How can I force the size of the embedding vector to be 2000?

[aropb]

I need to set the embedding vector size = 2000 to set the pgvector index, which has a limit of 2000. What is the best way to do this? The default model generates an embedding with a length of more than 4000.

[marcominerva]

Which model are you using? text-embedding-3-small has a dimension of 1536 values, while text-embedding-3-large uses 3072 values. And in both cases, you can shorten the embedding dimensions.

[aropb]

I'm using Qwen2.5. But it doesn't matter. I need to force the embedding size for any model. How to do it? Im using LLamaSharp.

[marcominerva]

It can't be valid for any model, because it depends if the model support shortening. You need to check if the model supports this feature and read in the documentation of your model how to specify the dimension. For example: https://platform.openai.com/docs/api-reference/embeddings/create#embeddings-create-dimensions

[aropb]

Is it probably better to use a separate model for embedding? Which one do you recommend?

[marcominerva]

I typically use the text-embedding-3-small or text-embedding-3-large models from OpenAI.