For Titan embedding model which distance metric to use

Question

As per best practice to choose distance metric  "Use the distance metric that matches the model that you're using". For Titan embedding model which distance metric to use.

Accepted Answer

Hello ,

Sudip

The Titan models are trained using cosine similarity, so the cosine distance metric would be the most appropriate choice.

Thanks,

Abhinav

Answer

Hi,

Your choice of distance metric is fairly open: the engines will return you embedding vectors and you choose which one you apply.

You have a wide choice well detailled at https://weaviate.io/blog/distance-metrics-in-vector-search

Cosine similarity is a frequent choice: it is normalized so always between 0 and 1.  If you choose a non-normlized distance, you become dependent on embeddings vector length, which may become an issue if you want to compare distance across embeddings engines.

Best,

Didier

For Titan embedding model which distance metric to use

Relevanter Inhalt