✅ Key Highlights
🧠
New Model Support
- • Qwen3-4B Instruct / Thinking
- • MiniCPM-V 4.5
⚙️
VLLM Engine Enhancements
- • Multi-model loading support
- • AWQ 8bit quantization support
- • VLLM upgraded to 0.10.2 in CUDA 12.8 image
🖼️
OpenAI Image Edit API Support
Direct compatibility with images/edits interface, enhancing compatibility with image editing and generation models.
🌐 Community Edition Updates
📦 Installation
pip
pip install 'xinference==1.11.0.post1'
Docker
Pull the latest image or update using pip in container
🆕 New Model Support
Qwen3-4B
Instruct / Thinking versions
MiniCPM-V 4.5
Vision Language Model
✨ New Features
OpenAI image edit API support
VLLM multi-model loading support (including Omni, image, video, audio models)
VLLM AWQ 8bit quantization support
CUDA 12.8 image upgraded VLLM to 0.10.2
🛠 Improvements & Fixes
Fixed UI button issue when n_gpu_layers=-1
Fixed CI build and CUDA 12.8 Dockerfile issues
Synchronized multimodal model JSON (audio, image, video, LLM)
🏢 Enterprise Edition Updates
🚀 Kubernetes Operator Preliminary Support
Automatic model replica scheduling and lifecycle management, providing unified interface for clustered inference
🔒 Stability Enhancements
Fixed several known issues, overall operation is more stable and reliable