.[FUNC:CompileOpOnly][FILE:op_compiler.cc][LINE:1119][THREAD:2876]
        [GraphOpt][FusedGraph][RunCompile] Failed to compile graph with compiler Normal mode Op Compiler[FUNC:SubGraphCompile][FILE:fe_graph_optimizer.cc][LINE:1385][THREAD:2876]
        Call OptimizeFusedGraph failed, ret:-1, engine_name:AIcoreEngine, graph_name:partition0_rank1_new_sub_graph1[FUNC:OptimizeSubGraph][FILE:graph_optimize.cc][LINE:126][THREAD:2876]
        subgraph 0 optimize failed[FUNC:OptimizeSubGraphWithMultiThreads][FILE:graph_manager.cc][LINE:1021][THREAD:860]
        build graph failed, graph id:24, ret:-1[FUNC:BuildModelWithGraphId][FILE:ge_generator.cc][LINE:1615][THREAD:860]
        [Build][SingleOpModel]call ge interface generator.BuildSingleOpModel failed. ge result = 4294967295[FUNC:ReportCallError][FILE:log_inner.cpp][LINE:161][THREAD:860]
        [Build][Op]Fail to build op model[FUNC:ReportInnerError][FILE:log_inner.cpp][LINE:145][THREAD:860]
        build op model failed, result = 500002[FUNC:ReportInnerError][FILE:log_inner.cpp][LINE:145][THREAD:860]

(Please search "CANN Common Error Analysis" at https://www.mindspore.cn for error code description)

----------------------------------------------------
- C++ Call Stack: (For framework developers)
----------------------------------------------------
mindspore/ccsrc/transform/acl_ir/acl_utils.cc:379 Run

[ERROR] DEVICE(14,ffff71ac00e0,python):2024-08-06-06:52:49.420.884 [mindspore/ccsrc/plugin/device/ascend/hal/hardware/ge_kernel_executor.cc:1007] LaunchKernel] Launch kernel failed, kernel full name: Default/Split-op5
[ERROR] RUNTIME_FRAMEWORK(14,ffff71ac00e0,python):2024-08-06-06:52:49.506.350 [mindspore/ccsrc/runtime/graph_scheduler/actor/kernel_async_launch_actor.cc:38] LaunchKernel] Failed to launch kernel: Default/Split-op5 and catch exception: 
----------------------------------------------------
- Kernel error:
----------------------------------------------------
Launch kernel failed: Default/Split-op5

----------------------------------------------------
- The Function Call Stack: (For framework developers)
----------------------------------------------------
In file /home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/ops/function/array_func.py:4576/        res = _get_cache_prim(P.Split)(axis, sections)(x)/
In file /home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/ops/function/array_func.py:4565/def _split_int(x, split_size_or_sections, axis):/
In file /home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/ops/function/array_func.py:4658/            res = _split_int(tensor, split_size_or_sections, arr_axis)/
In file /home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/ops/function/array_func.py:4612/def split(tensor, split_size_or_sections, axis=0):/
In file /home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindtorch-0.3.0-py3.9.egg/mindtorch/torch/functional.py:1264/    output = ms.ops.split(tensor, split_size_or_sections, dim)/
In file /home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindtorch-0.3.0-py3.9.egg/mindtorch/torch/functional.py:1262/def split(tensor, split_size_or_sections, dim=0):/
In file /tmp/code/iplapmindspore/models_ms/landmark_generator.py:224/        pose_embedding = torch.stack(torch.split(pose_embedding, T), dim=0) # (B,T,512)/
In file /tmp/code/iplapmindspore/models_ms/landmark_generator.py:203/    def forward(self, T_mels, T_pose, Nl_pose, Nl_content):/
In file /home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindtorch-0.3.0-py3.9.egg/mindtorch/torch/nn/modules/module.py:430/        return self.forward(*inputs, **kwargs)/
In file /home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindtorch-0.3.0-py3.9.egg/mindtorch/torch/nn/modules/module.py:429/    def construct(self, *inputs, **kwargs):/

