Both of our new embedding models were trained with a technique that allows developers to trade-off performance and cost of using embeddings. Specifically, developers can shorten embeddings (i.e. remove some numbers from the end of the sequence) without the embedding losing its concept-representing properties by passing in the dimensions API parameter. For example, on the MTEB benchmark, a text-embedding-3-large embedding can be shortened to a size of 256 while still outperforming an unshortened text-embedding-ada-002 embedding with a size of 1536.
Based on the API reference, this is available via a new parameter on the embedding request called dimensions (optional integer).
Describe the feature or improvement you're requesting
From the blog on new embedding models:
Based on the API reference, this is available via a new parameter on the embedding request called
dimensions
(optional integer).Additional context
I'm happy to write a PR for this.