NVIDIA posts Ising models

NVIDIA has released Ising, an open-source family of artificial intelligence models built to help researchers tune and run quantum computers. (nvidianews.nvidia.com) Quantum computers use qubits, which are fragile bits that can lose information from noise and drift. NVIDIA said Ising launches with two model lines: Ising Calibration, for tuning hardware, and Ising Decoding, for spotting and fixing errors fast enough to keep calculations on track. (developer.nvidia.com) On NVIDIA’s product page, the company said the first calibration model is a 35 billion-parameter vision-language model trained to read experimental charts from quantum processors and suggest calibration actions. NVIDIA said the models are available pre-trained, with data, tooling, and instructions for retraining and fine-tuning. (nvidia.com) The problem Ising targets is basic but stubborn: today’s qubits are noisy enough that useful machines need constant retuning and layers of error correction. NVIDIA said those two jobs are among the main bottlenecks in building hybrid systems that combine quantum processors with classical supercomputers. (nvidianews.nvidia.com) NVIDIA is fitting Ising into a larger quantum stack it has been building around CUDA-Q, its open-source platform for programming systems that split work across central processing units, graphics processing units, and quantum processing units. The company’s quantum pages also place Ising alongside NVQLink, its hardware interconnect for tying quantum processors to accelerated computing systems. (developer.nvidia.com, developer.nvidia.com) The release also extends NVIDIA’s broader 2026 push around open models and developer tooling. In January, the company said it was expanding open models, data, and tools across industries; Ising applies that strategy to quantum hardware research. (blogs.nvidia.com) NVIDIA’s developer materials describe Ising as a model family, training framework, and cookbook, not just a single download. The company said quantum teams can use it without deep machine-learning expertise, then adapt the models for their own devices and workflows. (developer.nvidia.com) NVIDIA is also framing the release as part of a community effort rather than a closed in-house system. Its forums direct questions about benchmarks, training frameworks, deployment, and documentation to the Ising GitHub issues page, signaling that the code and support process are meant to live in public. (forums.developer.nvidia.com) For now, Ising does not change the physics limits on qubits. It gives quantum labs a new set of open artificial intelligence tools for the slow, repetitive work that stands between experimental devices and usable machines. (developer.nvidia.com)