----------------------------------------------------
- C++ Call Stack: (For framework developers)
----------------------------------------------------
mindspore/ccsrc/runtime/graph_scheduler/actor/kernel_actor.cc:753 ExecuteLaunchKernelTask


  0%|          | 0/25 [05:50<?, ?it/s]

  0%|          | 0/332 [06:03<?, ?it/s]
Traceback (most recent call last):
  File "/tmp/code/iplapmindspore/train_landmarks_generator_ms_cuda.py", line 344, in <module>
    evaluate(model, val_data_loader)
  File "/tmp/code/iplapmindspore/train_landmarks_generator_ms_cuda.py", line 269, in evaluate
    predict_content = model(T_mels, T_pose,Nl_pose,Nl_content)  # (B*T,2,57)
  File "/home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/nn/cell.py", line 685, in __call__
    out = self.compile_and_run(*args, **kwargs)
  File "/home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/nn/cell.py", line 1006, in compile_and_run
    return _cell_graph_executor(self, *new_args, phase=self.phase)
  File "/home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/common/api.py", line 1631, in __call__
    return self.run(obj, *args, phase=phase)
  File "/home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/common/api.py", line 1670, in run
    return self._exec_pip(obj, *args, phase=phase_real)
  File "/home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/common/api.py", line 130, in wrapper
    results = fn(*arg, **kwargs)
  File "/home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/common/api.py", line 1650, in _exec_pip
    return self._graph_executor(args, phase)
RuntimeError: 
----------------------------------------------------
- Kernel error:
----------------------------------------------------
Launch kernel failed: Default/Split-op5

----------------------------------------------------
- The Function Call Stack: (For framework developers)
----------------------------------------------------
In file /home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/ops/function/array_func.py:4576/        res = _get_cache_prim(P.Split)(axis, sections)(x)/
In file /home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/ops/function/array_func.py:4565/def _split_int(x, split_size_or_sections, axis):/
In file /home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/ops/function/array_func.py:4658/            res = _split_int(tensor, split_size_or_sections, arr_axis)/
In file /home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/ops/function/array_func.py:4612/def split(tensor, split_size_or_sections, axis=0):/
In file /home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindtorch-0.3.0-py3.9.egg/mindtorch/torch/functional.py:1264/    output = ms.ops.split(tensor, split_size_or_sections, dim)/
In file /home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindtorch-0.3.0-py3.9.egg/mindtorch/torch/functional.py:1262/def split(tensor, split_size_or_sections, dim=0):/
In file /tmp/code/iplapmindspore/models_ms/landmark_generator.py:224/        pose_embedding = torch.stack(torch.split(pose_embedding, T), dim=0) # (B,T,512)/
In file /tmp/code/iplapmindspore/models_ms/landmark_generator.py:203/    def forward(self, T_mels, T_pose, Nl_pose, Nl_content):/
In file /home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindtorch-0.3.0-py3.9.egg/mindtorch/torch/nn/modules/module.py:430/        return self.forward(*inputs, **kwargs)/
In file /home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindtorch-0.3.0-py3.9.egg/mindtorch/torch/nn/modules/module.py:429/    def construct(self, *inputs, **kwargs):/

----------------------------------------------------
- C++ Call Stack: (For framework developers)
----------------------------------------------------
mindspore/ccsrc/runtime/graph_scheduler/actor/kernel_actor.cc:753 ExecuteLaunchKernelTask
failed

如果是自动微分报错的话，我把我的模型训练的代码，也就是train_step函数相关的代码。提了出来然后再做测试，输入用torch.rand生成的和我输入数据一致的随机张量。发现可以正常运行。那应该不是微分的问题？还是说是特定的数值（比如我的训练集）会导致这样的报错？

这里的调试任务和训练任务的差异点是什么？不是很理解这个描述，是说云脑镜像申请的模式吗？

我在官网上找到的GPU最新版本是mindtorch0.2.1+mindspore2.2.14（我在AutoDL上调试的），而Ascend调试任务只有mindtorch0.3版本的镜像可用。所以版本不一致，我晚点提供一个简化版的代码。

