Software Engineering Courses (SECourses)•3y ago

@Discord Admin I am having this problem, asked you some days ago in youtube comments, can you help m

@Discord Admin I am having this problem, asked you some days ago in youtube comments, can you help me here please?

felipeabdoOP•2/15/23, 11:57 AM

I was using Stable Difffusion normally like you demonstrated in the video, but today I got an error.

OutOfMemoryError Traceback (most recent call last)
<ipython-input-9-9817a4f7d2d0> in <module>
10
11 with autocast("cuda"), torch.inference_mode():
---> 12 images = pipe(
13 prompt,
14 height=height,

13 frames
/usr/local/lib/python3.8/dist-packages/diffusers/models/cross_attention.py in get_attention_scores(self, query, key, attention_mask)
221
222 attention_probs = attention_scores.softmax(dim=-1)
--> 223 attention_probs = attention_probs.to(dtype)
224
225 return attention_probs

OutOfMemoryError: CUDA out of memory. Tried to allocate 2.50 GiB (GPU 0; 14.76 GiB total capacity; 12.21 GiB already allocated; 1.22 GiB free; 12.26 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Now I'm only able to generate images with 256 x 256 px and lower.

Any idea how to fix this and generate images of 512 x 512 again?

Thank you!

MMinhalHnaqviOfficial I am 3d Artist and my brother is React,html.css,java developer

Furkan Gözükara SECourses•2/15/23, 11:59 AM

nice

SSesarr Is the art workflow based on stable diffusion?

Furkan Gözükara SECourses•2/15/23, 11:59 AM

not at all atm

Ffelipeabdo @Discord Admin I am having this problem, asked you some days ago in youtube comm...

Furkan Gözükara SECourses•2/15/23, 12:00 PM

welcome

Furkan Gözükara SECourses•2/15/23, 12:00 PM

this is in google colab?

felipeabdoOP•2/15/23, 12:00 PM

yes@

Furkan Gözükara SECourses•2/15/23, 12:00 PM

if so what is your batch size

felipeabdoOP•2/15/23, 12:00 PM

yes!

Furkan Gözükara SECourses•2/15/23, 12:00 PM

show me settings please

felipeabdoOP•2/15/23, 12:00 PM

felipeabdoOP•2/15/23, 12:01 PM

!accelerate launch train_dreambooth.py \
--pretrained_model_name_or_path=$MODEL_NAME \
--pretrained_vae_name_or_path="stabilityai/sd-vae-ft-mse" \
--output_dir=$OUTPUT_DIR \
--revision="fp16" \
--with_prior_preservation --prior_loss_weight=1.0 \
--seed=1337 \
--resolution=512 \
--train_batch_size=1 \
--train_text_encoder \
--mixed_precision="fp16" \
--use_8bit_adam \
--gradient_accumulation_steps=1 \
--learning_rate=1e-6 \
--lr_scheduler="constant" \
--lr_warmup_steps=496 \
--num_class_images=744 \
--sample_batch_size=4 \
--max_train_steps=4960 \
--save_interval=10000 \
--save_sample_prompt="felipe" \
--concepts_list="concepts_list.json"

Reduce the `--save_interval--save_interval` to lower than `--max_train_steps--max_train_steps` to save weights from intermediate steps.

`--save_sample_prompt--save_sample_prompt` can be same as `--instance_prompt--instance_prompt` to generate intermediate samples (saved along with weights in samples directory).

Ffelipeabdo !accelerate launch train_dreambooth.py \ --pretrained_model_name_or_path=$MODE...

Furkan Gözükara SECourses•2/15/23, 12:01 PM

ok this is training settings

Furkan Gözükara SECourses•2/15/23, 12:01 PM

you get error during training?

felipeabdoOP•2/15/23, 12:02 PM

no, I was using normally and suddenly one day it stopped working

felipeabdoOP•2/15/23, 12:03 PM

working only with 256 x 256 images or lower

Furkan Gözükara SECourses•2/15/23, 12:04 PM

show me

Furkan Gözükara SECourses•2/15/23, 12:04 PM

where you get error

Furkan Gözükara SECourses•2/15/23, 12:04 PM

when you click generate images?

felipeabdoOP•2/15/23, 12:04 PM

ok, I'll make a video from scratch

felipeabdoOP•2/15/23, 12:04 PM

just 1 sec

Furkan Gözükara SECourses•2/15/23, 12:04 PM

by the way

Furkan Gözükara SECourses•2/15/23, 12:04 PM

first try

Furkan Gözükara SECourses•2/15/23, 12:04 PM

num_samples = 1

felipeabdoOP•2/15/23, 12:05 PM

ah, ok!

Marcelo•2/15/23, 12:05 PM

Felipe are u brazilian?

felipeabdoOP•2/15/23, 12:05 PM

yep!

Marcelo•2/15/23, 12:05 PM

Show

felipeabdoOP•2/15/23, 12:08 PM

Tá funcionando pra vc ai direitinho?

