the result of onnx model validate on target has two different output, why?

cxf · ‎2025-07-29

hi

I have validate on the STM32N6 board by onnx model in STM32MX,but i have two different output ,like this:

m_outputs_1: (10, 3)/float64, min/max=[-282.578064, 481.370483], mean/std=[-2.619613, 298.301398], output m_outputs_2: (10, 1)/float64, min/max=[0.000000, 0.000000], mean/std=[0.000000, 0.000000], node_139 c_outputs_1: (10, 1, 1, 3)/float32, min/max=[-4.547175, 5.813805], mean/std=[0.008044, 4.321293], output c_outputs_2: (10, 1, 1, 1)/float32, min/max=[0.000800, 0.000800], mean/std=[0.000800, 0.000000], node_139

the m_outputs_1 is error, and c_outputs_1 is correct,why? what can lead to this problem?

please help me ,thanks very much!!!

I have upload the report.txt and onnx model in the attach.

cxf · ‎2025-08-13

Hi @SlothGrill

Thank you for your help very much!!!

Now,I modified the config.json in the \X-CUBE-AI\10.2.0\scripts\N6_scripts like this:

{

"compiler_type": "gcc",

"gdb_server_path": "E:/ST/STM32CubeIDE_1.18.1/STM32CubeIDE/plugins/com.st.stm32cube.ide.mcu.externaltools.stlink-gdb-server.win32_2.2.100.202501151542/tools/bin/",

"gcc_binary_path": "E:/ST/STM32CubeIDE_1.18.1/STM32CubeIDE/plugins/com.st.stm32cube.ide.mcu.externaltools.gnu-tools-for-stm32.13.3.rel1.win32_1.0.0.202411081344/tools/bin/",

"iar_binary_path": "",

"objcopy_binary_path": "E:/ST/STM32CubeIDE_1.18.1/STM32CubeIDE/plugins/com.st.stm32cube.ide.mcu.externaltools.gnu-tools-for-stm32.13.3.rel1.win32_1.0.0.202411081344/tools/bin/arm-none-eabi-objcopy.exe",

"cubeProgrammerCLI_binary_path": "E:/Program Files/STMicroelectronics/STM32Cube/STM32CubeProgrammer/bin/STM32_Programmer_CLI.exe",

"cubeide_path": "E:/ST/STM32CubeIDE_1.18.1/STM32CubeIDE"

}

and get the same two results by Julian E.'s five steps,like this:

By this way,I think I have validate on target successfully.

But when I go to STM32CubeMx,and click the validate on target,and check the Enabled,I still get the previous issues ，like this:

when I get the previous issues, I found that the config.json has been modified to like this:

{

"gdb_server_path": "E:/ST/STM32CubeIDE_1.18.1/STM32CubeIDE/plugins/com.st.stm32cube.ide.mcu.externaltools.stlink-gdb-server.win32_2.2.100.202501151542/tools/bin/",

"gcc_binary_path": "E:/ST/STM32CubeIDE_1.18.1/STM32CubeIDE/plugins/com.st.stm32cube.ide.mcu.externaltools.gnu-tools-for-stm32.13.3.rel1.win32_1.0.0.202411081344/tools/bin/",

"objcopy_binary_path": "E:/ST/STM32CubeIDE_1.18.1/STM32CubeIDE/plugins/com.st.stm32cube.ide.mcu.externaltools.gnu-tools-for-stm32.13.3.rel1.win32_1.0.0.202411081344/tools/bin/arm-none-eabi-objcopy.exe",

"cubeide_path": "",

"compiler_type": "gcc",

"cubeProgrammerCLI_binary_path": "E:/Program Files/STMicroelectronics/STM32Cube/STM32CubeProgrammer/bin/STM32_Programmer_CLI.exe",

"iar_binary_path": ""

}

As I have validate on target successfully by the first way(above-mentioned),

why it can generate this problem by use STM32CubeMX?

Is there any differences between two way?

Thanks!!!

I put the shell information into the attach.

SlothGrill · ‎2025-08-13

Hello
I'm not sure to fully understand, but there is definitely a problem with your last config.json: setting "cubeide_path" to "" is not similar to fully removing the line (in the way the config file is parsed).

Setting it to "", indicates to the tool that you expect to find cubeide in your current directory, then when trying to find gdb_server in your current directory, everything fails.

So, if this is x-cube-ai who erased your config.json, then this is problematic :), but the main thing is that you now know how to perform a validation on target after editing your config file.

I don't really understand if there are still issues pending:

After editing the config.json, you manage to do a validation on target through X-Cube-AI
so, you can now perform validations on target with X-Cube-AI, right ?
Could you please "summarize" the "two ways" in a synthetic manner ?

Thanks

Cheers.

cxf · ‎2025-08-13

Hi @SlothGrill

Thank you for your help very much!!!

The two methods are like this:

First method:

Go to: C:\Users\<your_user>\STM32Cube\Repository\Packs\STMicroelectronics\X-CUBE-AI\10.2.0\scripts\N6_scripts\
Copy your model to the model folder (or directly in N6_script folder)
Run the generate command:
stedgeai generate -m your_model --target stm32n6 --st-neural-art
Run the n6_loader.py script with DK board plugged in dev mode:
python n6_loader.py
Run the validate on target command:
stedgeai validate -m your_model --target stm32n6 --mode target -d serial:921600

Second method:

1:go to stm32cubeMX,click the validate on target,and check the enabled,the click ok like this

The results is that I validate on target successfully by the first method,but get a problem by the second method.

I think the reason is that config.json has been modified to "cubeide_path": "".but i do not know what lead to this

phenomenon？

Thanks!

SlothGrill · ‎2025-08-18

Hello
Agreed, it's a bug on X-Cube-AI, then, as setting the path to "" will most likely lead to buggy flow.

I guess X-Cube-AI checks if cubeide has been installed on "standard" paths (i.e. not E:), and fallback to setting the config to "" when not found.

We will post a bug for that.

Thanks.

cxf · ‎2025-08-19

Hi @SlothGrill when I can get the latest X-Cube-AI which can solve this bug? As i want to validate on target by STM32CubeMX. Thanks very much!!!

SlothGrill · ‎2025-08-20

Hello,

X-Cube-AI gets updated when new versions of STEdgeAI are released. So there will be no available versions with a fix before any new STEdgeAI release (i.e. do not expect anything before some months...).

For now, if you really want to use X-Cube-AI, you can try to reinstall CubeIDE properly and see if it fixes the issues (eventually removing the old versions that may lie around on your PC may help). Or try to install it on standard target directories (eg. c:\ST\...)

Best regards.