What are your settings to generate your importance matrix?
#8
by
froggeric
- opened
It seems your importance matrix quantisations are much better than others; congratulation!
Would you mind sharing the settings you use to train the imatrix, and your input file?
I used these settings, with the random words I found in a discussion on the llama.cpp github, I didn't create it, I just used it.
./imatrix -m /run/media/knut/HD/saily.fp16.bin -f /run/media/knut/HD/llama.cpp-current/20k_random_data.txt -o imatrix_sally.dat --chunks 100 -t 15
I will upload the text file into this repo now
KnutJaegersberg
changed discussion status to
closed