Pretraining Problem

[Ismail111Participant]

Hi guys! I’ve never created a video with Deepfacelab before. First of all, I heard that if I do pretraining, it will be faster and easier every time I make a df video. When I open the Deepfacelab folder, I directly run the 6) train SAEHD file. But I get an error at the last part because of the values I entered.

My System Features:

My Graphics Card: RTX 3060
11th Gen Intel(R) Core(TM) i7-11800H @ 2.30GHz 2.30 GHz
RAM: 16GB

Can you tell me what path I should follow and what values I should enter?

[deepfakeclubParticipant]

i always start with the default value’s

also if you are a beginner i would not reccomend trying to do all the complicated stuff at once

however this is what i would reccomend

train using the default values

delete the model that is already in the model folder

create a completely new one or import a pretrained model that you can find somewhere

this site offers really good comunity made models

also if this didnt help

check your workspace folder and see if everything is correct in there

[Ismail111Participant]

I deleted the previous model and made a new model. I tried to do SAEHD training again, but I got the same error again. I can’t find out why I’m getting this error. I’m about to go crazy…

[n] Enable pretraining mode ( y/n ?:help ) : y
Initializing models: 80%|##################################################4 | 4/5 [01:31<00:22, 22.88s/it]
Error: OOM when allocating tensor with shape[131072,256] and type float on /job:localhost/replica:0/task:0/device:GPU:0 by allocator GPU_0_bfc
[[node src_dst_opt/ms_inter_AB/dense1/weight_0/Assign (defined at C:\Users\Ersin\Desktop\DeepFaceLab\DeepFaceLab_NVIDIA_RTX3000_series_build_11_20_2021\DeepFaceLab_NVIDIA_RTX3000_series\_internal\DeepFaceLab\core\leras\optimizers\AdaBelief.py:37) ]]
Hint: If you want to see a list of allocated tensors when OOM happens, add report_tensor_allocations_upon_oom to RunOptions for current allocation info. This isn’t available when running in Eager mode.

[mrthongParticipant]

Lower your batch size

[Author]

Posts