search

madaan / self-refine

LLMs can generate feedback on their work, use it to improve the output, and repeat this process iteratively.
https://selfrefine.info
Apache License 2.0
596 stars 48 forks source link

Codex discontinued #4

Closed hideodeo closed 1 year ago

hideodeo commented 1 year ago

Hello. For code-related tasks, do you plan to update your code to replace the Codex with another model? Do you have a suggestion for the alternative model & do you plan to push the updated code?

madaan commented 1 year ago

Hi @hideodeo, we are working on Self-Refine + text-davinci-003, ChatGPT, and GPT-4. The results so far are quite promising, and we will release the code and results soon.

hideodeo commented 1 year ago

Sounds great! Looking forward to seeing results

hideodeo commented 1 year ago

Hi @madaan I'd like to quickly try ChatGPT with gsm/run.py. Can I simply change CODEX = "code-davinci-002" to CODEX = "gpt-3.5-turbo" in gsm/run.py?

I'm wondering if it requires more modifications because I remember the way to feed examplers is different in gpt-3.5-turbo compared to code-davinci-002 (e.g. specifying "role"). Thank you!

madaan commented 1 year ago

Yes! If you update https://github.com/reasoning-machines/prompt-lib/tree/main/prompt_lib it should just work. Prompt-lib now supports chat models.

hideodeo commented 1 year ago

Nice! Thank you for info!