Gemma 3 270M - Google’s Lightweight On‑Device Model
Google has released a TINY model that embodies the right tool for the job philosophy.
It’s a 270M (yeah, million not billion 😅)
It has support for full model fine tune support. For example, you can fine tune it using Transformers
And it is primarily intent for on-device LLM use cases.
Here is a quick demo of Bedtime story generator that runs entirely on your browser using Transformers.js👇
How to use it using Ollama?
Just run
ollama run gemma3:270m
When I tried it, the generated answer felt almost instant.
But in term of accuracy, it is not that good (as you might have already guessed).
How to use it in Browser?
The demo Bedtime Story Generator - a Hugging Face Space by webml-community actually runs in the model in your browser.
You can use Transformer.js to run it.
And you can refer to the bed time story demo on how they’ve implemented it.
https://huggingface.co/spaces/webml-community/bedtime-story-generator/blob/main/src/hooks/useLLM.ts
References
