GPU worker: No score output

I am running a self-hosted GPU worker connected to Codabench.org. It used to work and run submissions some months ago. Now, the worker successfully:

• Starts correctly
• Connects to RabbitMQ
• Receives submissions
• Downloads the submission metadata

However, before the ingestion container is started, the worker fails when attempting to update the submission status to "Preparing" through the Codabench API.

The API returns an HTTP 500 error, causing the worker to abort with:
compute_worker.SubmissionException: Failure updating submission data.

Worker Configuration
Worker image: codalab/competitions-v2-compute-worker:gpu1.1
docker-compose.yml
networks: worker_network: driver: bridge driver_opts: com.docker.network.bridge.enable_icc: "false" com.docker.network.bridge.enable_ip_masquerade: "true" services: worker: image: codalab/competitions-v2-compute-worker:gpu1.1 container_name: compute_worker volumes: - /codabench:/codabench - /var/run/docker.sock:/var/run/docker.sock env_file: - .env restart: unless-stopped logging: options: max-size: 50m max-file: 3 runtime: nvidia deploy: resources: reservations: devices: - driver: nvidia device_ids: ["0"] capabilities: [gpu] security_opt: - no-new-privileges:true networks: - worker_network

Relevant Logs

Worker starts normally:

compute-worker ready.
Received task: compute_worker_run[76b24fc3-7395-4acb-8c73-c89aff6a6eba]

The failure occurs immediately when updating the submission:

Updating submission @
https://www.codabench.org/api/submissions/773803/

with data = {
  'status': 'Preparing'
}

The response body is the Codabench “500 Internal Server Error” HTML page:
<p>Something went wrong on our end.</p>

The worker then attempts to mark the submission as failed:

Updating submission @
https://www.codabench.org/api/submissions/773803/
with data = {
  'status': 'Failed',
  'status_details': 'Failure updating submission data.'
}

but receives the same HTTP 500 error.

### Questions

1. Is there a known issue affecting submission status updates through the API?
2. Is the gpu1.1 worker image still compatible with the current Codabench production deployment?
3. Are there any recent API changes that require a newer worker version?
4. Is there additional server-side logging available for submission 773803 that could help diagnose the 500 error?

You may also want to mention that you previously tested:
codalab/competitions-v2-compute-worker:gpu1.4
New failure
The actual failure is now:

/codabench/tmp88f_wrtb/program not found, no program to execute
/codabench/tmp88f_wrtb/ingestion_program not found, no program to execute

followed by:

'ingestion_program_data':
https://.../ingestion_program.zip