Open comeback01 opened 12 months ago
Seems like the API is pretty similar: https://readme.fireworks.ai/docs/openai-compatibility
So you could try by setting the VITE_API_BASE
env var.
I was waiting for fireworks.ai to implement repetition penalty of some type since it's needed for any type of chat that lasts for more than a handful of interactions with LLaMa2 models, since LLaMa2 tends to get stuck in a catch-phrase rut without repetition penalty, eventually degrading to just repeating the same thing over and over.
Right now you can't adjust the model name in an OpenAI request, so I'm not sure if overriding the endpoint will work.
Hello,
I would like to use the API from https://app.fireworks.ai/
I wanted to know how to do it?
Thanks