← Back to TensorSync
Changelog
All notable changes to TensorSync are documented here. We follow Semantic Versioning.
Bug Fixes
- Fixed race condition in concurrent tensor uploads
- Resolved memory leak in long-running sync sessions
- Fixed incorrect delta calculation for sparse tensors
New Features
- Added support for PyTorch 2.3 distributed checkpoints
- New
--watch flag for automatic sync on file changes
- Introduced checkpoint versioning with rollback support
Improvements
- 30% faster delta compression for large tensors
- Improved error messages for authentication failures
- Reduced memory footprint by 25% in CLI
New Features
- WebSocket telemetry streaming for real-time metrics
- Support for custom compression algorithms
- Added
tensorsync telemetry CLI command
Breaking Changes
- API v1 endpoints deprecated, migrate to v2 (see migration guide)
- Minimum Python version increased to 3.9
New Features
- Edge caching for frequently accessed models
- Team collaboration features (shared repositories)
- SSO/SAML integration for Enterprise plan
Improvements
- 50% reduction in API latency (P95: 45ms → 22ms)
- Improved handling of network interruptions
New Features
- Rust SDK released (beta)
- Support for TensorFlow SavedModel format
- Automatic bandwidth throttling based on network conditions
Major Release
- Complete rewrite of synchronization engine in Go
- New gRPC-based API (replaced REST)
- 10x performance improvement for large-scale syncs
- Native support for distributed training frameworks
Subscribe to Updates
Stay informed about new releases and features: