Voyager SDK

Hello, I would like to use the Metis AIPU to perform micro benchmarks, so very simple operations on which i would have the maximum control. Therefore i was wondering if there is a way to read and write for example in the cores’ L1 memory “manually”? For the moment i couldn’t find any of these, the closest being the DmaBufAllocator but i would guess it is not directly related to the core itself? Thank you very much 

Hey everyone,I'm trying to boot Ubuntu 22.04 from a SATA SSD on a Metis Compute Board (RK3588) running the Axelera BSP.Setup Board: Metis Compute Board (RK3588) BSP: Axelera Voyager Linux 1.2.1 (Mender-managed A/B boot) Goal: Boot Ubuntu 22.04 from SATA SSD (/dev/sda1) What I've Done Built a full Ubuntu 22.04 arm64 rootfs on the SATA SSD Copied BSP kernel modules (6.1.148-rockchip-standard) into the rootfs Set correct fstab using UUID Verified systemd and libc are present and valid Used mender_post_setup_commands to override mender_kernel_root to /dev/sda1 after Mender setup The ProblemMender overwrites mender_kernel_root every boot with the eMMC partition.Even after overriding it, the kernel attempts to boot from SATA but fails and then falls back to eMMC via mender_try_to_recover, leaving me stuck at a splash screen.I currently do not have UART access, so I can’t see kernel logs during the failed boot attempt.Additional NoteIf the system gets stuck at the splash screen pe

Hello, I would like to run Voxtral-Mini for Speech-To-Text (STT) and qwen3 tts 0.6B for Text-To-Speech. (TTS)Qwen3 tts would probably fit into a M.2 Metis AI board, but voxtral is a 4B model, therefore it would probably need a M.2 Metia AI Max.I have a few questions to start:I find a plenty of examples running Metis with images, but audio is almost not existent. Do you support it? If yes, can you point me to the specific documentation, please? If I use the 16GB of RAM of M.2 Metis AI Max, can I have two models loaded and running in parallel on the board? Voice applications are very sensitive to latency, being able to load two models and keeping them ready to work would keep latency low.Thank you, Ottavio

Hello, I want to share that I think I found a bug in https://software.axelera.ai/ui/native/axelera-apt-source/axelera-runtime/ubuntu24/axelera-runtime_1.5.3-1_arm64.deb (and potentially other packages with the same version). pkg-config returns the wrong location for the headers of axruntime:(venv) ottavio@instance-20220410-1128:~/Projects/voyager-sdk$ pkg-config --cflags axruntime-I/opt/axelera/runtime-1.5.3-1/(venv) ottavio@instance-20220410-1128:~/Projects/voyager-sdk$but it shoud be (venv) ottavio@instance-20220410-1128:~/Projects/voyager-sdk$ pkg-config --cflags axruntime-I/opt/axelera/runtime-1.5.3-1/include(venv) ottavio@instance-20220410-1128:~/Projects/voyager-sdk$I am sure of these, because, without the final include directory, all the examples doing#include <axruntime/axruntime.h>will fail. Even if you change the code to#include <include/axruntime/axruntime.h>it will fail because it is trying to import axruntime.hpp which does not have the missing directory in the

Hello,I am trying to understand how to add support for Metis to my embedded system, which is based on Buildroot.So far, this is what I have:I need to compile a kernel supporting DMA-BUF (done). I need to compile the kernel drivers (this is what I am working on now). I need to have the axelera_runtime package in my system, because the final application will depend on it.The last point is still unclear to me. I expected to find it in the Voyager SDK repository, but it is not there. The SDK depends on runtime files installed by the installation script, which downloads them from https://software.axelera.ai/ui/packages.Could you please clarify whether the source code for axelera_runtime is available? Or is it closed source?Thank you, Ottavio

Hi Team,I’m working with the Voyager SDK (using docker container) and trying to build the axstreamer and gstaxstreamer components. The source files examples reference:#include <axruntime/axruntime.hpp> (examples/axruntime/axruntime_example.cpp)But I can’t find axruntime.hpp anywhere in the SDK.I’m planning to run inference using a webcam as the input on my mac, with a pipeline that includes ffmpeg and mediamtx for streaming.Is axruntime supposed to be part of the SDK repo, a separate dependency, or is there a script I missed?Thanks!Val.

Hi,I just finished setting up everything and now I’m trying to run an example pipeline using.\inference.py yolov7-640x480-coco cars_clip.mp4Building the model seems to work fine but when the inference starts, it either crashes after some images withDetecting... : 2%|█▋ | 37/1802 [00:01<00:39, 44.29frames/s][ERROR][axeCommandQueueExecuteCommandListsAsync]: Wait kernel failed with return code -1.terminate called after throwing an instance of 'std::runtime_error' what(): axr_run_model failed with Error at zeCommandQueueExecuteCommandLists(cmdqueue, n_cmdlists, cmdlists, nullptr): cmdqueue_run_cmdlists: 319: Exit with error code: 0x7FFFFFFE : ZE_RESULT_ERROR_UNKNOWNAborted (core dumped)or instantly withINFO : Could not exec vainfo: Command '['vainfo']' returned non-zero exit status 3.WARNING : Timeout for querying an inferenceERROR : InferenceStream terminated due to T

