Enable real classifiers in local dev by default

[mihow]

Summary

Local development now runs real models: the main Antenna docker compose stack starts the example processing service as ml_backend instead of the minimal stub, so jobs in a local stack produce real detections and classifications. Once #992 lands, that includes the global moth species classifier. The models download on the first job and are kept in a volume, so rebuilding the container does not download them again. CI is unchanged and keeps the fast, model-free minimal service.

List of Changes

Change (effect)	How	Notes
Local jobs run real models	Root docker-compose.yml: ml_backend builds and mounts processing_services/example	docker-compose.ci.yml keeps minimal
Models download once, not on every rebuild	New ml_backend_model_cache volume on /root/.cache (Hugging Face and torch caches)	Addresses the review suggestion about cache volumes
Developers know what to expect	README: which service each stack runs, the first-download size, how to start minimal next to the main stack

Testing

On a local stack built from this branch, with only ml_backend started: the first request to the zero-shot detector pipeline took 60 s and downloaded 3.6 GB of models into the cache volume; after recreating the container, the same request took 13.7 s with no download. Both returned 6 detections.

Still open (from the original description)

• Global moth model after the zero-shot detector: Add general detector + global moth model to example pipeline #992.
• The constant classifier returning a fixed box for testing, and default pipelines for new projects: not changed here.

To try it: docker compose build ml_backend && docker compose up -d, then use Register Pipelines on the project's processing services page.

Related: #992 (global moths in example), #1252 (the minimal service as a pull-mode worker).

🤖 Generated with Claude Code

https://claude.ai/code/session_01TNNRg9NDr6opFZgK3nxx5C

[netlify]

✅ Deploy Preview for antenna-preview canceled.

Name	Link
🔨 Latest commit	6806786
🔍 Latest deploy log	https://app.netlify.com/projects/antenna-preview/deploys/6aceb1225a9599c61c84eb1d

[coderabbitai]

📝 Walkthrough

The docker-compose.yml file was updated to change the ml_backend service's build context and volume mount paths from ./processing_services/minimal to ./processing_services/example.

Changes

Cohort / File(s)	Summary
Docker Compose Configuration docker-compose.yml	Updated ml_backend service build context and mounted volume path from minimal to example subdirectory

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~2 minutes

• Straightforward path substitution with no logic changes
• Single file affected with repetitive pattern (same update applied twice)

Pre-merge checks: 3

Passed checks: 3 passed

Check name	Status	Explanation
Title check		The title clearly and specifically describes the main change: enabling real classifiers in the local development environment by default, which aligns with the docker-compose.yml changes updating the M…
Docstring Coverage		No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Description check		The description is complete enough for review. It explains the purpose, lists the file and behavior changes, documents testing results and commands, identifies related issues, notes open work, and sta…

✨ Finishing touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch feat/default-processing-service

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[Copilot]

Pull Request Overview

This PR switches the default ML backend for local development from the minimal processing service to the example processing service, enabling real ML classifiers (transformers, PyTorch) instead of the basic mock implementations.

Key changes:

• Updates the ml_backend service in docker-compose.yml to use the example processing service with real ML models
• The CI environment continues to use the minimal service for faster testing

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The example service uses ML models (transformers, PyTorch) that download models on first use. Consider adding cache volumes to avoid re-downloading models on container restarts:

volumes:
  - ./processing_services/example/:/app
  - ./processing_services/example/huggingface_cache:/root/.cache/huggingface
  - ./processing_services/example/pytorch_cache:/root/.cache/torch

This is similar to the configuration in processing_services/example/docker-compose.yml and would improve startup performance after the initial model download.

[netlify]

✅ Deploy Preview for antenna-ssec canceled.

Name	Link
🔨 Latest commit	6806786
🔍 Latest deploy log	https://app.netlify.com/projects/antenna-ssec/deploys/6aceb122254a0a9d13359c3c