Any suggestion to pair the TFlite model predictions with the Gemini model?

Gemini API
,
1

Hi,

I have a tiny TFlite model running on the microcontroller for real time object detection and I would like to experiment with this model predictions send to the Gemini 2.0 model over C/C++ API for high level planning/reasoning, are there any suggestion to pair the TFlite model predictions with the Gemini model?

2

Hello!

If you are able to make an API call via TFlite model predictions, you can use Gemini 2.0 Flash following the cookbook for reference: https://github.com/google-gemini/cookbook/tree/main/gemini-2

3

Hi @camadi , is there any concrete example calling Gemini from TFlite either on mobile or embedded device for live please?

4

Just to clarify, TFLite appears to be specialized for on-device inference. Please note that it is currently not possible to run Gemini 2.0 Flash on-device. Therefore, if you intend to run Gemini on-device, Gemini 2.0 model isn’t available for that use case.

However, if you intend to take the TFLite predictions and feed them into Gemini 2.0 separately, that could be done via the API and there are examples in the cookbook I shared previously.

1 Like
5

Just clarify I intend to take the TFLite predictions and feed them into Gemini 2.0 via the API in C or C++

7

I’ve confirmed that currently, C++ API is not supported for you to use. Please refer to this documentation for the existing SDK.

9

Hi,
There’s always the REST API to integrate Gemini into other programming languages than available SDKs, eg. C# like I created an SDK, or C/C++ as required by @ramkumarkoppu