Coral summer updates: Post-training quant support, TF Lite delegate, and new models! - Google Developers Blog

This post announces several summer updates for the Coral Edge AI platform. Key updates include: support for post-training quantization for int4 models, enabling more efficient on-device inference; a new TensorFlow Lite delegate for Coral, allowing for easier integration of Coral hardware with TensorFlow Lite models; and the release of new models, including Gemma 3, which is a multimodal on-device SLM supporting text, image, video, and audio inputs. Additionally, new libraries for RAG and Function Calling are introduced, enabling developers to build AI features fully on the edge. MediaPipe now supports post-training quantization for int4 models, and TensorFlow Lite also supports this for Coral models.