Announcing vLLM v0.12.0, Ministral 3 and DeepSeek-V3.2 for Docker Model Runner
At Docker, we are committed to making the AI development experience as seamless as possible. Today, we are thrilled to announce two major updates that bring state-of-the-art performance and frontier-class models directly to your fingertips: the immediate availability of Mistral AI’s Ministral 3 and DeepSeek-V3.2, alongside the release of vLLM v0.12.0 on Docker Model Runner.
Whether you are building high-throughput serving pipelines or experimenting with edge-optimized agents on your laptop, today’s updates are designed to accelerate your workflow.
Meet Ministral 3: Frontier Intelligence, Edge Optimized
While vLLM powers your production infrastructure, we know that development needs speed and efficiency right now. That’s why we are proud to add Mistral AI’s newest marvel, Ministral 3, to the Docker Model Runner library on Docker Hub.
Ministral 3 is Mistral AI’s premier edge model. It packs frontier-level reasoning and capabilities into a dense, efficient architecture designed specifically for local inference. It is perfect for:
Local RAG applications: Chat with your docs without data leaving your machine.
Agentic Workflows: Fast reasoning steps for complex function-calling agents.
Low-latency prototyping: Test ideas instantly without waiting for API calls.
DeepSeek-V3.2: The Open Reasoning Powerhouse
We are equally excited to introduce support for DeepSeek-V3.2. Known for pushing the boundaries of what open-weights models can achieve, the DeepSeek-V3 series has quickly become a favorite for developers requiring high-level reasoning and coding proficiency.
DeepSeek-V3.2 brings Mixture-of-Experts (MoE) architecture efficiency to your local environment, delivering performance that rivals top-tier closed models. It is the ideal choice for:
Complex Code Generation: Build and debug software with a model specialized in programming tasks.
Advanced Reasoning: Tackle complex logic puzzles, math problems, and multi-step instructions.
Data Analysis: Process and interpret structured data with high precision.
Run Them with One Command
With Docker Model Runner, you don’t need to worry about complex environment setups, python dependencies, or weight downloads. We’ve packaged both models so you can get started immediately.
To run Ministral 3:
docker model run ai/ministral3
To run DeepSeek-V3.2:
docker model run ai/deepseek-v3.2-vllm
These commands automatically pull the model, set up the runtime, and drop you into an interactive chat session. You can also point your applications to them using our OpenAI-compatible local endpoint, making them drop-in replacements for your cloud API calls during development.
vLLM v0.12.0: Faster, Leaner, and Ready for What’s Next
We are excited to highlight the release of vLLM v0.12.0. vLLM has quickly become the gold standard for high-throughput and memory-efficient LLM serving, and this latest version raises the bar again.
Version 0.12.0 brings critical enhancements to the engine, including:
Expanded Model Support: Day-0 support for the latest architecture innovations, ensuring you can run the newest open-weights models (like DeepSeek V3.2 and Ministral 3) the moment they drop.
Optimized Kernels: Significant latency reductions for inference on NVIDIA GPUs, making your containerized AI applications snappier than ever.
Enhanced PagedAttention: Further optimizations to memory management, allowing you to batch more requests and utilize your hardware to its full potential.
Why This Matters
The combination of Ministral 3, DeepSeek-V3.2, and vLLM v0.12.0 represents the maturity of the open AI ecosystem.
You now have access to a serving engine that maximizes data center performance, alongside a choice of models to fit your specific needs—whether you prioritize the edge-optimized speed of Ministral 3 or the deep reasoning power of DeepSeek-V3.2. All of this is easily accessible via Docker Model Runner.
How You Can Get Involved
The strength of Docker Model Runner lies in its community, and there’s always room to grow. We need your help to make this project the best it can be. To get involved, you can:
Star the repository: Show your support and help us gain visibility by starring the Docker Model Runner repo.
Contribute your ideas: Have an idea for a new feature or a bug fix? Create an issue to discuss it. Or fork the repository, make your changes, and submit a pull request. We’re excited to see what ideas you have!
Spread the word: Tell your friends, colleagues, and anyone else who might be interested in running AI models with Docker.
We’re incredibly excited about this new chapter for Docker Model Runner, and we can’t wait to see what we can build together. Let’s get to work!
Quelle: https://blog.docker.com/feed/
