For AI Studio truncation updates, join

[unkempt palm]

🤞

[full zinc]

🙂

[dry slate]

Third

[glad oak]

fourth!

[opal vessel]

Fifth?

[cosmic lintel]

6?

[opal vessel]

Seems to be working atm

[novel hemlock]

fr?

[mellow iris]

Seems to be working a lot better at least

[dry slate]

Wait...

[wet field]

yeah seems to be working for me a lot more, doesn't take 300+ seconds to generate a response and finishes the message

[timber ember]

If it's available for some of you, but not all... keep in mind that they may be rolling it out slowly and you got lucky which node you hit. {:

[novel hemlock]

yaaayyyy

[dry slate]

[mental otter]

not solved here 😦 will keep waiting

[opal vessel]

Yeah it works when it works /:

[full zinc]

nvm, i got my hopes up, it only worked for the essay prompt.

[dry slate]

4 out of 4 responses working for me bless

[restive canyon]

That worked for me, interesting...

[dry slate]

Seems like not a fluke for me 🙏 thank god

Or google

Idk

[full zinc]

interesting, the response is complete for me now

[full zinc]

same prompt as this

[full escarp]

i noticed that it tends to complete the whole thing if you lower the safety settiings

set all of them to block few

[full zinc]

alright

[full zinc]

for this i didn't change any settings btw

might been a fluke but i used it 3x now and all gave complete response

[snow arrow]

Still truncating for me, nomatter the prompt 😭

[full escarp]

@snow arrow try this

[snow arrow]

Already did, thx

[full escarp]

did it work ?

[snow arrow]

Aha! Refreshed the page, working now 👍

[timber ember]

If it is still rolling out - it may be hit or miss. Really. {:

[unkempt palm]

Hm, looks like I can't use (at)everyone. The production push should be done, but there might be some lagging bits. Try a refresh or cache clear as well!

[full escarp]

amazing

[full zinc]

peepohypers

ctrl + shift + r to reload it and override the locally cached version

[dry slate]

[torpid sage]

it was working but after like 9 messages it stopped working and started getting cut off again.

[unkempt palm]

Yes, I just heard that the fix didn't actually finish deploying due to the other issue around the API requiring billing :/ so this is still an issue. This is an official OMG for the team.

[torpid sage]

ok tysm for the clarification

[vocal stag]

anther way to refresh cache on a page in chrome is press f12 then you can right click on the reload circle arrow on the url bar and > empty cache and hard reload

tho im still not having any luck getting 1.5 to work in aistudio

still truncated 😦

[torpid sage]

i think im probably just gonna wait till tommorow/day after tommorow and try again then

[dry slate]

[unkempt palm]

Hi all, confirming we're aware the issue is still happening. I'll update again if there's anything meaningful to report.

[sacred lotus]

My hypothesis is that one of the experts have scaped the MOE architecture due to hard working conditions. find that expert and bring it back to work 😂

[pastel sphinx]

sadly the 1000 word essay thing doesn't seem to work for me

[torn turret]

hey, same issue here. Gemini 1.5 takes super long time to generate answer (at least 50 secs) even for simplest questions

and its response are always short and incomplete

[viral pebble]

you'll see above that they're still working on it

[restive idol]

[vocal stag]

We back y'all full length messages working! Thanks google team!

[dry slate]

🙏

[unkempt palm]

Hi all! I have not gotten official confirmation that things are resolved, but it sounds like some of the mitigation work has been helpful based on recent messages here? Fingers crossed!

[vital merlin]

It's been working well for me for the past few hours (although I've only actively been using it for the past hour)

[dry slate]

Works for me except for 1 specific statistics question for some reason

[cosmic lintel]

i am just having the same issue again

[indigo geode]

likely not the same issue. notice the alert sign next to the model. hover with mouse to see details

[novel hemlock]

it probably cutoff answering in the middle because the response it was generating might be considered harmful (according to the model)

[unkempt palm]

Hi all; realized I never mentioned but the fix should have been fully deployed as of ~9:30 PM PT yesterday (conveniently, 24 hours ago!). Please let me know if you're having any issues.

[grave beacon]

Hi WIlliam, yes i`m still having problem with the gemini api for web client, javascript... can you assist me?

i'm using the import map "@google/generative-ai": "https://esm.run/@google/generative-ai"

provides in google tutorial in "gemini for web"

Or if someone here can help me ... haha

[timber ember]

That's a question for the broader changer, not this thread. More people may be watching that.

[torn turret]

hey, just tried the gemini 1.5 again, and even a simple prompt will cause internal error

[timber ember]

Do you have multiple logins in this browser?
GMail or Workspace account?
Try in Incognito Window

[torn turret]

great, works now, thx!

btw, although the response now is much better as for the completeness, sometimes it'll still cut off after generating ~150 tokens

[timber ember]

Make sure you Send Feedback when it does. Make it clear in the report this is new.

[torn turret]

done🫡

[unkempt palm]

@torn turret Does it happen consistently? If so, what is the prompt you're using?

[torn turret]

2-3 times in total. It’s a pretty simple prompt
“Give a comprehensive intro tutorial of coding with c++, phrase in an organized way”

[pastel sphinx]

Working a bit better today, but now I'm getting a huge lag problem when going over 40K tokens

On a machine with 64GB ram and a 5800X... 😬

[torpid sage]

I might be wrong but I think the processing is cloud side

[unkempt palm]

Thanks, and this is with 1.5 Pro?

[torn turret]

yup

[unkempt palm]

Trying the prompt you shared (“Give a comprehensive intro tutorial of coding with c++, phrase in an organized way”) and I'm seeing full responses right now. Is this still an issue for you?

[pastel sphinx]

Yeah, but clearly something is being stored locally

Given conversation history has to be rendered on the client

And the more I add, the laggier it gets and the more ram the tab uses

[unkempt palm]

How are you using AI Studio where you're seeing this performance issue? A chat prompt with a big upload gets you to 40k tokens immediately? Or a long conversation?

Also, what browser/OS are you using?

[timber ember]

We had a report in the main chat about Firefox and Safari both taking up a ton of memory when adding a video.

(Just as a data point)

[torn turret]

Just tried again, looks great to me this time👍

[pastel sphinx]

The latter, in Firefox/Linux