Selfhosted instance cannot use gpt-3.5

dglaefcke commented 1 year ago

Please check that this issue hasn't been reported before.

[X] I searched previous Bug Reports didn't find any similar reports.

Expected Behavior

Hello, I am currently facing a problem with the key I provide via .env file. The key we provide is not allowed to use gpt-4 (we are on the waitlist), but as far as I understand, you can select the used model in the preferences, where we selected "gpt-3.5-turbo". So I expect it to use gpt-3-5-turbo model, instead of the gpt-4.

Current behaviour

When we start an agent, it directly prompts: "ERROR your API key does not have GPT-4 access. You must first join OpenAI's wait-list. (This is different from ChatGPT Plus)" Is it possible to use another model or is it currently only usable with gpt-4, even if we set it on another model?

Steps to reproduce

Provide Key via .env file
Make sure it prompts, that the env file is used
Start browser and locate to the instance
Login
Change settings to use gpt-3.5-turbo model
Create an agent

Possible solution

No response

Which Operating Systems are you using?

[ ] Android
[ ] iPhone/iPad
[X] Linux
[ ] macOS
[ ] Windows

Acknowledgements

[X] My issue title is concise, descriptive, and in title casing.
[X] I have searched the existing issues to make sure this bug has not been reported yet.
[X] I am using the latest version of AgentGPT.
[X] I have provided enough information for the maintainers to reproduce and diagnose the issue.

JPDucky commented 1 year ago

Try running the startup script, ./setup.sh instead directly making the .env file. It will prompt you to paste your key at start. Let me know if it's still acting up after that

realblackbot commented 1 year ago

I have exactly the same problem, and I've added my OpenAPI key using the setup.sh script. Tried to google this, but all the suggestions (add billing to my openai account, make sure the correct api key is used... etc) are already fine, still this error keeps making my AgentGPT instance unusable :(

realblackbot commented 1 year ago

@dglaefcke Did you manage to find a solution please? 🙏

dglaefcke commented 1 year ago

@realblackbot no sry, I had no time for the last week. I'll post the results of my next test session :)

jasangill1 commented 1 year ago

Hello @dglaefcke are you still experiencing these issues if so please private message me on Discord my at is @namesnipe or join the reword discord https://discord.gg/7ya2Kp3T so we can troubleshoot with your API Key

Looking forward to hearing from you!

jasangill1 commented 1 year ago

Hello, @dglaefcke @realblackbot would one of you mind sharing some console logs for this issue??

realblackbot commented 1 year ago

Hi @jasangill1, thank you very much for looking into our problem! 🙏 I attach the console log both from the frontend (browser) and the backend (linux cli). If you need anything else, I would be happy to do what you want if you instruct me! 😉

back-end-console_2023-07-02_2144-1.txt front-end-console-export-2023-7-2_21-44-1.txt

jasangill1 commented 1 year ago

Hello @realblackbot Seems like your next.js started correctly, Could you provide me with your network tab console logs as well!! This will help me get to the root of the problem!!!

realblackbot commented 1 year ago

Sure, here it is. Thank you very much again! 🙏 192.168.64.6_network_tabexport[23-07-04 13-27-12].har.zip

jasangill1 commented 1 year ago

Hello @realblackbot Ill check you logs and get back to you ! In the mean time could you try running it in production ? Seems like this works for alot of users !!!

realblackbot commented 1 year ago

@jasangill1 This is my 'production' env, as I just want to run the AgentGPT backend in an Ubuntu Server VM on my laptop. 😉 Thank you very much again! 🙏

jasangill1 commented 1 year ago

Hello @realblackbot could i pick your brain a bit and ask why do you wanna run the backend on the VM? Seems like there are a lot of 'handshakes' happening which makes it hard to replicate

realblackbot commented 1 year ago

Hey, @jasangill1, I just wanted to contain it, that's why I used a VM. But if you say I have a chance this issue is solved if I run it natively on my mac, then I will do that!

jasangill1 commented 1 year ago

Hello @realblackbot, Yeah please try to run it natively on your mac and let me know if that made a difference!! Ill try to replicate this error and get the bottom of this!!

realblackbot commented 1 year ago

@jasangill1 I've managed to solve it by installing AgentGPT on an x86 Ubuntu Server pc box instead of on an Ubuntu for ARM VM on my MacBook Pro with an Apple M1 (ARM) CPU. I'm not sure the issue was caused by the 'exotic' CPU or a misconfiguration from me, but hey, it's working in the end! 😎

Thank you very much @jasangill1 for spending time on the problem, your help is highly appreciated! 🙏

tdpires commented 1 year ago

I am having the same issue on Ubuntu server. Model in settings says gpt-3.5-turbo and I have the correct API, but it outputs: ERROR your API key does not have GPT-4 access. You must first join OpenAI's wait-list. (This is different from ChatGPT Plus) I am on the waiting list for gpt-4 API access

jasangill1 commented 1 year ago

Hello @tdpires If your on the waitlist you wont have access to gpt-4 API until it has been approved !!

reworkd / AgentGPT