ModelCat AI Introduces Model Retargeting for Cross-Device AI Model Portability

Sunnyvale-based ModelCat AI has introduced a new capability called Model Retargeting, enabling AI model portability across various silicon devices, announced in a press release.

The feature allows users to transfer existing AI models between different hardware types, including GPUs, CPUs, NPUs, and analog computation platforms, without retraining or accessing original datasets. Using its proprietary agentic AI technology, ModelCat automates the optimization process to maintain performance across parameters like accuracy, power, memory, and inference speed.

Model Retargeting supports retargeting models in Keras V2 format and is currently available in limited beta. Supported silicon includes hardware from NXP, Silicon Labs, Alif Semiconductor, and STMicroelectronics. Future releases will expand compatibility to additional model formats such as Keras V3.

According to ModelCat, this capability enables customers to take advantage of new hardware innovations quickly, reducing the time required to adapt models for new inference devices from weeks or months to just hours.