Zum Inhalt

DeepExtension Release Notes

Version Overview

Version Release Date Core Theme Status
v2511 Nov 2025 Experience Optimization & Feature Enhancement 🆕 Latest
v2509 Sep 2025 Redefining Full Lifecycle Management for AI Models ✅ Stable
v2507 Jul 2025 Multimodal Vision Models Debut ✅ Stable
v2505 May 2025 Opening a New Era of Efficient AI Training ✅ Stable

🆕 Version 2511 (Latest Release)

Release Date: November 2025

✨ User Experience Optimization

🔄 Training Flow Improvements

  • Real-time Status Sync: Training, saving, and deployment tasks appear in the model list immediately upon submission, no longer requiring manual status checks.

  • Quick Error Handling: Displays error messages immediately upon training failure and provides an option to interrupt training.

📊 Enhanced Model Evaluation

  • Intelligent Error Handling: User-friendly error prompts with specific modification suggestions.

  • Partial Success Support: Allows viewing results from other samples even if some fail.

🎯 New Fine-tuning Examples

  • Qwen LLM Example: Practical training on biographical data.

  • Stable Diffusion Example: Practical guide for generating vintage avatar-style images.

🐛 Bug Fixes

  • Fixed abnormal display in the model evaluation preview feature.

  • Optimized training status detection mechanism.

  • Improved interface loading performance.

🚀 Version 2509

Release Date: September 2025

Core Upgrades

🔄 Intelligent Lifecycle Management

  • Non-linear Workflow: Supports freely designed model iteration paths.

  • Complete Version Control: Automatically records training parameters, data, and results.

  • Model Lineage Tracking: Clearly view model evolution history.

🛠️ Environment & Deployment

  • Custom Training Environments: Flexibly configure dependency libraries and compute resources.

  • Seamless Deployment: Supports standard platforms like Ollama.

  • Containerization Support: Pre-configured Stable Diffusion 3.5 Medium environment.

🎨 Expanded Model Types

  • Image Generation Models: Based on Stable Diffusion technology, supports text-to-image and image-to-image.

  • Multimodal Models: Full lifecycle management on a unified platform.

💼 Enterprise Features

  • Multi-role team collaboration.
  • Enterprise compliance support.

👁️ Version 2507

Release Date: July 2025

Major Updates

📸 Multimodal Vision Models

  • Full Pipeline Support: Fine-tuning, inference, and deployment for vision models like Qwen-VL, Llama-Vision.

  • Dataset Visualization: Multi-image configuration, native Bounding Box visualization.

  • Visual Supervised Fine-Tuning: Launch fine-tuning tasks with custom data.

🎯 DeepPrompt Enhancement

  • Full support for image-to-text scenarios.
  • Single/Multi-image input support.
  • Custom task instructions.

⚙️ Training Flexibility

  • Fully Customizable Parameters: Supports string, int, float, bool types.

  • Frontend Direct Configuration: Parameters passed directly to Python scripts.

🔍 Evaluation Capability Upgrade

  • Multi-image inference support.
  • Automatic prediction result comparison.
  • Simultaneous display of image and text content.

🏗️ Version 2505 (Initial Stable Release)

Release Date: May 28, 2025

Core Features

📊 Full Lifecycle Management

  • End-to-end solution from model development and training to deployment.

  • Reduces trial-and-error costs and improves iteration efficiency.

🔧 Flexible Adaptation

  • Supports training models of various scales.

  • Breaks hardware and algorithmic limitations.

📈 Evaluation System

  • Business-oriented evaluation standards.

  • Objectively validates model usability.

⚡ Experimentation Capability

  • Rapid experimental iteration.

  • Accelerates decision-making cycles.

🔒 Security & Control

  • Supports private deployment.

  • Enterprise-grade security standards.

Technical Architecture Evolution

v2505 → v2507

  • Expanded from basic large model support to multimodal vision models.

  • Enhanced training parameter flexibility.

  • Improved evaluation system.

v2507 → v2509

  • Restructured lifecycle management system.

  • Introduced non-linear workflows.

  • Expanded model type support.

v2509 → v2511

  • Optimized user experience details.

  • Enhanced error handling capabilities.

  • Enriched practical example library.

🎯 Use Cases

✅ Enterprises building vertical domain-specific large models
✅ Research institutions for efficient training experiments
✅ Developers for rapid AI application iteration
✅ Multimodal vision task processing
✅ Image generation and editing applications

🔗 Resources

  • 🌐 Official Website: www.deepextension.ai

  • 📚 Documentation: https://deepextension.readthedocs.io/en/latest/

  • ⬇️ Source Code: https://github.com/DeepExtension-AI/DeepExtension

🚀 Getting Started

Recommendations for New Users:

  1. Start with v2511 for the best user experience.

  2. Refer to the new fine-tuning examples to get up to speed quickly.

  3. Utilize the optimized training flow to improve development efficiency.

For Existing Users Upgrading:

  • Smooth upgrade, compatible with existing projects and configurations.

  • Enjoy a smoother training and evaluation experience.

  • Explore more application scenarios with the new examples.

💬 Feedback & Support

We are continuously improving – your feedback is very important to us!
If you encounter issues or have suggestions, please contact us via:

🐛 Issue Reporting: GitHub Issues
💡 Feature Suggestions: Official Website Feedback Form

DeepExtension – Extend Your AI Capabilities, Not Complexity.