I want to create some simple /v1/completions mode calls, and I'd like to offer a dropdown for which model to use. However not all models are available for complete (like "gpt-3.5-turbo" is a chat only model). Is there a way to limit the model call/filter it. As the documentation seemingly offers no filter methods, and the response payload doesn't mention the supported modes. Or do I need to hardcode the 6 options. (cushman also does not seem to be listed in model endpoint comptability)
#Models call => determine which ones are available for which mode
thorn jolt
winter scroll
You'll want to hard code them. Good catch on cushman missing on that page, they recently overhauled it and added that section.
thorn jolt
That's what I "dreaded". I was hoping to first do a models call so I wouldn't have to adapt it as new models get added/removed
In case you can pass it on: there's a small typographical error with davinci-002 as well:
text-davinci-002 /v1/completions`
civic stump
I want to create some simple /v1/completions mode calls, and I'd like to offer a...
yeah unfortunately nothing to do but hardcode the 6 options
and then create delegators for executing between the different options
like e.g if the model is turbo, use the turbo endpoint with correct setup
thorn jolt
like e.g if the model is turbo, use the turbo endpoint with correct setup
That's what I'll be doing albeit the inverse. I do hope in the future the option gets added. But only so many hands available of course.
civic stump
yeah, and different models will have different inputs // ways to get completions
It's pretty tough to standardize all models
but much easier for us on our end to just create respective delegating functions that just slightly adjust parameters before a request and etc when we're making projects
thorn jolt
Sorry, I might have phrased it poorly @civic stump . I just meant it as in: either a param on the models call so I can tell whether I want the available models for complete/chat/insert/edit, or that the generic model result object also contains an availableFor array. Because either way now we need to hardcode a bit for the transformation, and that's something I'd like to avoid as much as possible. Especially since in this case I just wanted to make a simple project using completions, but wanted to dynamically fetch models.