Gemma 3 270M support, please

[rmarioni]

Google DeepMind juts dropped the small language model w/270M parameters which is very mobile/device friendly. Aware that Gemma 3 nano E1B also is availble but this model is still a big stretch for mobile apps. Kindly ask for some support on the RN bindings for Gemma 3 270M. This is going to be huge.

Has anyone started or looking into support for the model?

[msluszniak]

Hi, I created an issue from this feature request. I will create a separate issue on ExecuTorch repo to add support for this model. Once it will be supported, we will be able to port this to React Native ExecuTorch :))

[msluszniak]

Ok, as I received information that this model can be exported by optimum executorch, see this: pytorch/executorch#14941, we need to make it possible to support such models in RNE. Now, I will escalate one task that needs to be done before we can port it, namely #609 since this model was probably exported with these kernels.