NVIDIA has introduced Ising Calibration 1.5, an open-source vision language model (VLM) intended for the automated calibration of quantum computers. This model interprets diagnostic outputs from quantum processors and suggests tuning parameters to maintain their operation. The release includes a new NVFP4-quantized version, enabling deployment on a single GPU or an NVIDIA DGX Spark.
Key Points
- Ising Calibration 1.5 is a 31-billion-parameter vision language model for diagnosing and tuning quantum processors.
- A new NVFP4-quantized version allows deployment on a single GPU or an NVIDIA DGX Spark.
- The model achieves an 11.4% reduction in size at BF16 precision compared to its predecessor.
- It is trained on diverse datasets from multiple qubit modalities, including superconducting qubits, quantum dots, and ions.
- Performance is evaluated using the QCalEval benchmark, demonstrating strong zero-shot and in-context learning capabilities.
- Ising Calibration 1.5 outperforms all open models and is competitive with leading closed models in quantum calibration plot interpretation tasks.
- Full-parameter checkpoints, quantized versions, open datasets, and deployment blueprints are available under the OpenMDW License.
Context
According to NVIDIA, Ising Calibration 1.5 advances AI-based quantum processing unit (QPU) calibration by analyzing unfamiliar diagnostic results without requiring prior training examples. It can also use examples from related experiments when available. The model's reduced size at BF16 precision facilitates deployment of agentic calibration workflows in local lab environments.
Why It Matters
This release provides quantum computing builders and operators with an open-source tool for automated QPU calibration, potentially reducing manual intervention and improving operational efficiency. The model's ability to perform zero-shot and in-context learning on diagnostic data can streamline the bring-up and retuning of quantum hardware.
What To Do
- Review the full-parameter checkpoints for Ising Calibration 1.5 available on Hugging Face.
- Explore the NVFP4-quantized version for deployment on a single GPU or NVIDIA DGX Spark.
- Examine the open datasets and deployment blueprints provided under the OpenMDW License.
- Investigate the NVIDIA Nemo Agent Toolkit for integration support in automated quantum calibration workflows.
Keep Exploring
/atlas/claude-family /atlas/gpt-family /atlas/gemini-family /techniques/role-objective /techniques/constraints