NVIDIA Ising Enables Fully Automated Quantum Computer Calibration
AI-powered quantum processor calibration is now faster and more efficient.
1What is NVIDIA Ising Calibration 1.5?
NVIDIA Ising Calibration 1.5 is a vision-language model designed to automate the diagnosis and tuning of quantum processors. With 31 billion parameters, it can run on a single GPU or NVIDIA DGX Spark using NVFP4 quantization technology. The model reduces size by 11.4% at BF16 precision, making it easier to deploy in lab environments. It is trained on data from various qubit modalities and evaluated using the QCalEval benchmark.
2Technical Features of the Model
NVIDIA Ising Calibration 1.5 supports NVFP4 quantization, enabling deployment on a single GPU or NVIDIA DGX Spark. The model is 11.4% smaller at BF16 precision, simplifying deployment. It integrates with the NVIDIA Nemo Agent Toolkit to support automated quantum calibration workflows. Full-parameter checkpoints, quantized versions, and open datasets are available under the OpenMDW License.
3Training and Datasets
NVIDIA Ising Calibration 1.5 is trained on data from various qubit modalities, including superconducting qubits, quantum dots, ions, neutral atoms, and electrons on helium. These datasets are provided by partners and collaborators. The model is evaluated using the QCalEval benchmark, demonstrating competitive performance in zero-shot and in-context learning tasks.
4Performance and Benchmarking
NVIDIA Ising Calibration 1.5 shows strong performance on the QCalEval benchmark. The model outperforms previous versions by 10% in zero-shot tasks and by 86.68% when using examples from related experiments. It surpasses open models and remains competitive with leading closed models.
5Zero-Shot and In-Context Learning
NVIDIA Ising Calibration 1.5 excels in both zero-shot and in-context learning. Zero-shot learning analyzes results independently, while in-context learning evaluates results in the context of related samples. These capabilities are crucial for automating quantum processor tuning and retuning operations.
6Deployment and Integration
NVIDIA Ising Calibration 1.5 offers optimized tokens per second performance on NVIDIA DGX Spark, ensuring efficient local throughput. The model is available through NVIDIA NIM and NVIDIA Build. The OpenMDW License from the Linux Foundation provides flexibility for quantum processor developers and operators to deploy anywhere.
7Open Source and Community Support
The NVIDIA Ising model family is fully open source. Weights, datasets, benchmarks, and recipes are provided, allowing users to modify, deploy, and fine-tune their own models. Full-parameter checkpoints are available on Hugging Face. A ready-to-use agent harness blueprint is also available on the NVIDIA/Quantum-Calibration-Agent-Blueprint GitHub repository.
8The Future of Quantum Computing
NVIDIA Ising Calibration 1.5 represents a significant step forward in quantum computing and AI. By automating quantum processor calibration, the model enables faster and more efficient operations in lab environments. This advancement will play a critical role in the future applications of quantum computing.