Verba: Changing OpenAI Embeddings & Chunk Size

Is it possible to easily change the Verba code so we can try using the new OpenAI “text-embedding-3-large” model? We’re curious to see if that improves retrieval for unusual words.

And is it possible to play with changing chunk size?


Verba has now a modularized structure that easily allows new embedding components.

This is the one that controls the current OpenAi embedding:

We should have some contents around this new model soon.

You can probably experiment with it by adapting this recipe:

Let me know if this helps.


Thanks. I think that means that for now there isn’t an easy way to use the 3072 dimension “text-embedding-3-large” embedder with Verba.

Not without changing the code.

But hopefully we get it soon.

It would be interesting to point this out as an issue:

Done – issue created. Thanks Duda.

1 Like