相同 cann环境下进行onnx 转换，在 310B 上执行推理报错

[983183947]

操作步骤

1. 我先在910芯片的 cann7+mindspore2.2+mindspore_lite2.2.14上执行 onnx -->mindir转换得到.mindir文件
2.我在200 DK 开发板上执行步骤 1 得到的.mindir文件并进行推理，但是报错

说明

1. 目前只能在 200 DK 的310B板子上进行转换，但是大部分模型都转不动，运行几个小时后会报错。
2. mindcv 下的模型（resnet50）可以转成功

报错文件：paddle2mindir.log

环境(310B)

环境（910）

代码

import numpy as np
import mindspore_lite as mslite

# init context, and set target is ascend.
context = mslite.Context()
context.target = ["Ascend"]
context.ascend.device_id = 0
# build model from file
MODEL_PATH = "./paddle_clas_cann7_m22.mindir"
model = mslite.Model()
model.build_from_file(MODEL_PATH, mslite.ModelType.MINDIR, context)

报错

(mindspore) -sh-5.1# python infer.py 
[ERROR] ME(2974,e7ffe882c020,python):2024-08-10-23:30:56.704.903 [mindspore/lite/src/extendrt/kernel/ascend/model/model_process.cc:586] Load] Call aclmdlLoadFromMem failed, ret = 545002
[ERROR] ME(2974,e7ffe882c020,python):2024-08-10-23:30:56.705.027 [mindspore/lite/src/extendrt/kernel/ascend/model/model_infer.cc:150] Load] Load model model failed.
[ERROR] ME(2974,e7ffe882c020,python):2024-08-10-23:30:56.705.082 [mindspore/lite/src/extendrt/kernel/ascend/src/custom_ascend_kernel.cc:147] Init] Load om data failed.
[ERROR] ME(2974,e7ffe882c020,python):2024-08-10-23:30:56.705.137 [mindspore/lite/src/extendrt/session/single_op_session.cc:199] BuildCustomAscendKernelImpl] kernel init failed CustomAscend
[ERROR] ME(2974,e7ffe882c020,python):2024-08-10-23:30:56.706.811 [mindspore/lite/src/extendrt/session/single_op_session.cc:221] BuildCustomAscendKernel] Build ascend kernel failed for node: custom_0
[ERROR] ME(2974,e7ffe882c020,python):2024-08-10-23:30:56.706.877 [mindspore/lite/src/extendrt/session/single_op_session.cc:306] CompileGraph] Failed to Build custom ascend kernel
[ERROR] ME(2974,e7ffe882c020,python):2024-08-10-23:30:56.706.933 [mindspore/lite/src/extendrt/cxx_api/model/model_impl.cc:417] BuildByBufferImpl] compile graph failed.
Traceback (most recent call last):
  File "/opt/tmp/infer.py", line 12, in <module>
    model.build_from_file(MODEL_PATH, mslite.ModelType.MINDIR, context)
  File "/root/anaconda3/envs/mindspore/lib/python3.9/site-packages/mindspore_lite/model.py", line 111, in wrapper
    return func(*args, **kwargs)
  File "/root/anaconda3/envs/mindspore/lib/python3.9/site-packages/mindspore_lite/model.py", line 262, in build_from_file
    raise RuntimeError(
RuntimeError: build_from_file failed! Error is Common error code.

[zhouyifeng888]

确认下你这边说的200DK应该是200I DK A2吧？就拿200I DK A2来说，有不少开发者反馈过，由于A2板子内存小（4GB规格的内存），会出现模型转换失败的情况，或者mslite直接加载mindir进行推理的过程，其中也有转换编译的过程，也需要消耗一定的内存，A2板子很可能内存不够，可以试一下设置一下交换内存swap再看看（比如设置8G或者16G）

[983183947]

我换了 orange pi有 24G 内存。也是一样的问题