Skip to main content
Version: 1.5.0

Version 1.5

GenAI Studio version 1.5 series has been released:

  • Initial version v1.5.0 was released on 2026/09/09.
  • No patch version is available yet.

✨ What's New​

Multi-GPU Support for LoRA Fine-Tuning​

This release introduces automatic multi-GPU support for LoRA fine-tuning. When a model cannot be fine-tuned on a single GPU, GenAI Studio automatically uses multiple GPUs for fine-tuning, enabling more efficient utilization of available GPU resources.

New Built-In Inference Engine​

In addition to the default Ollama inference engine, this release adds Phison aiDAPTIVLink as a new built-in inference engine option. Built on vLLM, Phison aiDAPTIVLink delivers the high performance of vLLM while also supporting KV cache offloading to AI SSDs.

warning
  1. Ollama remains the default inference engine after GenAI Studio is installed.
  2. To use Phison aiDAPTIVLink as the system inference engine, an AI SSD must be installed on the host running GenAI Studio. Otherwise, this option will not be available on the corresponding settings page.

🚀 Enhancements​

Model Support Updates​

This release adds support for the following large language models in LoRA fine-tuning: Qwen3.6-27B, Qwen3.6-35B-A3B, and Qwen/Qwen3-Coder-30B-A3B-Instruct. For the complete list of supported models, please refer to Supported Models.

⚙️ Kernel Upgrades​

  • Phison Inference Middleware: Added aiADPTIVLink 3.0 NXUN303.AA as a built-in inference engine with KV cache offloading support. Users can switch between the available inference engines after installation.

📦 Third-Party Packages​

GenAI Studio uses the following third-party packages:

  • dcgm-exporter (4.5.2-4.8.1-ubuntu22.04)
    This package sends GPU monitoring metrics to Prometheus for collection.
  • Flowise (2.2.7-patch.2)
    GenAI Studio provides automated RAGOps features and workflows through this package.
  • Grafana (12.3)
    This package displays the monitoring data collected by Prometheus into a system resource monitoring dashboard.
  • llama.cpp (full-cuda-b8882)
    This package provides GenAI Studio with the ability to convert models into the GGUF file format.
  • node-exporter (1.10.2)
    This package sends the required host monitoring metrics to Prometheus for collection.
  • Ollama (0.32.6)
    This package is the default model inference server after installation.
  • OpenVINO (2.1.0)
    This package provides the ability to convert models into the OpenVINO file format.
  • Phison aiDAPTIVLink (NXUN205.A1/NXUN303.AA)
    GenAI Studio utilizes this package to provide users with the Full Parameter model fine-tuning feature (NXUN205.A1), and the KV cache offloading feature (NXUN303.AA).
  • PostgreSQL (16.13)
    This package provides relational database capabilities.
  • Prometheus (3.10.0)
    This package collects all the metric data required for monitoring to serve as the data source for Grafana.
  • Qdrant (1.17)
    This package provides vector database storage capabilities.
  • Unsloth (2026.8.22)
    GenAI Studio utilizes this package to provide users with the ability to perform LoRA fine-tuning on models.
  • vsFTP (3.0.5)
    This package provides GenAI Studio with the function to allow external systems to download converted model files.