#500 Error
foggy sequoia
ah i had to make a thread
I don't see what's wrong with it and I assume I can't do anything about it since it's an issue on Google's end, but is there any way around it?
scenic pond
that's interesting...
foggy sequoia
yea, cause it's something that surely would have happened to more people before (even though this has only been out for a week), but I can't find any examples of people doing this the correct way, if there's any
uneven burrow
Yeha. So Ik this error
It's mainly becz you internet is slow or the api timed out
Fix your internet or try using VPN in a place where Gemini isn't restricted.
Or maybe, google servers had issue
foggy sequoia
a request timeout wouldn't be a 500 error, and my internet is fine. Also, loading any image, even pretty big images, isn't an issue if i send it as the first message. The issue happens in this specific case when i send the image in the second message
uneven burrow
Another thing, you didn't mention the model in your code
Gemini with vision cannot have multi turn chats
uneven burrow
Nah.
foggy sequoia
when i send just text in my second message it does work fine
when contents is this:
contents: [
{ role: "user", parts: [{ text: "hello?" }] },
{ role: "model", parts: [{ text: "hi, what can i help you with?" }] },
{ role: "user", parts: [{ text: "how are you?" }] },
]
it works fine
foggy sequoia
does it fallback to gemini-pro if it's just text?
uneven burrow
Not gemini-pro-vision
foggy sequoia
but the model i put is pro vision
uneven burrow
BECZ gemini-pro-vision cannot have multi turn chats along with images
foggy sequoia
huh
so it does fallback to just gemini-pro if there's only text and otherwise it fails?
i feel like that's not properly explained in the docs if that's the case
and surely they should change it to return a 400 message if it isn't meant to take multiple messages
https://huggingface.co/spaces/ysharma/Gemini-PRO-Vision-Chat well this works
so what's going on
odd coyote
Vision is not multi turn now.
YSharma's implementation on hugging face is a nice workaround.
Not sure what's happening with Vertex AI's SDK or service. It should not be a 500 error.