Changelog
Latest features, improvements, and updates to the Langtrain platform.
- v2.5.0
- Mar 2026
v2.5.0 (March 4, 2026)
New Features
- Knowledge Graph: Added a 3D interactive graph visualization that dynamically maps workspace entities and semantic relationships
- Brand Aesthetics: Rolled out the 'Light Luxury' design system across all UI primitives, including refined circular shapes, premium color palettes, and Osmo-inspired navigation components
- Admin Controls: Improved superadmin data table styling for dark mode and automated workspace creation for new signups
- Resolved 500 errors and Next.js async parameter issues on Knowledge API endpoints
- Streamlined the public landing page by removing redundant Stats and Social Proof sections
- Stabilized authentication flow and cross-subdomain session handling
- Verified and strengthened Admin API stability and error logging
v2.4.0 (January 24, 2026)
New Features
- Micro-HUDs: Global rollout with context-aware metrics and randomized positioning
- Pipeline Visuals: Enhanced 'Beam' animation and sticky interactions
- Refined UX: Improved CTA layouts and annotations
- Fixed Pipeline HUD overlay issues
- Resolved CTA annotation visibility and alignment
- Mobile responsiveness improvements
v2.3.0 (December 31, 2024)
New Features
- Agent Builder: Visual workflow builder for AI agents
- DeepSeek V3 models (70B, 236B) now available
- Phi-4 (14B) added to model hub
- OpenAI-compatible inference endpoints
- Streaming responses for chat completions
- 40% faster LoRA training on H100 GPUs
- Improved documentation with better navigation
- Enhanced model search and filtering
v2.2.0 (December 15, 2024)
New Features
- 100+ open-source models now available
- New providers: DeepSeek, Qwen 2.5, Gemma 2
- Dynamic model registry with real-time updates
- Webhook notifications for training events
- New
/api/v1/chat/completionsendpoint (OpenAI-compatible) - Batch inference support with
/api/v1/chat/completions/batch
v2.1.0 (November 2024)
New Features
- Llama 3.3 support (8B, 70B, 405B)
- Distributed training with DeepSpeed ZeRO-3
- Real-time training metrics dashboard
- GGUF/ONNX model export
- 50% faster LoRA fine-tuning
- Memory optimization for 70B+ models
- Improved error messages and logging
v2.0.0 (October 2024)
Major Release
- Complete platform redesign
- New Python SDK (
pip install langtrain-ai) - Multi-cloud deployment (AWS, GCP, Azure)
- Enterprise SSO support
- API v1 deprecated—migrate to v2 endpoints
v1.9.0 (September 2024)
Features
- QLoRA 4-bit training support
- Custom evaluation metrics
- Model comparison tools
- Fixed memory leak in long training jobs
- Resolved auth issues in Python SDK