It seems that this gguf weights run almost as the same speed as original bf16
#1
by
for1096
- opened
I'm trying 720 * 1280 * 33, it generates on 85s/it, so for 20 steps around 1800 seconds for each video on my 4070.
But when I use the original bf16 weights it perform almost the same.
Now I wonder if the gguf weights are loaded and converted to bf16 to run? I don't know, I'm not familiar with this.
Just wondering if anyone else feel the same confusion.
okay it seems that it's that the gguf weights don't fit into my vram since it only 12gb, so it's my hardware problem
for1096
changed discussion status to
closed