Skip to content

v0.1.41 - Update Platform API contract (#66)

Choose a tag to compare

View all tags
@github-actions github-actions released this 15 Sep 00:41
· 29 commits to main since this release
v0.1.41
7eb11b9
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

🌟 Summary

Ultralytics Platform API v0.1.41 adds reliable training callbacks and checkpoint uploads, with a regenerated, type-safe Python SDK for easier integration with the Ultralytics Platform. 🚀

📊 Key Changes

  • Training checkpoint uploads 📦

    • Added an endpoint to request secure upload URLs for model checkpoints.
    • Supports associating uploads with a specific model and training run.
    • Includes typed responses containing the upload URL, storage path, and model ID.
    • Available through both synchronous and asynchronous Python SDK methods.

  • Training event callbacks 📈

    • Added support for sending training lifecycle events, including:
      • Training started and progress updates
      • Epoch completion
      • System metrics
      • Console output
      • Checkpoint saved
      • Training completed or failed
    • Responses can report model and run identifiers, cancellation requests, warnings, and stale worker instances.

  • Regenerated Python SDK 🐍

    • Updated ultralytics-platform from version 0.1.40 to 0.1.41.
    • Added typed upload_checkpoint and metrics methods for both regular and asynchronous clients.
    • Added response type definitions to improve editor support and validation.

  • API contract and release process improvements ✅

    • Synchronized the OpenAPI specification with the live Platform API.
    • Added deterministic SDK generation and checksum validation to help ensure the published SDK matches the approved API contract.

🎯 Purpose & Impact

  • More robust remote training workflows: Training workers can report progress, metrics, failures, and completion status directly to the Platform.
  • Improved checkpoint handling: Checkpoints can be uploaded securely and linked to the correct training run, making model recovery and persistence more dependable.
  • Better automation and monitoring: Developers can build services that react to training events, detect failures, and honor cancellation requests.
  • Easier Python integration: Typed synchronous and asynchronous SDK methods reduce manual API work and make integrations clearer and less error-prone.
  • No new model architecture changes: This release focuses on Platform API and SDK capabilities rather than changes to YOLO model performance or structure.

What's Changed

Full Changelog: v0.1.40...v0.1.41

Contributors

UltralyticsAssistant
Assets 3
Loading