Hello, I try to use a YOLO model on 8 camera streams in parallel on the Metis PCIe card. 20-30 fps per camera stream is all I need. With the 548 fps end-to-end stated for YOLOv8s in https://axelera.ai/metis-aipu-benchmarks, it should be possible to reach ~68 fps per camera. As a small test I wrote a python script, which creates a inference stream with 8 videos as input:from axelera.app import config, display, inf_tracersfrom axelera.app.stream import create_inference_streamdef run(window, stream): for frame_result in stream: window.show(frame_result.image, frame_result.meta, frame_result.stream_id) fps = stream.get_all_metrics()['end_to_end_fps'] print(fps.value)def main(): tracers = inf_tracers.create_tracers('core_temp', 'end_to_end_fps', 'cpu_usage') stream = create_inference_stream( network="yolov5s-v7-coco", sources=[ str(config.env.framework / "media/traffic1_1080p.mp4"), str(config.env.framework / "media/traffic1_

Hello everyone,I am new to Axelera and working with a Metis device (4 GiB, PCIe, Voyager SDK). Device: metis-0:1:0 Memory: 4GiB Clock: 800MHz SDK: voyager-sdk  I am trying to deploy the InsightFace model for face recognition and would appreciate guidance on model conversion, supported formats, and deployment steps.Thank you in advance,Akash

Hello,I am facing an issue with a Metis Compute Board where any inference command (or any command using the accelerator) immediately fails with the error:ERROR: No devices found This happens regardless of the model or the command I run.Yesterday, I managed to temporarily recover by removing and reinstalling the Docker container with the Voyager SDK, after which inference started working again. I thought this was because I did not use the metis in a month so maybe it needed to be reinstalled.However, today the same error appears again, even though I did not change anything in my setup.I have already tried rebooting the board, but the problem persists. At the moment, no inference works at all, and reinstalling the container no longer fixes the issue.I would like to know if this is a known issue and if there is a recommended recovery procedure for this situation.Thanks in advance for your help.Best regards,Thibaut

Hello,I’m currently working with the Metis Compute Board and the Voyager SDK, and I had a question regarding development outside of the SDK.Is it possible to run a script like a custom inference or a similar script directly on the board without going through the Voyager SDK pipeline ?I’m mainly asking to better understand the level of low-level access available on the Metis (runtime, APIs, or tools outside the SDK).Thanks in advance for your help.Best regards,

Hello,Following the question I had on running a GEMM only model, I have settled for a model that has Id Conv → Flatten → Gemm . I was wondering if the Flatten operation is just rearranging some pointer values or if it is doing a memory operation (like copying the tensor or something like that)?Thank you very much,

Hi,I'm using NanoPC-T6. I followed the Https://support.axelera.ai/hc/en-us/articles/31859388491794-Bring-up-Voyager-SDK-in-NanoPC-T6 you sent me. However, when I proceed with the test, the test will not proceed with the following error.https://community.axelera.ai/voyager-sdk-2/axelera-runtime-tar-gz-256?postid=2749#post2749 Someone has had the same issue in this community and I need help if there was a fix. (venv) pi@pi-desktop:~/voyager-sdk$ AXELERA_CONFIGURE_BOARD=,20 ./inference.py yolov8l-coco-onnx ./media/traffic1_480p.mp4 --no-display2026-01-02 14:31:47.360506524 [W:onnxruntime:Default, device_discovery.cc:164 DiscoverDevicesForPlatform] GPU device discovery failed: device_discovery.cc:89 ReadFileContents Failed to open file: "/sys/class/drm/card1/device/vendor"Failed to create device manager: libaxruntime.so: axr_create_context: libaxruntime.so: cannot open shared object file: No such file or directoryUnsupported tracer: core_temp: valid tracers are: cpu_usage, end_to_end_fps,

Bonjour,Je n’arrive pas à récupérer le fichier `axelera-runtime.tar.gz`. Le script `install.sh` affiche “Runtime Installed OK” mais ne télécharge ni n’extrait quoi que ce soit.Aucune URL n’apparaît dans les logs et `libaxruntime.so` est absent de mon système.Merci de me fournir un lien de téléchargement direct pour installer le runtime manuellement.Cordialement,   

Similar to https://community.axelera.ai/voyager-sdk-2/run-simple-gemm-model-1127, I got IndexError when converting a Gemm-only model via “compile -i model.onnx -o model --overwrite”. However, I don’t have Flatten layer. Could you help on checking what’s going wrong? The model I use and logs are attached below, and here are my system information:System: Ubuntu 22.04 x86_64 SDK: version 1.5.2 cloned from https://github.com/axelera-ai-hub/voyager-sdk

