Changelog

Major Update

Features
- Added support for Mistral 7B and CodeLlama models
- New distributed training with DeepSpeed integration
- Enhanced monitoring dashboard with real-time metrics

Improvements
- 50% faster LoRA fine-tuning performance

Fixes
- Fixed memory leak in long-running training jobs

Patch Update

Fixes
- Resolved authentication issues with Python SDK
- Fixed dataset upload for large files (>1GB)

Improvements
- Improved error messages for failed deployments

Major Release

Features
- Complete UI redesign with improved user experience
- New JavaScript/TypeScript SDK
- Multi-cloud deployment support (AWS, GCP, Azure)
- Advanced security features and SOC 2 compliance

Breaking Changes
- API v1 deprecated, please migrate to API v2

Patch Update

Features
- Added support for custom evaluation metrics

Improvements
- Reduced training startup time by 30%

Fixes
- Fixed issue with gradient accumulation in multi-GPU setups

Minor Update

Features
- QLoRA support for memory-efficient fine-tuning
- New model comparison tools

Improvements
- Enhanced dataset validation and preprocessing