Appear API is returning Internal Server Error when the request is getting higer thant 202k token, please nvidia can fix that or let us know the real context windows limitation instead say 1M ?
for real
Experiencing the same, happened at 170k.
One question, @bozoweed. Are you still experiencing the same issue? From what Iāve seen, youāre not the only one.
The real question is: are they going to fix it? Will it stay like this? Or is it even a bug in the first place, or is it intentional?
Because if the model is advertised as supporting a 1M token context window, then the expectation is that it should actually provide that capacity, right?
Iām asking mostly out of curiosity, since the model is still relatively new on NVIDIAās infrastructure. I assume they may still be making adjustments or changes.
they can let it like this like they have done with deepseek v4 pro/flash, they are also not accept the 1M context as excepted ā¦ itās nvidia stuff and appear they donāt give a ā ā ā ā
Since yesterday I have only Internal Server Error or Timeout connection when working with model GLM5.2 in every scenario (I use kilocode)
i have keep working on roo-code from my side and still working ( really slow but working )
Do you change settings like connection timeout or api response timeout?
i donāt think i have juste made few change to follow vs-code update
Right now itās unusable, showing error āDEGRADED function cannot be invokedā
maybe we should mail bombing the support to froce them to fix the model š¤·āāļø
Is this the reason it keep getting āmessageā:āRequest failed with status code 400ā error on sillytav?
I think NVIDIA team works on this because my problem from yesterday was solved since some time. But now ādegraded functionā is coming))))
May be GLM5.2 change something and nvidia nim team canāt be in time with some wrapper
So I guess we have to wait and suffer
And communicate there to show our faith in best future :)
why it so fast for me
maybe becaause you are lucky, if itās so fast you can maybe try to ask them to fix that please š
idk it response time hover around from 1 - 7 second. Also how to ask them to fix?
using the support email help@build.nvidia.com
they have to fix the context to 1M but also the others errors , itās completely broken since early this morning.
yes. only 202k context windows.
and DeepSeek flash/Pro only 256K context
Working fine now so far, thanks for the help and support!