Bringing open AI models, training frameworks, data sets, and workflows to the NVIDIA platform for quantum-GPU supercomputing.
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.
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.
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.
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.
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.
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.
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.
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.
The NVIDIA Ising family AI-accelerates two key workloads for developing and operating quantum processors at scale.
Quantum Error Correction
Control and Calibration
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.
Keeping quantum processors operational requires continual tuning to account for hardware imperfections or misalignments. Current approaches to this calibration are neither scalable nor fast enough, relying on human intervention or simple algorithms. NVIDIA Ising Calibration provides an open model capable of rapidly interpreting the status of quantum hardware and connects to an agent to automate its correction.
Opening the AI Needed by the Entire Quantum Ecosystem
NVIDIA Ising is being adopted by research institutes, national laboratories, and quantum hardware and software builders around the world.
Use the tools, models, and datasets from NVIDIA Ising to make the breakthroughs you need to scale qubits into useful quantum applications.
Get the latest news on the future of useful quantum-GPU supercomputing.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。