Unable to install full Voyager SDK (only partial ~80 MB install, missing runtime and ax_runner)

Add :  I did the same thing but with my generated token and it still does the same thing, the folder is approximatively 80 MB. ( I followed the manipulation in the txt file on github)

Plz tell me if it’s normal and it will install on the metis automatically (in this case something is wrong with my metis) or if it’s a wrong manipulation that causes this weight for my sdk folder.

Ty

Hi Thibaut,

Thanks for the detailed info. I think what you’re seeing (80 MB SDK, missing ax_runner) could mean the installer couldn’t fetch binaries from Axelera’s private Artifactory.

Possibly the token isn’t being passed correctly?

Running this inside your Docker container should give us the full install:

export AXELERA_ARTIFACTORY_TOKEN=your_token_here
./install.sh --all

Let me know how it goes!

Nope…

Unfortunately, even after doing that inside my Ubuntu 22.04 Docker container, the SDK installation still ends up partial (~80 MB total) and I’m still missing the ax_runner module.
The installer finishes almost instantly with “Installation complete”, without downloading any binaries from Artifactory.

Here’s what I’ve already tried:

• Clean reinstall from a fresh Ubuntu 22.04 container (Docker)

• Installed sudo inside the container (so that the installer wouldn’t skip commands)

• Exported AXELERA_ARTIFACTORY_TOKEN correctly before running the installer

• Verified that Artifactory is accessible (I can list axelera-runtime and other packages with apt search)

• Tried both ./install.sh --all and ./install.sh --all --media --user ... --token ...

No matter what, the SDK remains around 80 MB and python3 -m ax_runner --help returns

No module named ax_runner

Installing directly on the Metis board doesn’t work either, since several commands used in install.sh aren’t available there (for example sudo, add-apt-repository, etc.).

Would it be possible to have a manual download link or an alternative installation method (for example a .tar.gz package or a full .deb bundle) so I can install the full SDK and its binaries offline or inside Docker?

Thanks again for your help,

Thibaut

It just clicked that you’re ultimately looking to run this on the Metis Compute Board, is that right? We have a guide on getting it running, which might be helpful: https://support.axelera.ai/hc/en-us/articles/26405136028434-Getting-Started-with-Metis-Compute-Board

Also, it runs Yocto, rather than Ubuntu, which could be a stumbling block here too. Let me know if this helps!

Yes, exactly — I’m trying to get the SDK running on my Metis Compute Board.

The issue is that my Metis no longer has any of the original Axelera files. Everything is gone — no setup_axelera_environment.sh, no start_axelera.py, no voyager-sdk.tar.gz, etc.

That’s why I was trying to install the SDK inside a Docker container on my PC (Ubuntu 22.04), and then compress it with: 

tar czf voyager-sdk.tar.gz /home/voyager/voyager-sdk

to transfer it later to the Metis board.

But even after installing with the token (using ./install.sh --all --user ... --token ...), the SDK inside Docker is still incomplete — only about 80 MB total, and ax_runner is missing.

So, since the Metis is empty and I can’t reinstall the SDK directly on it (the installation script fails due to missing packages and sudo), I’m looking for a way to restore or transfer the full environment back onto the Metis.

Is there perhaps a recovery image or a way to reinstall Voyager Linux + SDK on the board itself?

Or a solution to have the full sdk installed on my docker container ?

Thanks a lot for your help!
— Thibaut

Ah okay, I’m with you. Perhaps this might help - if you jump down to point 3.3. Anything there that helps with this?

https://support.axelera.ai/hc/article_attachments/29949067454738

Maybe ! I’ll test this right now thank you !

Awesome! Let me know how it goes 👍

It worked ! thanks a lot

Ah, sweet! So you’ve been able to get your Metis Compute Board back to square one, ready to start on the project? That’s great to know.

What’s your project, by the way? Curious to know what you’re building!

I’m just testing this board by using differents yolo models and codes that i use with other boards ! Out of curiosity !

Nice! Be interesting to hear which YOLO models you’ve been testing, how they performed, any adjustments you had to make - that sort of thing. It’s really handy for the whole community to hear about the different models people have been working with on Metis and Voyager 👍

… 

Still not complete haha my sdk is only 64 Mo :

ubuntu@antelao-3588:~$ du -sh voyager-sdk/
64M    voyager-sdk/

I still can’t use the sdk to load yolo models…

I create a new post to discuss it because i thought that it was a different error but it’s the same…

Plzzzz help 

Hi!
Like mentioned in the other post, ax_runner is not a component of Voyager SDK, seems like an AI assistant hallucinated that for you (it happens :)

Following the install guide and the quick start guide should get you running.

If you plan to keep using LLM assistants, a suggestion would be to download our entire repo as a text file and attach it to the LLM prompt as context. You can download the repo as an entire text file from tools like gitingest: https://gitingest.com/axelera-ai-hub/voyager-sdk