Summary
- NVIDIA Ising Calibration 1.5 is an open-source VLM that interprets quantum processor diagnostic results to determine calibration methods
- An NVFP4 quantized version is available for the first time, enabling deployment on a single GPU or DGX Spark
- It achieved the best performance among open models on the QCalEval benchmark and proved competitive with some closed models
NVIDIA has unveiled 'NVIDIA Ising Calibration 1.5,' a vision-language model (VLM) designed to automate quantum computer calibration. The 31B-parameter model is built to interpret diagnostic graphs from quantum processing units (QPUs) and determine how they should be recalibrated.
This version strengthens both zero-shot performance—analyzing unfamiliar diagnostic results without prior training examples—and in-context learning (ICL) performance, which draws on related experimental cases. At BF16 precision, the model size is 11.4% smaller than the previous version. For the first time, an NVFP4 quantized version is also available, allowing deployment directly in lab settings on a single GPU or NVIDIA DGX Spark.
Trained on data from diverse qubit modalities
The model was trained on partner-provided data spanning multiple qubit modalities, including superconducting qubits, quantum dots, ions, neutral atoms, and electrons on helium. Its performance was evaluated using the QCalEval benchmark, which measures the ability to interpret diagnostic results, classify them, assess fitness, and recommend next steps.
On QCalEval, Ising Calibration 1.5 showed an 86.68% improvement over the previous model when leveraging related examples. It recorded the highest performance among open-source models and was reportedly competitive even with closed models such as Fable 5 and GPT 5.6 Sol.
NVIDIA said it has released the full-parameter checkpoint, quantized version, training dataset, and deployment blueprint under the OpenMDW license, and that integration with the NVIDIA Nemo Agent Toolkit supports building automated quantum calibration workflows.
Comments