Help Needed with Docker Submission for Task 2: Classification Spot-Check

Touché
1

Hello everyone,

I’m trying to submit my solution for Task 2: Classification Ads of ADS-in-RAG 2025 via the Docker interface, but I’m running into a persistent issue:

  1. Docker image & inference script
  • I’ve built an inference-only image that bundles my model.pkl, a script.py that auto-discovers input.jsonl, runs predict(), and writes predictions.jsonl to $outputDir.
  • Locally, when I manually docker run against the spot-check file (mounted as /data/input.jsonl), it produces the expected predictions.jsonl.
  1. tira-run push
  • Using

lua

CopyEdit

tira-run \
  --input-dataset=advertisement-in-retrieval-augmented-generation-2025/ads-in-rag-task-2-classification-spot-check \
  --image=<my-registry-path>/tira-submission:0.0.1 \
  --push true \
  --tira-client-token=… \
  --tira-docker-registry-user=… \
  --tira-docker-registry-token=…
  • It either:
    • Fails locally with “Your software did not produce any output” (even though Docker Desktop is running), or
    • Skips local test and then never shows the image under the Docker Submissions tab.
  1. What I’ve checked
  • Verified Docker is running and tira-cli login/verify-installation both succeed.
  • Ensured the image is properly tagged and pushed to the registry I’m authorized on.
  • Confirmed the spot-check dataset ID exactly matches what the CLI prints as available.

Questions

  • Dataset path: Is ads-in-rag-task-2-classification-spot-check the correct identifier for the spot-check in Task 2?
  • Push workflow: What’s the minimal tira-run invocation that both runs the container locally (to generate predictions.jsonl) and registers the image?
  • Registry creds: Are there any additional flags or user names I should include to ensure the image lands under my team’s Docker submissions?

Any pointers on what I might be missing or common pitfalls here would be hugely appreciated—especially since the deadline is approaching. Thank you!

2

Hi Tarek,

Thanks for reaching out!

From your description, everything looks good, I think it might only be a minor thing.
Could you please send me a link to your code (potentially in a private message), so that I can help you to finalize the submission?

Thanks and best regards,

Maik