What are the valid embedding input values?

API
1

I am attempting to create the embeddings for each sentence in a book. So after removing all sentences less than 3 characters, I have a list of strings of length 2305. When I pass this list to the embedding input, I get "(LIST) is not valid under any of the given schemas - ‘input’. So it seems there are some values that are not being accepted. What are they? Are there certain characters which the embedding model does not accept?

2

It seems like you might be passing the entire array to the Embeddings API, instead of each individual sentence.

3

Yes I am. It accepts either a string or array. It works fine if I pass it a list of strings around 250 length. Is there a size limit?

4

I haven’t seen an limits specific to batch size, so maybe you’re hitting the 350,000 TPM rate limit?

If not, maybe there is a batch limit that just isn’t well documented.

5

If you’re still thinking this might be the problem, I would send one sentence per request so that you can see which is causing the error, and investigate it from there.

6

Yeah good suggestions. I was hoping for some documentation to clear up the guesswork.

Edit: @wfhbrian I was able to embed it by breaking the calls up. The problem was I was exceeding the 8192 token limit. Duh…

1 Like