配置ms.set_context(device_target="CPU")试试呢？先看下其他硬件条件下流程是否正常。

这里的调试任务和训练任务的差异点是什么？不是很理解这个描述，是说云脑镜像申请的模式吗？

是的，就是云脑中分的那些调试任务、训练任务、推理任务等。如下图：

配置ms.set_context(device_target="CPU",mode=0)会报错：

Traceback (most recent call last):
  File "/tmp/code/iplapmindspore/train_landmarks_generator_ms_cuda.py", line 351, in <module>
    loss,L1_loss,Velocity_loss = train_step(T_mels, T_pose, Nl_pose, Nl_content, T_content)
  File "/tmp/code/iplapmindspore/train_landmarks_generator_ms_cuda.py", line 333, in train_step
    (loss, _, L1_loss, Velocity_loss), grads = grad_fn(T_mels, T_pose, Nl_pose, Nl_content, T_content)
  File "/home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/common/api.py", line 741, in staging_specialize
    out = _MindsporeFunctionExecutor(func, hash_obj, input_signature, process_obj, jit_config)(*args, **kwargs)
  File "/home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/common/api.py", line 130, in wrapper
    results = fn(*arg, **kwargs)
  File "/home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/common/api.py", line 360, in __call__
    raise err
  File "/home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/common/api.py", line 357, in __call__
    phase = self.compile(self.fn.__name__, *args_list, **kwargs)
  File "/home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/common/api.py", line 449, in compile
    is_compile = self._graph_executor.compile(self.fn, compile_args, kwargs, phase, True)
ValueError: For 'BiasAddGrad', input tensor's dimension is 3, when data_format is NCHW the last dimension size should greater than 1.

----------------------------------------------------
- C++ Call Stack: (For framework developers)
----------------------------------------------------
mindspore/ccsrc/plugin/device/cpu/kernel/bias_add_grad_cpu_kernel.cc:45 Resize
failed

配置ms.set_context(device_target="CPU")会报错：

Traceback (most recent call last):
  File "/tmp/code/iplapmindspore/train_landmarks_generator_ms_cuda.py", line 351, in <module>
    loss,L1_loss,Velocity_loss = train_step(T_mels, T_pose, Nl_pose, Nl_content, T_content)
  File "/tmp/code/iplapmindspore/train_landmarks_generator_ms_cuda.py", line 333, in train_step
    (loss, _, L1_loss, Velocity_loss), grads = grad_fn(T_mels, T_pose, Nl_pose, Nl_content, T_content)
  File "/home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/ops/composite/base.py", line 626, in after_grad
    return grad_(fn_, weights)(*args, **kwargs)
  File "/home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/common/api.py", line 130, in wrapper
    results = fn(*arg, **kwargs)
  File "/home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/ops/composite/base.py", line 602, in after_grad
    out = _pynative_executor.grad(fn, grad_, weights, grad_position, *args, **kwargs)
  File "/home/ma-user/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/common/api.py", line 1297, in grad
    return self._executor.grad(grad, obj, weights, grad_position, *args, *(kwargs.values()))
ValueError: For 'BiasAddGrad', input tensor's dimension is 3, when data_format is NCHW the last dimension size should greater than 1.

----------------------------------------------------
- C++ Call Stack: (For framework developers)
----------------------------------------------------
mindspore/ccsrc/plugin/device/cpu/kernel/bias_add_grad_cpu_kernel.cc:45 Resize
failed

你在调试环境中是直接执行训练程序，没有对环境做出过修改么，看到你训练模式下的这个报错应该是数据格式算子不支持，那在同样环境下，调试任务应该也是不支持会报错才对。

@davislee 以调试模式的结果为准。mindtorch0.3的镜像在训练模式没有做过正式的测试。在训练模式中是通过docker run启动的任务，并没有进入到bash的环境，可能会出现意料之外的错误。

你在调试环境中是直接执行训练程序，没有对环境做出过修改么，看到你训练模式下的这个报错应该是数据格式算子不支持，那在同样环境下，调试任务应该也是不支持会报错才对。

安装过其他的python依赖包（通过python执行shell脚本的方式），但是在训练任务和测试任务中程序运行的过程都是一模一样的，不知道为什么调试环境没有报错但是训练环境报错了

