EmbeddingGemma 2: An open, lightweight multimodal embedding model

I really appreciate that EmbeddingGemma 2 is under the Apache 2.0 license.

For embedding models in particular, I don't think it makes sense to use a closed, proprietary, hosted-only model.

Most applications of embedding models involve calculating thousands or even millions of embedding vectors and storing them for later comparison.

If your model is proprietary, the vendor is likely someday going to decide to stop offering that model. They'll have a better model to replace it, but you still need to pay to re-calculate those millions of stored existing vectors.

(In April 2024 OpenAI offered to "cover the financial cost of users re-embedding content with these new models" - https://openai.com/index/gpt-4-api-general-availability/ - but I don't think that's something we can rely on from every provider.)

Notably, I don't want to host the model myself. I'd much rather pay a provider for a hosted model while knowing that if they ever stop hosting it I can run the open weights version myself - or find another vendor who can do that for me.

It's also very neat that this can be used for "Jev"-like tasks with text and image.

https://developers.google.com/edge/mediapipe/solutions/decis...

Hats off to google for offering OSS (or at least open weights + license) a model that would be probably pretty closed to what they would ship in their Android phones.

Note that unlike prior on device embedding models, this seems to be trained with MRL, not MatFormers, meaning you don’t get to shrink the model weights alongside the lower dimensional embeddings, unfortunately. Likely there’s not good research for how to do MatFormers for multimodal yet?

Would be good to see how it compares to the embedding models from https://www.voyageai.com/ for text. I have used these a few times in the past and have found them superior to the Qwen models compared to here.

Finally. I was getting annoyed that there's been an inflection point in how LLMs/agents work but there hasn't been a good moderate-size embeddings model, and this one is multimodal too! 270M for text only is great compared to older embedding models, and a total 440M for text + vision is also fair.

I also may or may not have a tool for much faster local embedding creation that I calibrated for EmbeddingGemma but didn't want to release until a better embedding model came along.

So what are some use cases people have found for running these sized multimodals on their device? What is it accurate on, and what is the hallucination rate like?

for text, benchmarks are identical to the first EmbeddingGemma.

but you can use this new one and enable/disable what you don't need.

can keep only text for ex.

Parameter count split is interesting:

740M total (270M text, 170M vision, 300M audio)

why isn't this being compared to siglip2 (also from google)? because that one isn't fully multimodal? or because it's a different org/team?

I'm considering adding this in my harness after some testing, this seems like a really nice embedding model!

rag transformations would be legendary