Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Editor mode doesn't use GPT4 (defaults to text-davinci-003) #24

Open
2 tasks done
pistachiomatt opened this issue Mar 26, 2023 · 12 comments
Open
2 tasks done

Editor mode doesn't use GPT4 (defaults to text-davinci-003) #24

pistachiomatt opened this issue Mar 26, 2023 · 12 comments
Labels
bug Something isn't working triage Issue that is to-be-triaged and prioritized accordingly

Comments

@pistachiomatt
Copy link

Verify it's not a duplicate bug report

Describe the Bug

  1. Change genieai.openai.model in settings.json to GPT4
  2. Confirm GPT4 is working in conversation view by observing speed and eloquence response
  3. Switch to editor mode and prompt a question. The response speed is way too fast to be GPT4; it must still be using gpt-3.5-turbo

Please tell us if you have customized any of the extension settings or whether you are using the defaults.

The genieai.openai.model is already set to GPT4, and the conversation view does successfully use GPT4.

Additional context

No response

@pistachiomatt pistachiomatt added the bug Something isn't working label Mar 26, 2023
@genieai-info
Copy link

Hello @pistachiomatt - the editor view is currently defaulted to text-davinci-003 for various reasons:

  1. text-davinci-003 is the only model that'd officially respect the stop parameter to ensure you are getting a code response.
  2. The nature of Editor View is not conversational but iterating within a context. So the initial implementation was done using davinci model

Though, we are still considering switching the Editor View to respect your model choice and provide a better messaging possible for the best experience. That is, if we could find a way to ensure GPT-3.5 / GPT-4 can really respond only using code (Which is relevant since otherwise you will get bunch of linter errors in Editor with non-code/non-commented text)

@genieai-info genieai-info changed the title Editor mode doesn't use GPT4 Editor mode doesn't use GPT4 (defaults to text-davinci-003) Mar 27, 2023
@pistachiomatt
Copy link
Author

pistachiomatt commented Mar 27, 2023

You can get GPT4 to follow a format. You can prompt it to "only respond with a code block" or "respond with this JSON format: { full_code: "<code only answer that…>" }". Lowering the temperature might also help.

@cmp-nct
Copy link

cmp-nct commented Apr 1, 2023

davinci-003 is a magnitude more expensive than 3.5-turbo and less capable in my tests.
all you need is a proper prompt and a stop condition inside your code instead of using stop words. Just drop the connection once stop is reached.

@genieai-info
Copy link

@cmp-nct @pistachiomatt thank you! we are aware of it and working on the best solution to provide customizability of models in Editor View. Please bear with us and we'll keep here posted once we rollout an update 🙏🏽

@veonua
Copy link

veonua commented May 9, 2023

Azure OpenAI Service: Unsupported data type Error: This means you didn't select the right base model for your Azure OpenAI service. Make sure to select the base GPT model you used to create your Azure OpenAI Service deployment

well, if the editor is not using GPT4.
What do we do in Azure then, should we setup two endpoints to have conv and editor mode running?

GPT4 is way superior in code-generation tasks

@LL1yt
Copy link

LL1yt commented Oct 12, 2023

Is it possible to add a setting that will switch the model for editor view (let there be a warning for potential problems)? I'm willing to accept temporary inconvenience, but on the other hand I can use gpt-4 at my own risk in this mode.

@alanwilter
Copy link

I'd definitely love to have GPT4 in the Editor view.
And to what extension can GPT4 API be compared to Chat GPT Plus? Is there a way of passing images?

@genieai-info genieai-info added the triage Issue that is to-be-triaged and prioritized accordingly label Nov 24, 2023
@cloudbring
Copy link

This is still broken for commit message even if you change the model to GPT4 (mine is currently gpt-4-1106-preview) but, commit message still uses text-davinici-003

@martinguenther
Copy link

text-davinici-003 is deprecated since 2024-01-04 (see https://platform.openai.com/docs/deprecations/instructgpt-models).
I get the following error when working in editor mode:
An error occured. ChatGPT error 404: { "error": { "message": "The model text-davinci-003has been deprecated, learn more here: https://platform.openai.com/docs/deprecations", "type": "invalid_request_error", "param": null, "code": "model_not_found" } }
This seems related to #155.
Did I miss something, or is the editor mode currently not usable due to it being defaulted text-davinci-003?

@cmp-nct
Copy link

cmp-nct commented Jan 30, 2024

Just use continue.dev, they are far better than Genie

@murali-marimekala
Copy link

text-davinici-003 is deprecated since 2024-01-04 (see https://platform.openai.com/docs/deprecations/instructgpt-models). I get the following error when working in editor mode: An error occured. ChatGPT error 404: { "error": { "message": "The model text-davinci-003has been deprecated, learn more here: https://platform.openai.com/docs/deprecations", "type": "invalid_request_error", "param": null, "code": "model_not_found" } } This seems related to #155. Did I miss something, or is the editor mode currently not usable due to it being defaulted text-davinci-003?

I am observing the same issue. Any help ?

@5tigerjelly
Copy link

same issue, I guess it isn't maintained anymore?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
bug Something isn't working triage Issue that is to-be-triaged and prioritized accordingly
Projects
None yet
Development

No branches or pull requests

10 participants