@davislee 以调试模式的结果为准。mindtorch0.3的镜像在训练模式没有做过正式的测试。在训练模式中是通过docker run启动的任务，并没有进入到bash的环境，可能会出现意料之外的错误。

可是调试模式有4小时的限制，我想训练模型的话好像只能到训练任务中，但训练任务会报这个错误。

又复现了bug

import mindtorch.torch as torch
import mindspore as ms
import mindtorch.torch.nn as nn


class SimpleModel(nn.Module):
    def __init__(self):
        super(SimpleModel, self).__init__()
        self.conv = nn.Conv1d(
            in_channels=2,  # 输入通道数
            out_channels=512,  # 输出通道数
            kernel_size=131,  # 卷积核大小，选择与输入序列长度相同的值
            stride=1,  # 步长
            padding=0,  # 填充
        )

    def forward(self, Nl_pose, Nl_content):
        # (B,T,1,hv,wv) (B,T,2,74) (B,N_l,2,74)  (B,N_l,2,57)
        N_l = Nl_content.size(1)

        Nl_ref = torch.cat([Nl_pose, Nl_content], dim=3)  # (B,Nl,2,131=74+57)
        Nl_ref = torch.cat([Nl_ref[i] for i in range(Nl_ref.size(0))], dim=0)  # (B*Nl,2,131)

        ref_embedding = self.conv(Nl_ref).squeeze(-1)  # (B*Nl,512)

        ref_embedding = torch.stack(torch.split(ref_embedding, N_l, dim=0), dim=0)  # (B,N_l,512)

        return ref_embedding  # (B*T,2,57)



ms.set_context(device_target="Ascend", device_id=4, ascend_config={"precision_mode": "allow_fp32_to_fp16"})

# 定义输入

Nl_pose = torch.randn(128, 15, 2, 74)
Nl_content = torch.randn(128, 15, 2, 57)

# 将输入打包
inputs = [Nl_pose, Nl_content]

# 定义模型
model = SimpleModel()

# 定义优化器和损失函数
optimizer = torch.optim.Adam(model.parameters(), lr=0.001)
criterion_L1 = torch.nn.L1Loss()

# 定义前向传播并定义grad函数


def forward_fn(inputs):
    predict_content = model(*inputs)
    L1_loss = criterion_L1(predict_content, torch.zeros(*predict_content.shape))
    print(f'L1_loss:{L1_loss}')
    return L1_loss, predict_content


grad_fn = ms.ops.value_and_grad(forward_fn, None, optimizer.parameters, has_aux=True)
grad_fn(inputs)
print("success")

报错

[ERROR] KERNEL(270214,fffdd3fff160,python):2024-09-19-01:46:09.655.677 [mindspore/ccsrc/plugin/device/ascend/kernel/acl/acl_kernel_mod.cc:260] Launch] Kernel launch failed, msg: Acl compile and execute failed, op_type_:Unpack

