Imatrixes

#1
by Nexesenex - opened

Hey Richard,

Could you share your iMatrix files when you share a gguf repo, and tell which calibration data you used, how many chunks and with which context size?
Several of the LCPP tensor quants and quants strategies are becoming obsolete, while the iMatrix files are worth being kept and shared for future quantization types and quant strategies/custom quants.
Thanks in any case!

Hi, not doing imatrix as I dont have compute for that, so cant share

Any recommendations for dataset in case I manage to get compute?

This is the standard dataset used by serious quantizers who have the necessary compute :
https://gist.github.com/bartowski1182/eb213dccb3571f863da82e99418f81e8

Ok thank you! One day... one day...

Sign up or log in to comment