eager heath
Applio detects my GPU and has even used it once or twice before but i've found that it has completely started neglecting my GPU and has only been using my CPU. The training started off fine but now ive been stuck on 317 epochs for the past 30 mins, please help.
I'm on 3.2.9 btw
my first few trains went smooth but now it only wants to use my cpu and i can't find a fix anywhere...
tepid lotus
oops, never mind
go to Applio, Training page, expand advanced settings at the top
show what is there
4GB is not really enough to train anything
unless you enable bf16 and checkpointing
tepid lotus
okay, so it detected your gpu and should be using it
but, as I said, 4GB is barely enough
eager heath
yeah thats what i thought but it hasnt been the entire time
tepid lotus
batch 4 maybe
eager heath
i think i was using batch 4
eager heath
alright i will
tepid lotus
alright i will
if you see anything significant in shared memory
you're over the limit
and your traning will be slow af
eager heath
alright
placid lark
yeah thats what i thought but it hasnt been the entire time
not only that but your laptop could also overheat in the long training session esp if you concern about the lifespan
last steppe
Applio detects my GPU and has even used it once or twice before but i've found t...
Training on the laptop GPU is possible, but often slower than that desktop counterpart. The VRAM size of this GPU, the dedicated VRAM 4GB is considered very low for any training task, while another shared 3.9 GB can either come from your main PC RAM or the integrated GPU (Intel UHD Graphics). While there is 7.9 GB memory combined, the program might still use the main GPU VRAM (4GB) for most parts, which would still be slow for training, regardless of any setting like batch size.