----------------------------------------------------
- Ascend Error Message:
----------------------------------------------------
E61001: 2024-09-19-01:46:09.274.889 In op [te_unpack_076dce63228e6644175f2a39b4445a6c3013edac76e87619911779e7acbabd45_1], [input number is too much][THREAD:271372]
        TraceBack (most recent call last):
        Failed to compile Op [Unpack7]. (oppath: [Compile /home/shuziren/Ascend/ascend-toolkit/8.0.RC2/opp/built-in/op_impl/ai_core/tbe/impl/dynamic/unpack.py failed with errormsg/stack: File "/home/shuziren/miniconda3/envs/iplap/lib/python3.10/site-packages/tbe/common/utils/errormgr/error_manager_vector.py", line 284, in raise_err_specific_reson
    raise RuntimeError(args_dict, msg)
RuntimeError: ({'errCode': 'E61001', 'op_name': 'te_unpack_076dce63228e6644175f2a39b4445a6c3013edac76e87619911779e7acbabd45_1', 'reason': 'input number is too much'}, 'In op [te_unpack_076dce63228e6644175f2a39b4445a6c3013edac76e87619911779e7acbabd45_1], [input number is too much]')
], optype: [Unpack])[THREAD:271372]
        Compile op[Unpack7] failed, oppath[/home/shuziren/Ascend/ascend-toolkit/8.0.RC2/opp/built-in/op_impl/ai_core/tbe/impl/dynamic/unpack.py], optype[Unpack], taskID[40]. Please check op's compilation error message.[FUNC:ReportBuildErrMessage][FILE:fusion_manager.cc][LINE:748][THREAD:271372]
        [SubGraphOpt][Compile][ProcFailedCompTask] Thread[281462595072352] recompile single op[Unpack7] failed[FUNC:ProcessAllFailedCompileTasks][FILE:tbe_op_store_adapter.cc][LINE:961][THREAD:271372]
        [SubGraphOpt][Compile][ParalCompOp] Thread[281462595072352] process fail task failed[FUNC:ParallelCompileOp][FILE:tbe_op_store_adapter.cc][LINE:1009][THREAD:271372]
        [SubGraphOpt][Compile][CompOpOnly] CompileOp failed.[FUNC:CompileOpOnly][FILE:op_compiler.cc][LINE:1112][THREAD:271372]
        [GraphOpt][FusedGraph][RunCompile] Failed to compile graph with compiler Normal mode Op Compiler[FUNC:SubGraphCompile][FILE:fe_graph_optimizer.cc][LINE:1420][THREAD:271372]
        Call OptimizeFusedGraph failed, ret:-1, engine_name:AIcoreEngine, graph_name:partition0_rank1_new_sub_graph1[FUNC:OptimizeSubGraph][FILE:graph_optimize.cc][LINE:119][THREAD:271372]
        subgraph 0 optimize failed[FUNC:OptimizeSubGraphWithMultiThreads][FILE:graph_manager.cc][LINE:1012][THREAD:270734]
        build graph failed, graph id:6, ret:-1[FUNC:BuildModelWithGraphId][FILE:ge_generator.cc][LINE:1608][THREAD:270734]
        [Build][SingleOpModel]call ge interface generator.BuildSingleOpModel failed. ge result = 4294967295[FUNC:ReportCallError][FILE:log_inner.cpp][LINE:161][THREAD:270734]
        [Build][Op]Fail to build op model[FUNC:ReportInnerError][FILE:log_inner.cpp][LINE:145][THREAD:270734]
        build op model failed, result = 500002[FUNC:ReportInnerError][FILE:log_inner.cpp][LINE:145][THREAD:270734]

(Please search "CANN Common Error Analysis" at https://www.mindspore.cn for error code description)

----------------------------------------------------
- C++ Call Stack: (For framework developers)
----------------------------------------------------
mindspore/ccsrc/transform/acl_ir/acl_utils.cc:379 Run

[ERROR] DEVICE(270214,fffdd3fff160,python):2024-09-19-01:46:09.655.806 [mindspore/ccsrc/plugin/device/ascend/hal/hardware/ge_kernel_executor.cc:1156] LaunchKernel] Launch kernel failed, kernel full name: Bprop/gradStack/Unstack-op0
[ERROR] RUNTIME_FRAMEWORK(270214,ffff992bd010,python):2024-09-19-01:46:09.700.558 [mindspore/ccsrc/runtime/graph_scheduler/actor/actor_common.cc:327] WaitRuntimePipelineFinish] Wait runtime pipeline finish and an error occurred: 
----------------------------------------------------
- Kernel error:
----------------------------------------------------
Launch kernel failed: Bprop/gradStack/Unstack-op0

----------------------------------------------------
- C++ Call Stack: (For framework developers)
----------------------------------------------------
mindspore/ccsrc/runtime/graph_scheduler/actor/kernel_actor.cc:917 ExecuteLaunchKernelTask

环境：

• mindtorch 最新的master分支
• mindspore 2.3.1
• Cann：Ascend-cann-kernels-910_8.0.RC2_linux、Ascend-cann-toolkit_8.0.RC2_linux-aarch64
• NPU：910A
• Python： 3.10.14
• gcc：7.3.0