Even the Vercel online version is showing same Error: Something went wrong with the request, please check the Network log

[KannamSridharKumar]

I have credit in my OpenAI account and I'm using a short text. The error is not due to my OpenAI account or quota limit. Few others have also reported the same issue.

Same error with online version too - https://graphgpt.vercel.app/

Thanks.

[KannamSridharKumar]

After some debugging I found the issue is
"openai fetch failed ConnectTimeoutError: Connect Timeout Error"

I'm not an expert javascript, I tried a few suggestions made online but still couldn't fix the issue.

[moaldeen]

have you solved the problem

[KannamSridharKumar]

no, I couldn't

[Xemin0]

I've checked the network log, the fetch request returns a 404 error at the OpenAI Completions endpoint.
I have checked the default parameters in src/App.js, seems like text-davinci-003 is alrdy deprecated on Jan 4 2024.

The solution might be replacing that model with simply gpt-3.5-turbo or anything else in the list

and also here mentioned some models for language inference

[KannamSridharKumar]

@Xemin0 In the past I've updated the model, it didn't work. I tested again just now with latest models. Same error.
Pls let me know if it works for you and what changes you made.
Thanks,

[Xemin0]

@Xemin0 In the past I've updated the model, it didn't work. I tested again just now with latest models. Same error. Pls let me know if it works for you and what changes you made. Thanks,

sry, I was running on an old mac and was having some trouble installing the dependencies. I did not test it myself.

Now as everything's settled on my old machine, I just tested a few models on the Completion Endpoint, here are some results:

• davinci-002 : SyntaxError: Unterminated string in JSON at position 3062 (line 1 column 3063)
• babbage-002 : works
• gpt-3.5-turbo-instruct: works
• new models : 404

I also tested the Chat Completion Endpoint with new models

• gpt-4o-mini : 400 bad request

so i guess babbage-002 and gpt-3.5-turbo-instruct are the ones that are working. using davinci-002 may need some changes in the code that handles the JSON part.

while if to use the chat completion endpoint, the request header might need some modification (I did go through the API document yet).

I am not an expert in Node.js so I didn't change anything else other than the requested model string, but let me know if the two models work on your end.

[Xemin0]

btw I tumbled upon this endpoint compatibility list, where it does say, davinci-002, babbage-002, gpt-3.5-turbo-instruct are the supported models among the latest ones.

[KannamSridharKumar]

many many thanks, gpt-3.5-turbo-instruct is working for me too.

I never used node.js and I spent hours trying to fix the code and gave up.

btwn this is my youtube channel - https://www.youtube.com/@SridharKumarKannam

Thanks a lot again...