FFurkan Gözükara SECourses num_samples = 1

felipeabdoOP•2/15/23, 12:09 PM

worked with one sample

Marcelo•2/15/23, 12:09 PM

Eu não cheguei a usar esse colab ainda, utilizo geralmente o do automatic1111 mesmo

FFurkan Gözükara SECourses are you using automatic1111 or what else?

Arvy•2/15/23, 12:12 PM

Google Colab

FFurkan Gözükara SECourses num_samples = 1

felipeabdoOP•2/15/23, 12:16 PM

but why this time with only one sample?

felipeabdoOP•2/15/23, 12:17 PM

and now it's not working even with 1 sample

GP•2/15/23, 12:44 PM

Hey all! Trying to setup controlnet with Automatic as per the latest video

GP•2/15/23, 12:44 PM

but I'm hitting an error I'm not sure how to fix

GP•2/15/23, 12:44 PM

once I hit generate I get this in CMD

GP•2/15/23, 12:45 PM

GP•2/15/23, 12:59 PM

Solved the first part of the problem, but now I have this:

Diverso (1050 Ti)•2/15/23, 1:10 PM

I can't connect to colab, it writes that I used limits of GPU usage((((

DDiverso (1050 Ti)I can't connect to colab, it writes that I used limits of GPU usage((((

felipeabdo•2/15/23, 1:10 PM

wait some hours and it will work again

felipeabdo•2/15/23, 1:11 PM

happened to me too

Ffelipeabdo but why this time with only one sample?

Furkan Gözükara SECourses•2/15/23, 1:12 PM

u have 10 samples not 1

GGP Solved the first part of the problem, but now I have this:

Furkan Gözükara SECourses•2/15/23, 1:13 PM

did you try with base sd 1.5

FFurkan Gözükara SECourses did you try with base sd 1.5

GP•2/15/23, 1:13 PM

Hi! Yeah, I'm using v1-5-pruned.ckpt

Diverso (1050 Ti)•2/15/23, 1:13 PM

By the way, if someone will be interested, I tried to run colab on my tablet)))) It always disconnects((((

GGP Hi! Yeah, I'm using v1-5-pruned.ckpt

Furkan Gözükara SECourses•2/15/23, 1:13 PM

it says yours corrupted

Furkan Gözükara SECourses•2/15/23, 1:14 PM

i would make a fresh install and download sd 1.5 again

Furkan Gözükara SECourses•2/15/23, 1:14 PM

i have links in description

@Discord Admin I am having this problem, asked you some days ago in youtube comments, can you help m

Reduce the `--save_interval--save_interval` to lower than `--max_train_steps--max_train_steps` to save weights from intermediate steps.

`--save_sample_prompt--save_sample_prompt` can be same as `--instance_prompt--instance_prompt` to generate intermediate samples (saved along with weights in samples directory).

Similar Threads

Reduce the
`--save_interval`
`--save_interval`
`--save_interval`
`--save_interval` to lower than
`--max_train_steps`
`--max_train_steps`
`--max_train_steps`
`--max_train_steps` to save weights from intermediate steps.

`--save_sample_prompt`
`--save_sample_prompt`
`--save_sample_prompt`
`--save_sample_prompt` can be same as
`--instance_prompt`
`--instance_prompt`
`--instance_prompt`
`--instance_prompt` to generate intermediate samples (saved along with weights in samples directory).

@Discord Admin I am having this problem, asked you some days ago in youtube comments, can you help m

Reduce the --save_interval--save_interval to lower than --max_train_steps--max_train_steps to save weights from intermediate steps.

--save_sample_prompt--save_sample_prompt can be same as --instance_prompt--instance_prompt to generate intermediate samples (saved along with weights in samples directory).

Similar Threads

Reduce the --save_interval--save_interval--save_interval--save_interval to lower than --max_train_steps--max_train_steps--max_train_steps--max_train_steps to save weights from intermediate steps.

--save_sample_prompt--save_sample_prompt--save_sample_prompt--save_sample_prompt can be same as --instance_prompt--instance_prompt--instance_prompt--instance_prompt to generate intermediate samples (saved along with weights in samples directory).

Reduce the `--save_interval--save_interval` to lower than `--max_train_steps--max_train_steps` to save weights from intermediate steps.

`--save_sample_prompt--save_sample_prompt` can be same as `--instance_prompt--instance_prompt` to generate intermediate samples (saved along with weights in samples directory).

Reduce the
`--save_interval`
`--save_interval`
`--save_interval`
`--save_interval` to lower than
`--max_train_steps`
`--max_train_steps`
`--max_train_steps`
`--max_train_steps` to save weights from intermediate steps.

`--save_sample_prompt`
`--save_sample_prompt`
`--save_sample_prompt`
`--save_sample_prompt` can be same as
`--instance_prompt`
`--instance_prompt`
`--instance_prompt`
`--instance_prompt` to generate intermediate samples (saved along with weights in samples directory).