Discussions

To be compatible with OpenAI API implementation Voyage AI API should also deliver "prompt_tokens" in the "usage" part of the response for Embeddings.

Hi there, can you please list the languages that Voyage AI's embedding model natively supports?

Hi, I'm using Typescript library to get voyage-3.5 embedding. Despite the documentation contains description of output_dimension parameetr of a request, Typescript library does not support it. method embed() takes a VoyageAI.EmbedRequest type object which contains only input, model, inputType, truncation, and encodingFormat properties. There is neither output_dimension, nor output_dtype. Even if I try to create a request of any type and include output_dimension: 2048 in it, the embed method still returns 1024 embedding vector. How can I request a different embedding size?

Running into too many concurrent streams exception -- I take that is a bit different than requests per minute or maximum tokens being exceeded. How many concurrent streams are allowed?

I'm on a big, high accuracy needed project. so I want to know that voyage multimodal embedding api will do it's job with best performance for sinhala language

Lots of new stuff in the 3rd generation of embeddings like voyage-3-large, 3.5 series and contextual embeddings.

Is a multimodal reranker in the works?

Here is my code

We're currently using OpenAI and we're used to couple hundred millisecond response times. Benchmarking voyage and recall is definitely improved but it comes at the cost of 10X slower embeddings? Am I doing something wrong.

I see reference to count_tokens in python. What about typescript or just a REST api is fine