Detail version :
Cooler_26.08.24.01
Keyboard_25.07.25.01
VGA_26.08.24.01
Power_26.08.24.01
Mouse_26.08.24.01
SingleLed_26.08.24.01
MBUpdate_26.07.22.01
MBEasyTune_26.08.06.01
MBStorage_26.09.08.01
RGB_Sync_Control_26.08.31.01
rgbMotherboard_26.08.27.01
Notebook_26.08.28.01
Sidekick_26.08.04.01
GCC Main Program_26.08.28.01

Release Highlights
1. Linux Kernel 7.0 Compatibility Issue Resolution
 Provided a solution for users experiencing Linux Kernel 7.0 compatibility issues. Added a downgrade procedure from Linux Kernel 7.0 to 6.17 to the User Manual.
2. Graphics Driver Compatibility Issue Resolution
 Provided a solution for graphics driver compatibility issues. Added instructions for removing and reinstalling the graphics driver to the User Manual.

1. Added Smart Assign Mode
2. Improved User Interface
3. Added support to GIGABYTE GeForce series graphics cards

Release Highlights
1. New User Interface Design
Introduced a completely redesigned UI/UX to provide a more intuitive and seamless user experience.
2. Workflow Optimization
Streamlined operational workflows and simplified user interactions to improve overall efficiency.
3. Database Management System
Added a comprehensive database management system with enhanced data visualization and management capabilities.
4. Fine-tune Benchmark
Introduced Fine-tune Benchmark to monitor and evaluate hardware performance during the fine-tuning process, enabling efficient comparison and analysis of system performance.
5. Fine-tune Validation
Added a Fine-tune Validation mechanism to ensure model quality and improve the reliability of validation results.
6. Fine-tune Report Download
Added support for downloading Fine-tune analysis reports, making it easier to save, review, and share results.
7. Notification Center
Added a notification center that enables users to monitor task progress and track completed tasks in real time, improving visibility and workflow management.

1 .Upgrade system to Ubuntu 24.04.1
2 . Update ROCm drivers to version 7.1
3 . Upgrade NVIDIA CUDA toolkit to 13.0
4 . Add support for AMD R9700 GPU

1. Newly supports certain legacy GIGABYTE motherboards using the latest BIOS.
2. LLM/LMM Fine-tuning: Added support for Qwen-2.5-VL training.
3. Validation: Feature removed, replaced by Inference.
4. Multimodal RAG: Optimized RAG functionality, improved document library loading issues.
5. Inference: Fixed text-to-text LoRA model loading program, added LoRA model mounting feature for text-to-image.
6. Model Download: Focused on GIGABYTE tested and verified models, each model is provided with the link to its Hugging Face repo, and disclose error messages.
7. Machine Learning: Changed to the most recommended package for each topic.
📌Available on Linux (Ubuntu 24.04.2 LTS) and WSL2 (Ubuntu 24.04.2 LTS) on Windows 11.
📌Please update to the latest BIOS version for AI TOP Motherboards to ensure the stable use.

1. Support other old GIGABYTE motherboards with the newest BIOS;
2. Dataset Creator: Automatically transforms unorganized datasets into structured Q&A lists, ready for fine-tuning with optimal efficiency.
Support Languages: English, S. Chinese, T. Chinese, German, Japanese, Korean;
3. LLM/LMM Fine-tuning: AI TOP Utility now supports the latest 405B-parameter models, enabling both LLMs for text generation and LMMs for generating images and videos;
4. Validation: Helps users quickly validate the quality of fine-tuned models from LLM/LMM fine-tuning;
5. Multimodal RAG: Utilizes the new RAG feature to select the most suitable models for fine-tuning, with in-app validation for instant performance feedback and fast improvements;
6. Model Download: A new feature that automatically downloads Hugging Face models for training and deployment;
7. SSD Mounting: Allows users to automatically mount 1 to 2 NVMe SSDs, offloading training memory to SSDs for improved performance;
8. Multi-node Optimization: Reduces VRAM usage and increases model training speed when using multiple GPU cards or cluster-based training;
9. Model Converter: Converts various model formats (.safetensors, .bin) into the GGUF format using different quantization methods (f32, f16, bf16, or 8-bit) to reduce model size or improve performance, and is optimized for deployment on Linux and Windows (LM-Studio);
10. Inference: Supports real-time inference for state-of-the-art models with different formats (safetensors, GGUF). Remote access via QR code from smartphones, laptops, or tablets — no internet required, ensuring secure and private usage;
11. Machine Learning: Prebuilt ML project templates let users build end-to-end ML workflows with zero coding;
12. Bug Fixes: Resolved various issues to improve stability and performance;
📌 Please update your AI TOP Motherboard to the latest BIOS version for optimal functionality.