Right now we use only client.chat.completions. But we could also send stuff via client.completions.
(Probably useless for OpenAI models)
This would be useful for base models, but:
- We could run weird experiments where we use chat models via the completions API
- Maybe we could implement prefill that way?
This is a major task and likely not worth it. Needs thinking.
Right now we use only
client.chat.completions. But we could also send stuff viaclient.completions.(Probably useless for OpenAI models)
This would be useful for base models, but:
This is a major task and likely not worth it. Needs thinking.