Bringing open AI models, training frameworks, data sets, and workflows to the NVIDIA platform for quantum-GPU supercomputing.
Accelerating Quantum Computing Breakthroughs with AI
Scaling quantum computing demands state-of-the-art AI, yet purpose-built models have remained out of reach for the quantum computing community.
NVIDIA Ising brings these missing tools to the NVIDIA quantum platform, making them available to the entire quantum ecosystem. This open-source family of AI models spans key quantum workloads, starting with Ising Calibration for automating the rapid tuning of quantum processors and Ising Decoding, which accelerates the real-time decoding required for quantum error correction. All models are available pre-trained and include guidelines, data, and tooling for retraining, fine-tuning, and deployment.
Video
The Missing Toolset for Quantum-GPU Supercomputing
NVIDIA Ising brings purpose-built AI to the NVIDIA quantum-GPU supercomputing platform, complementing the NVIDIA® CUDA-Q™ software platform and NVQLink™ hardware interconnect.
Learn how AI is driving breakthroughs for key quantum workloads, and how NVIDIA Ising is bringing it to the quantum computing ecosystem.
Models
AI for Quantum Processor Development
NVIDIA Ising Calibration
This first-of-its-kind, open 35B parameter Vision Language Model is fine-tuned to infer calibration actions from QPU experimental data. It outperforms all other models on a suite of six tests measuring calibration performance and easily works with an agent to fully automate QPU calibration.
NVIDIA Ising Decoding
A pair of open 3D CNN models for performing pre-decoding, optimized to be both fast and accurate (with 0.9M or 1.8M parameters). Ising Decoding ships with models working with a depolarizing noise model for surface codes of any distance and includes a new training framework to support any noise model through PyTorch and CUDA-Q.
Benefits
The AI Tools Quantum Computing Has Been Waiting For
Performant
Achieve substantial speedups versus traditional solvers while maintaining high performance. Ising Calibration outperforms all other approaches across a suite of six tests, while Ising Decoding outperforms the state of the art with 2.5x improvement in speed and 3x improvement in accuracy. Read the model architecture paper for more details.
Open
NVIDIA Ising was released with permissive licensing and documented data provenance, training methods, data sets, and tools to fine-tune and quantize the models. Allowing developers to train or fine-tune for their own hardware, and with proprietary data.
Trusted and Benchmarked
NVIDIA Ising models provide robust verification, physics-consistency, and uncertainty quantification (UQ). All models are evaluated with transparent, reproducible benchmarks defined against reputable baselines. See the benchmark paper for definitions and results.
Easy to Use
Models come pre-trained for common use cases, and a cookbook of workflows provides domain experts simple steps to train or fine-tune models for their specific use cases. NVIDIA NIM™ microservices also provide instant setup.
Use Cases
AI for Key Quantum Workloads
The NVIDIA Ising family AI-accelerates two key workloads for developing and operating quantum processors at scale.
-
Quantum Error Correction
-
Control and Calibration
Quantum Error Correction
For quantum processors to run, qubit errors must be continually corrected by quantum error correction codes. This requires terabytes of qubit measurement data to be processed thousands of times per second by demanding (classical) decoding algorithms. NVIDIA Ising makes this possible by providing ready-to-use AI solutions for decoding.
