New Gemma 3N models support?

[Nikola-Milovic]

https://huggingface.co/google/gemma-3n-E4B-it-litert-preview

Google just unveiled gemma 3n made for edge devices and mobile phones. It's still in preview but seems promising

[chmjkb]

Hi @Nikola-Milovic!
Thanks for your interest! It's not open source yet, but we’ll definitely take a look once it’s released. As far as I know, ExecuTorch does plan to support MoE models, and we’ll be keeping an eye on that as it evolves.

[CarlosZiegler]

What about Gemma 3 https://huggingface.co/blog/gemma3

[moloti]

It has been released: https://developers.googleblog.com/en/introducing-gemma-3n-developer-guide/. I'm really interested to see it implemented in the library.

[chmjkb]

Hey, we've seen the release 👀, for now we're waiting for broader VLMs support within ExecuTorch and once that is out Gemma 3n will be the first thing on our TODO list, so stay tuned - we'll keep you updated 😆

[juv85]

Hello, I'm really looking out to run Gemma 3n on react-native, since it is not yet integrated to execu-torch, can someone please suggest an alternative or resource I can make use of to get it working in my react-native app?

[juv85]

I am looking ont that project => https://github.com/cactus-compute/cactus

Ok just checked this and thanks a lot, it is quite interesting as well.

Unfortunately they don't support Gemma 3n yet as well, just Gemma 3, I will just add as issue so it can be taken into consideration

[chmjkb]

Thank you for such an extensive answer @juv85, we'll be doing all we can to bring that model to our library.

[rbgoncalves]

My project’s use case is focused on using Gemma 3n, for the same reasons @juv85 mentioned, especially its on-device ASR capabilities.

[juv85]

My project’s use case is focused on using Gemma 3n, for the same reasons @juv85 mentioned, especially its on-device ASR capabilities.

Check this -> link

Someone made a custom module to make inference specifically with Gemma 3n for React-Native (Expo)

For my use case I need it in bare react native so I forked the repo to tune it. That's the link to the original repo I shared.
Hope this helps.

ps: I am not very proficient with native modules so this might take some time to get ready, opened to any help from the community - link to my fork

[rmarioni]

Gemma 3 270M support please!

[IlirEdis]

Please support EmbeddingGemma

[jakmro]

We might look into it. Could you also tell me why the currently supported models in our library are not enough for you? The EmbeddingGemma model is quite large for local usage.