hollowstrawberry / kohya-colab

Accessible Google Colab notebooks for Stable Diffusion Lora training, based on the work of kohya-ss and Linaqruf
GNU General Public License v3.0
599 stars 87 forks source link

Lora trained first time > not works ..Second try > lora with noise in epoch 10 NOT WORKS #100

Closed manuelpepe2024 closed 6 months ago

manuelpepe2024 commented 6 months ago

sorry for that but saddly not works ....and i dont understand why i use it in collab a few days ago and works amazing as always ....please if someone can tellme how to use old version while get fix ...please ...ayuda por favor

ArmyOfPun1776 commented 6 months ago

After an update to Kohya_ss a few days ago the trainer broke. The Creator is aware of the issue and is working on getting it fixed.

kaekitsune commented 6 months ago

set the network alpha to 1 (for now) :)

ArmyOfPun1776 commented 6 months ago

DO NOT SET THE NETWORK ALPHA TO 1. You are not fixing the issue! Your LoRa will be poor quality unless you're Training something super generic like a ball.

Alpha Net should be set to at least half of the Net Rank. (i.e.: 32R-16A, 64R-32A, 128R-64A).

Whatever you're trying to train can wait a couple of days. Don't sacrifice Compute time for nothing. Just wait.

ArmyOfPun1776 commented 6 months ago

I feel like I need to prove a point here. So here is an example for those telling people to use Net Alpha 1. These are "Bare Generations". ( meaning LoRa::.7, Activator )

Here are the first 3 bare generations with a LoRa trained at Net Rank 32 Net Alpha 1:

00000-1174718702 00001-265992758 00002-4116064999

Her are the first 3 bare generations of my first LoRa ever. It is, by far the worst quality LoRa I have ever made without trying. Net Rank 32 Net Alpha 16:

00003-3634950913 00004-2937727342 00005-3452292457

DON'T TRAIN AT NET ALPHA 1.

tglunzer01 commented 6 months ago

Just want to confirm my LoRa models are not working after training here. Never had issues before. So hopefully it get's fixed soon.

hollowstrawberry commented 6 months ago

https://github.com/hollowstrawberry/kohya-colab/issues/98