​@Spanner  what i get on this code? by launching “axllm.py my.yaml”try: network_yaml_info = yaml_parser.get_network_yaml_info( include_collections=['llm_local', 'llm_cards', 'llm_zoo'] ) parser = config.create_llm_argparser( network_yaml_info, description='Perform LLM inference on an Axelera platform' )How to configure the yaml file what’s inside? can i specify like  llm_local = ip:portllm_cards = /dev/metis0llm_zoo = https://huggingface.com/my/llm.zip…………Orllm_zoo: - url: ${HUG_RELEASE_URL}/llm/myhillarius.zip - url: ${HUG_RELEASE_URL}/llm/codellama:8B.zip - url: ${HUG_RELEASE_URL}/llm/r1-1776:70B.zipBy duck tapeing models to my yaml

r1-1776:70Bcodellamacodegemmapost follow ups by other users

Me and a few other folks would be very interested in having Ollama support the AIPUs. There is a wide range of different inference models on https://ollama.com/. From what I understand these would need to get recompiled to run on the AIPU. I could only find a handful of models in the model zoo and no information about how to recompile any given model. Can you provide instructions on how to recompile other models available online?

Quick update on Voyager SDK 1.5.2, which was just rolled out and sorts out a few issues from 1.5.1 while adding some quality of life improvements.What's fixed:The main focus here was stability. This version sorts the build failures that were hitting ImageNet networks when using HINT instructions, fixed NormaliseCL so it properly handles inputs that aren't 4-channel, and stopped those segfaults in YOLO decoders when they got an unexpected number of input tensors.New bits:Runtime can now execute AXM files directly, which should make deployment a bit more straightforward. It adds selectable network protocol for RTSP sources in inference.py (gives you more flexibility with inputs), and there are new create_inference_net overloads so existing code compiles without needing a context parameter.The docs have been updated to reflect the AXM execution changes, plus general tidy-ups across tutorials and references.That's it! Nothing earth-shattering, just the solid incremental improvements we lov

Hi!First of all, thank you very much for your work — the Axelera tools are really impressive and perform exceptionally well!I’m currently working on deploying LLMs on the METIS PCIe accelerator using the Voyager SDK v1.5.1.  From my understanding, the LLMs currently available are already precompiled, and we simply download the compiled model files.Would it be possible to access the deployment or compilation process so that we can deploy our own fine-tuned LLMs — even if they share the same architecture as the models already available in the model zoo? 

I get this error when I try to build examples:venv) aetina@aetina:/axelera/voyager-sdk-release-v1.5$ make examplesmake -C operatorsmake[1]: Entering directory '/mnt/microsd/axelera/voyager-sdk-release-v1.5/operators'make[1]: Nothing to be done for 'gst_ops_install'.make[1]: Leaving directory '/mnt/microsd/axelera/voyager-sdk-release-v1.5/operators'make -C examplesmake[1]: Entering directory '/mnt/microsd/axelera/voyager-sdk-release-v1.5/examples'ninja -C Release installninja: Entering directory `Release'[1/7] Building CXX object axinferencenet/CMakeFiles/axinferencenet_DPIR.dir/axinferencenet_DPIR.cpp.oFAILED: axinferencenet/CMakeFiles/axinferencenet_DPIR.dir/axinferencenet_DPIR.cpp.o /usr/bin/c++   -O3 -DNDEBUG -std=gnu++20 -fPIE -MD -MT axinferencenet/CMakeFiles/axinferencenet_DPIR.dir/axinferencenet_DPIR.cpp.o -MF axinferencenet/CMakeFiles/axinferencenet_DPIR.dir/axinferencenet_DPIR.cpp.o.d -o axinferencenet/CMakeFiles/axinferencenet_DPIR.dir/axinferencenet_DPIR.cpp.o -c /mnt/micros

After months of joint development, Ultralytics YOLO models now run seamlessly on Axelera AI hardware. So that’s native support through the Voyager SDK, including their new models. What does this mean for us?🔹 Energy-efficient, high-performance inference at the edge🔹 One-step export & deployment of Ultralytics models🔹 Proven performance across vision tasks, from detection to pose estimation And the cool part:🎯 8K Video Processing with Intelligent Tiling: Tiny objects, no detail lost🧠 Multiple Models on One Chip: 4 YOLO models, 1 Metis🏌️ Golf Swing Analysis: Real-time feedback using Ultralytics pose models There’s more info just gone live over in the Ultralytics docs, and you can check out some demos of it in action by ​@Steven Hunsche below! 

Hi, Spanner-san. I was able to get the camera demo working by applying the fix suggested by FUKU55-san in another thread (--disable-opencl). (x) # ./inference.py resnet50-imagenet-onnx usb:0 --show-stats (o) # ./inference.py resnet50-imagenet-onnx usb:0 --show-stats --disable-opencl It seems that the integration between GStreamer and OpenCL is unstable in SDK 1.5.1.

I’ve already redeployed the model:./inference.py yolo11m-custom dataset --show-device-fps --no-display --pipe gstarm_release_ver: g13p0-01eac0, rk_so_ver: 9INFO    : Using dataset valWARNING : Using deprecated heuristic transpose detection. Redeploy the model to include explicit output infoSegmentation fault