Commit e4ff3dc

committed

iq1s_blocks16: uint32_t codebook is also better in CUDA

TG-128 is now 204 t/s up from 194 t/s. PP-512 is 5890 t/s, so significantly better than other quants

1 parent d090b0c commit e4ff3dcCopy full SHA for e4ff3dc

1 file changed

+278

-525

lines changed

+278

-525

lines changed

Comments

(0)