master-cb40dfc
cb40dfca · llama : only use Q6_K for output weights if tensor size is multiple of 256 (#1932) ·