Error in input stream... all day long

Nvm. The “fix” with 4096 maxTokens worked for a few hours, then the same error as before we set a maxTokens variable re-appeared for a few hours, and now it works again… no code changes.

Unstable.

Facing the same issue with my custom GPT, it throws error in input stream as soon as it reaches 1k tokens in output (model limit is 8k).
Was working seamlessly a couple of days ago.
Is there some place we can report this behaviour?

@the.brainiac are you using Azure OpenAI?

I made a thread in the Microsoft forums, since we use Azure OpenAI. Hope we get some answers there. If anyone else in this thread also use Azure and have the same issues as we do, please click the “I have the same question” button in this thread:

Actually my error is related to custom GPT. (GPT Builder)
I’ve noticed that in recent days GPT-4 is unable to output more than 1k tokens.
Have there been any recent updates to the models?

Seems like it?
This was posted on the OpenAI blog yesterday:

But should it impact users of “regular” GPT-4?

1 Like

I think they nerfed GPT-4 just to make turbo look good :slightly_smiling_face:

3 Likes

I have not experienced any “Error in input stream” errors so far today.
Did this issue suddenly resolve itself? Without any explaination?

I made a custom GPT to help me study for Calc. Last time I used it nonstop for 24 hours right before my final and passed with a 95%. Now I thought I would do the same but for all my exams. Last time I could crank out easily 20 msg’s per hour with no problem and no limit reached (a bit forgetful but a small price to pay).
Now I cant get passed 12 messages sent on my end before I get am constantly getting this error. Not that I ever should have been gambling my calc grades with GPT…

1 Like

Only just recently got plus. This is all that I get all day long. Get about 5 - 10 prompts in than get the “Error in input stream” message and nothing I do will fix it. Even editing the message and saying something completely different does nothing.

Even tried waiting a few hours then Regenerate. Nothing.

I think I may have found a solution. This was only tested once, but I clicked on regenerate and before it could finish, I stopped the generation. I’m not sure if it was mentioned here or not and I don’t have time to read this all (exam), but after interrupting the generation, so far I’ve been able to continue without issue so far (~15min).

Edit: That particular GPT session worked without issue for the remainder of the night and morning, however I dont think cramming the night before is going to work anymore and I 100% failed the exam :sob:

1 Like