Different text embedding model

[loftusa]

I see that this package uses text-davinci-003 as its standard model: https://github.com/Methexis-Inc/terminal-copilot/blob/main/copilot/main.py#L186

but OpenAI recommends the new text-embedding-ada-002 for nearly all use cases, which is substantially cheaper and better: https://beta.openai.com/docs/guides/embeddings/embedding-models

Regardless, one of the code models (e.g., code-davinci-002) is pretrained specifically on code examples and might be better.

Are there any plans to switch, or is there a good reason we're using text-davinci-003?

[github-actions]

Hey! Thanks for sharing feedback with the community!

[JoelKronander]

@loftusa in my experience text-davinci-003 has worked best. Feel free to make a PR for adding a command line flag to switch to a different model though, should be a great starting PR.!

[shantanuo]

I think it is important to be able to change the model if I am not getting the expected command. For e.g.

copilot kill the orphaned python process --model=codex

Ref: https://github.com/yasyf/gpt-do

[2mawi2]

@JoelKronander @shantanuo this can be closed after the GPT migration