GPT 4 Vision examples do not run
NightMachinery opened this issue · 6 comments
What happened?
$ go install github.com/tbckr/sgpt/cmd/sgpt@latest
$ sgpt -m "gpt-4-vision-preview" -i "https://upload.wikimedia.org/wikipedia/en/c/cb/Marvin_%28HHGG%29.jpg" "what can you see on the picture?"
unknown command "what can you see on the picture?" for "sgpt"
How can we reproduce this?
_
sgpt version
version: dev
commit: unset
commitDate: unset
### Search
- [X] I did search for other open and closed issues before opening this
### Project
- [X] I have verified that I am using the Go implementation of SGPT (this project) and that it is not another project
### Code of Conduct
- [X] I agree to follow this project's Code of Conduct
### Additional context
_No response_
Hey, thanks for your bug report.
It seems like you are using an old version of SGPT. Please install the latest version of SGPT and test again.
Ah, I think you found an error in the docs.
Could you please try the following command to install the latest version:
go install github.com/tbckr/sgpt/v2/cmd/sgpt@latest
I am going to close this issue. If it still does not work, please feel free to reopen the issue.
@tbckr Indeed, the problem went away when I installed v2. However, the outputs are truncated very shortly for my requests:
> command sgpt --temperature 0 --max-tokens 40000 -m gpt-4-vision-preview -v -i 1.png 'This is a screenshot of a part of a paper. Rewrite it in org-mode. You can use `\( ... \)` to insert inline LaTeX.'
time=2024-01-16T10:38:20.785+03:30 level=DEBUG msg="Loading config"
time=2024-01-16T10:38:20.785+03:30 level=DEBUG msg="Config file not found - using defaults"
time=2024-01-16T10:38:20.785+03:30 level=DEBUG msg="No mode provided via command line args - using default mode"
time=2024-01-16T10:38:20.785+03:30 level=DEBUG msg="Chat session manager initialized"
time=2024-01-16T10:38:20.785+03:30 level=DEBUG msg="OpenAI client created"
time=2024-01-16T10:38:20.785+03:30 level=DEBUG msg="Loading personas from path: /Users/evar/Library/Application Support/sgpt/personas"
time=2024-01-16T10:38:20.785+03:30 level=DEBUG msg="could not find custom persona"
time=2024-01-16T10:38:20.785+03:30 level=DEBUG msg="Loaded default prompts"
time=2024-01-16T10:38:20.785+03:30 level=DEBUG msg="No persona provided"
time=2024-01-16T10:38:20.785+03:30 level=WARN msg="The GPT-4 Vision API is in beta and may not work as expected"
time=2024-01-16T10:38:20.785+03:30 level=DEBUG msg="Added prompt message"
time=2024-01-16T10:38:20.785+03:30 level=DEBUG msg="Added prompt message"
time=2024-01-16T10:38:24.175+03:30 level=DEBUG msg="Received response"
Certainly! Here is the content of the image written in Org-mode with LaTeX:
time=2024-01-16T10:38:24.175+03:30 level=DEBUG msg="Printed response"
time=2024-01-16T10:38:24.175+03:30 level=DEBUG msg="Received message from OpenAI API"
Should I open another issue?
Hi @NightMachinery,
I oberserved the same issue. I think it is an issue in the OpenAI API. You can open an issue if you want to.
@tbckr No, I am certain it's a bug here. I have sent requests using their own Python library and it's okay. But the default for the max returned tokens is different in the API (and much lower). So there is a need to set this parameter correctly.