NVIDIA Releases Ising
- NVIDIA published 'Ising', an open quantum‑AI model family for hybrid quantum‑classical systems. - The release is promoted as the first open quantum‑AI model family aimed at hybrid workflows. - For most startups this is research‑stage, but it signals AI research branching into alternative compute paradigms beyond conventional scaling (marktechpost.com).
Quantum computers store information in qubits, which are fragile enough that stray heat, vibration, or timing errors can knock calculations off course. NVIDIA said on April 14 it released Ising, an open model family built to use artificial intelligence to help tune and correct those systems. (nvidianews.nvidia.com) NVIDIA described Ising as the first open family of quantum-AI models for hybrid workflows, meaning classical chips and quantum processors split the job instead of either one working alone. The company said the package includes models, training frameworks, datasets, cookbooks, and deployable services on its quantum-GPU software stack. (nvidia.com) The launch starts with two tracks. Ising Calibration uses a 35-billion-parameter vision-language model to read lab data and help adjust quantum hardware, while Ising Decoding ships smaller models with 1.79 million and 0.91 million parameters for quantum error-correction decoding. (github.com) Calibration is the work of constantly retuning a quantum machine so its qubits behave as intended. NVIDIA said its benchmark dataset, QCalEval, is meant to test whether models can analyze calibration experiments and recommend the next setting changes. (github.com) Error correction is the second bottleneck: quantum hardware makes mistakes far more often than ordinary computers, so systems need a fast way to spot and patch likely errors before they spread. NVIDIA said today’s best quantum processors fail about once every thousand operations, while useful large-scale systems would need error rates closer to one in a trillion. (developer.nvidia.com) In its announcement, NVIDIA said Ising’s decoding models were up to 2.5 times faster and 3 times more accurate than traditional approaches. The company also said the calibration model delivers its strongest published AI-based calibration results so far. (investor.nvidia.com) The release fits NVIDIA’s larger push to make quantum computing a hybrid computing problem, with graphics processors handling simulation, training, and control around the quantum chip. Its Ising pages position the models alongside CUDA-Q and other tools for what NVIDIA calls quantum-GPU supercomputing. (nvidia.com) The code is public now. NVIDIA’s GitHub repository links to Hugging Face model weights, the QCalEval benchmark, an Ising-Decoding training framework, and a blueprint for an agentic calibration workflow for quantum computers. (github.com) That does not mean useful quantum computers are suddenly here. What NVIDIA released is software for two narrow but stubborn engineering problems, and the company’s pitch is that better classical AI can help quantum hardware get to the point where broader applications become practical. (developer.nvidia.com)
