I am trying to build a translation service.
It is very common that a terminology table is needed when we are translating a domain specific project (like game) or very large project involving collaborations.
I believe file-search is the easiest way to add terminology support. And I have tested it in Playground and quite satisfied. I made a Assistant and upload a terminology file, check the file-search option, and the model output did tend to follow terminology file to translate.
However, I found that I have to use thread to take advantage of assistant and file-search, which I do not want to.
Because I do not need state.
If I am building a tool like translate version of ChatGPT I am quite happy to use Threads but now I am building a tool that can automatically translate massive sentences one by one. You can image a game translation table in which sentences do not have necessarily have context relationship.
In a word, I hope that I can use file-search in normal stateless Chat completion query. I do notice Assistant messages
but I cannot simply turn the terminology table to assistant messages and send it in each query because it can be very big, like , thousands of terms.
Is there any best practice I can achieve that?