Skip to main content
Question

Avialable compiled models for ARM64 devices (Jetson Orin NX)

  • November 24, 2025
  • 6 replies
  • 30 views

Hi guys,

After a small break, I was able to get the Metis M2 running on my Jetson Orin NX.

However, is it possible that when using ‘axdownload’, the models are not precompiled? 
I tried to download ‘mobilenetv2-imagenet’, and got this error when trying to load it in. So, my guess is, I incorrectly assumed that the downloaded folder included a precompiled one? Or is there maybe a list of models which do include a precompiled model?

Cheers (more info below),

Sebbe

Error

-----------------------

ERROR][zeModuleCreate]: Failed with error code: 0x70000004
[ERROR][zeModuleCreate]: Build log:
sh: 1: riscv64-unknown-elf-gcc: not found
Command failed (32512)

Failed to build module: 
sh: 1: riscv64-unknown-elf-gcc: not found
Command failed (32512)

[ERROR][zeKernelCreate]: Failed with error code: 2013265925
Error at zeKernelCreate(module_handle, &desc, &kernel_handle): kernel_create: 409: Exit with error code: 0x78000005 : ZE_RESULT_ERROR_INVALID_NULL_HANDLE
Failed to create handles
Failed to initialise AICoreKernel kernel_function
Failed to initialize V1Executor
Traceback (most recent call last):
  File "/root/.pyenv/versions/3.10.6/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/root/.pyenv/versions/3.10.6/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/workspace/inference/radtest_metis.py", line 165, in <module>
    main()
  File "/workspace/inference/radtest_metis.py", line 161, in main
    run_det(args)
  File "/workspace/inference/radtest_metis.py", line 48, in run_det
    ctx.load()
  File "/workspace/inference/metis_runner.py", line 104, in load
    self._instance = self._conn.load_model_instance(
  File "/opt/voyager-sdk/venv/lib/python3.10/site-packages/axelera/runtime/objects.py", line 496, in load_model_instance
    _raise_error(self.context)
  File "/opt/voyager-sdk/venv/lib/python3.10/site-packages/axelera/runtime/objects.py", line 83, in _raise_error
    raise exc(f"{err}: {msg}")
axelera.runtime.objects.InternalError: AXR_ERROR_INTERNAL_ERROR: Error at zeKernelCreate(module_handle, &desc, &kernel_handle): kernel_create: 409: Exit with error code: 0x78000005 : ZE_RESULT_ERROR_INVALID_NULL_HANDLE


How I load the model (I’m still getting used to how the model loading works, so might have created an error here?)

def load(self) -> None:
self._ctx = _pick_ctx()
 
# Load deployed model.json (capability-based)
self._model = _load_model(self._ctx, self.model_json)
 
# Tensor metadata (API differs across releases)
self._input_infos = _get_inputs(self._model)
self._output_infos = _get_outputs(self._model)
 
# Connect to device (None = auto)
# v1.4 signature: device_connect(device=None, num_sub_devices=1, **kwargs)
self._conn = self._ctx.device_connect(self.device, self.batch_size)
 
# Create runnable instance
# v1.4: Connection.load_model_instance(model, **kwargs)
self._instance = self._conn.load_model_instance(
self._model,
num_sub_devices=self.batch_size,
aipu_cores=self.aipu_cores,
)
 
print("[METIS] Imported runtime:", _AXR.__name__)
print("[METIS] Loaded:", self.model_json)
print(" Device :", self.device or "auto")
print(" AIPU cores :", self.aipu_cores)
for ii in self._input_infos:
print(" Input :", getattr(ii, "name", "?"),
"shape:", getattr(ii, "shape", "?"),
"dtype:", getattr(ii, "dtype", "?"))
for oi in self._output_infos:
print(" Output:", getattr(oi, "name", "?"),
"shape:", getattr(oi, "shape", "?"),
"dtype:", getattr(oi, "dtype", "?"))


 -----------------------------------------------------

Hope this provides enough context, cheers!

6 replies

  • Author
  • Cadet
  • November 25, 2025

[Update] Got it fixed! Was mostly things with wrong $PATH variables when trying to install voyager-sdk in a docker container. All scripts run perfectly fine now. 


Spanner
Axelera Team
Forum|alt.badge.img+2
  • Axelera Team
  • November 25, 2025

Ah, great work ​@sebbe ! I was literally just researching your first post, and it pinged to tell me there’d been a reply 😄 Great to hear it’s sorted.

What is it you’re working on?


riQchet
Cadet
  • Cadet
  • November 25, 2025

Hey there, 

first of all: awesome idea! I was planning to do exactly that combination with the upcoming Metis M.2 Max. Hoping it will still come before 2026 


  • Author
  • Cadet
  • November 26, 2025

@Spanner Mostly some personal projects and testing the applications. But getting the sdk to work in a docker container with a Jetson Orin NX was the first step ;);), so glad that that worked out! 


Spanner
Axelera Team
Forum|alt.badge.img+2
  • Axelera Team
  • November 27, 2025

Great stuff! I don’t know if it helps in your specific use case(s) ​@sebbe, but did you see that the new SDK release (1.5) now supports Ubuntu 24.04 natively, with no Docker container required?


  • Author
  • Cadet
  • November 27, 2025

Hi! Yes! Well, I’m mostly working with Docker containers for a project, so I needed it to work with Docker!