i am using the api key of nvidia nim models and suddenly i see that the deepseek v4 pro has been removed and replaced with deepseek v4 flash 0731 model. That model is giving random output and is hallucinating a lot. Is anybody facing the similar issue or is it only me?
I thought I was the only one that was having that issue. It was working fine yesterday for about two hours, and then all of a sudden, I get maybe one or two prompts in, and it goes completely crazy and just start spitting out random stuff that has nothing to do with what I’m trying to work on. I tried to troubleshoot it, but nothing would fix it.
I’m having the same problem; the model can’t generate 3 stable outputs without starting to “hallucinate,” reverting to GLM 5.2.
theyre still testing the model, its available to the api but not exposed to the nvidia models website for a reason, might not even be the right model yet till they release it online
I have been trying it since they made the switch 2 days ago and for the moment i haven’t got any issue. The answers it gives me are pretty similar to the ones V4 Pro used to give me so far.
Although it would be nice if NIM’s devs could give us an explanation on why V4 Pro got removed out of the blue (i get V4 Flash as it got updated, but why Pro?.)
theyre large models, they needed space for deepseek flash 0731 and probably an upcoming larger flagship model