Quantization improvements for k_quants#2707

Merged

ikawrakow merged 12 commits intomasterfrom

ik/better_q234_k

Aug 22, 2023

Commits on Aug 22, 2023

Improve LLaMA-2 2-, 3- and 4-bit quantization
Iwan Kawrakow
committed
Minor 4-bit quantization improvement
Iwan Kawrakow
committed
Some more fine tuning
Iwan Kawrakow
committed
Adding make_qkx2_quants
Iwan Kawrakow
committed
Another minor improvement
Iwan Kawrakow
committed
Q2_K improvement
Iwan Kawrakow
committed
Iterating
Iwan Kawrakow
committed
Revert Q5_K back to make_qkx1_quants
Iwan Kawrakow
committed
Better Q6_K
Iwan Kawrakow
committed
make_qkx2_quants is better for Q5_K after all
Iwan Kawrakow
committed
Fix after rebasing on master
Iwan Kawrakow
committed
Fix for changed tensor names
Iwan Kawrakow
committed