Uri Alon
Uri Alon
>My understanding is therefore, that it is only a _FAISS index of the state_, $h_e$ that gets added to the datastore. (Thus, as per the paper, neither the keys $W_k...
Hi @cem2ran , Thank you for your interest in our work. I'm not sure. Did it just hang after loading the checkpoint? Did it print any stack trace when you...
Thanks @SharkWipf , it would certainly help.
Wow, thanks a lot @SharkWipf . If you manage to run more experiments we (and I'm sure that other users) would love to hear about them.
Hi @AshwinRamachandran2002 , Thank you for your interest in our work! Your reading is correct, and you are looking at the right places in the code: 1. In assigning a...
Hi @jordancole21 , Thank you for your interest in our work! The idea there is that we are duplicating the model before overriding its functions to inject Unlimiformer. One quick...
Sorry, I checked and my previous suggestion doesn't work. Do you have any idea of how to duplicate the model object, including its quantization settings? If not, I'd recommend just...
What kind of cuda errors? Out of memory?
Hi @jordancole21 , I don't know. I'm not sure whether the 4-bit is the problem, or is it anything else. Did you get any stack trace? I wonder if the...
I can't tell which step is it in, because `unlimiformer.py` does not appear in the traceback. Without Unlimiformer - does it work well?