What -ctx and -chunks parameters did you use to make the iMatrix of the Lllama 2 70b?

#9
by Nexesenex - opened

You should share your parameters to make your iMatrix, man, because you obviously know best. ^^

According to this thread I don't really know. Tons of other people who know better :-)

But if you want to know what I have done, I have simply created the importance matrix using

./imatrix -m <some_model> -f tests/wiki.train.raw -o some_model.imatrix --chunks 100

and then used it to quantize the model

./quantize --imatrix some_model.imatrix <some_model> iq2xxs.gguf iq2_xxs

Sign up or log in to comment