Can't upload my model in any way :-(

Support
1

Hello,

I am trying to upload my software and I can’t make it work, unfortunately.

If I try to upload the Docker container, I run this command:

tira-run --input-dataset generative-ai-authorship-verification-panclef-2026/pan26-generative-ai-detection-smoke-test-20260330-training --image final_submission-server

And I get an error, which ends in this:

docker.errors.NotFound: 404 Client Error for http+docker://localhost/v1.54/containers/create?platform=linux%2Famd64: Not Found (“image with reference final_submission-server was found but its platform (linux/arm64) does not match the specified platform (linux/amd64)”)

I tried to do a code submission, and it does not work either… I just can’t make it work.

I have a docker that runs well in my computer. I can load it and run the program inside the Docker image.

Thanks in advance,

Manex

2

Dear Manex,

Thanks for reaching out!
The problem above seems to be that the image was build for arm64 but the required platform is amd64. You can add --platform linux/amd64 during your docker build, that should resolve the problem.

You can also invite me to your github repository, I can submit the solution. My account on Github is mam10eks.

Best regards,

Maik

3

I tried that solution, but it just didn’t work for me. I am not the best Docker user, though…

I just added you to my Github repo.

I hope it works. The dockerfile contains the command that should be run to train the model and test it with an input file, like this:

python /app/manex_approach_v1.py $inputDataset $outputDirectory

Thanks a lot, Mark :slight_smile:

4

Thanks, I look into it, and will report back in case I have questions :slight_smile:

Best regards,
Maik

1 Like
5

Dear @manexagirrezabal,

I have a short question, I think the software trains many models, is this possible? That would not be a valid submission to the task at the moment.

Submissions to the task should already be trained, as the interface of the task asks for submissions that are already trained and only do the inference/predictions.

Could you please adjust the software accordingly?

E.g., python /app/manex_approach_v1.py INPUT OUTPUT should only make predictions with some finalized approach, no training.

Thanks in advance!

Best regards,

Maik

6

Oh! I see. Let me adjust it and I will let you know when I change it :-).

Thanks!

Manex

1 Like
7

Cool, thanks!

8

Hello Mark,

I just updated the Github repository with a program that just does inference.

I also include how to run the program in the README file from the Github repo.

It is a rather simple method, so it should not take too long (~1 minute in my computer, and mostly because of POS tagging).

Thank you very much,

Manex

9

Cool, thanks, this is very good!

The software did fail because it because it tried to access the test labels:

Traceback (most recent call last):
  File "/app/manex_approach_v1_inferenceonly.py", line 152, in <module>
    testdatay = [line['label'] for line in lines]
  File "/app/manex_approach_v1_inferenceonly.py", line 152, in <listcomp>
    testdatay = [line['label'] for line in lines]
KeyError: 'label'

The labels are not available during test.

I made this minor change to not access the labels during inference: https://github.com/manexagirrezabal/pan2026_GAI/commit/fa550bd0266da757a02877dfd9c32d167843c525.

With this change everything seems to be fine and I submitted the software to TIRA (it got the name adventurous-drawbridge, you can change the name). I also started the software on all datasets, and everything looks good, it produced valid results and valid evaluations. (The evaluation scores on the test sets are blinded until after the deadline.)

So everything looks good, thanks :slight_smile:

Best regards,

Maik

10

Ah! Thanks a lot, Mark!!! :slight_smile:

I’m looking forward t see what happens :slight_smile:

Kind Regards,

Manex

1 Like