Google describes EmbeddingGemma 2 tasks for on-device video search and routing in AI Edge

AIGoogle's MediaPipe Decision Task uses EmbeddingGemma 2 to classify multimodal inputs for routing or action prediction in under 100 ms without fine-tuning. The post links to documentation for developers to try the feature.

AIMediaPipe's new Decision Task uses EmbeddingGemma 2 to classify multimodal inputs for routing or action prediction in under 100 milliseconds. The task needs no fine-tuning, according to Google, which points developers to the Google AI Edge Gallery to experiment with it.

AIGoogle says developers building with EmbeddingGemma 2 in Google AI Edge need no custom code for complex data matching. MediaPipe Tasks, part of the AI Edge suite, handles image resizing, tokenization, and search ranking so apps can retrieve exact video timestamps in milliseconds.