Please subscribe and like the video to help me keep motivated to make awesome videos like this one. :) Inference code: gist.github.com/abhishekkrthakur/ebe82f7140d3b5bff925320d59972b14 Training part is in video description!

@Pakmandesign

Жыл бұрын

What version of autotrain are you using? My doesn't let me run autotrain setup. and threw this error usage: autotrain [] AutoTrain advanced CLI: error: argument {app}: invalid choice: 'setup' (choose from 'app')

@abhishekkrthakur

Жыл бұрын

@@Pakmandesign always on latest release: pypi.org/project/autotrain-advanced/

@gunasekhar8440

Жыл бұрын

Should we change the prompt? Or else we need to use only those words except images?

@abhishekkrthakur

Жыл бұрын

@@gunasekhar8440 prompt can be anything. just remember it as you have to use the same keyword during inference. in my case "sks dog"

@gunasekhar8440

Жыл бұрын

@@abhishekkrthakur really very thanks for the reply

Very helpful tutorials on the latest updates Abhishek, thanks!!

I don't see the option to select after creating a new space with docker and autotrain.. It's running on A10G but the backend options are totally different and there's no option to select projecttype. The tabs available to me are llm, text, and tabular

A beginner's question maybe, "How can I train the model on different things" like this dog example you shared. How can I build on that and add more concepts for the same model.

doest work. mem error in last step. tried on colab as well gcp office machin

Does this use .txt file or filename captions? Or is it just a single 'sks' keyword type prompt for every image?

Great Tutorial!

Thank you for a great video. I got some problem. When training is end and it's time to save lora weights and checkpoint, i get cuda out of memory. Do you have any idea how to fix it? solution: use --checkpointing-steps 100000 to save only lora weights

Thanks for your tutorial. Do you know if it is possible to use Kohya for SDXL dreambooth training ?

Can you help me? I'm trying to use the huggingface "AutoTrain_Dreambooth" notebook on Kaggle, with two T4 Gpus, to train faces on sdxl Worked, but it's only using one gpu I tried to launch using the accelerator changing !autotrain dreambooth \ to !accelerate launch autotrain dreambooth \ But this error message appears "can't open file 'kaggle/working/autotrain': [Errno 2] No such file or directory" How can i fix that? Or is there any other way to force the notebook to use two Nvidia t4?? Sorry to bother you

What might i be doing wrong if it is giving me an error after like 25 minutes of working?

Thanks for sharing :)

how to add different objects to same model like cat , dog , car etc

i see that we have a --resume-from-checkpoint parameter. can we have continue training one model that we start in other colab session? i trained one model until 1300 steps but i think we need more train. if i can resume this checkpoint i will not spen a lot of hours in processing data. i tryed add this parameter in colab and local but allways get error.

Is a 4070Ti 12gb enough? I've seen that the recommended setup is 16gb vram. I don't care if it takes longer, what I don't want is to run into too many OOM errors

how can i use realistic vision v5.1 instead of Stable Diffusion XL?

Amazing work, I am going to try this on a local machine I am buying new GPU for this and I will need to used it for local LLMa what GPU you think will work the best and I am going to try this on FPGA next just see how this can perform.

Great tutorial! So if I'm following the idea here, you are training to make an object (the dog in the pictures) to be prompted into different styles basically. But what about if you want to train it for a style that you can turn any object into that style? So basically how does the training differ if you want to train it to output a specific style across different objects?

@harnageaa

Жыл бұрын

think about this, the model learns the way the pixels get arranged in an imagine. If ur image have dogs, then he learns the pixels are shaped in a dog shape, so u get dogs.., if u get images with an 'artist' name in mind, then all the pixels will be shaped in that specific style, then once you train on "anime girls" for example from a hentai artist. You will still get "objects", drawn in his style even if you had zero objects with that stype in the training dataset. Does that make sense? if u think of shape, just find shape of what u want, if u want style just add only that type of style in dataset. same for pose (pose = shape for AI)

@harnageaa

Жыл бұрын

@@BarbSmith-fw7vbthen you will get all 10 + combined. Basically you might get 'random' styles, Some people train with tags each image(filewords)? at least in SD 1.5, and that meant when you use certain tags you retrieve that data. Simpler would be a Lora, where u have 10 images, and 10 txt for each image, where u have 'tags', assuming each file has different "artstyle tag" in the txt file then when you use that tag with the artstyle name you will get it. Also how u think these 'generalistic' models are made of that can create literally any type of stylistic image. They just add every genre and train 1 model. So I guess if u want to retrieve a certain style, u would need a 'tag'. In dreambooth SD1.5 I used to use 'concepts', where you had 4 keywords for dreambooth (I didn't used filewords), and I would teach the AI 4 keyword and everytime you use one of them, he gives exactly what I want (1 a pose, 1 a style, a manga style and a generalistic term that generate default images when prompted, so you use 1 word to generate what the model is suppose to). You will have to experiment to see what exactly happen when u use 10 artstyles in 1 . My guess, all in one + combined at the same time

Thanks for getting this up so quick!! A couple notes: 1. it's !autotrain setup --update-torch (not upgrade) 2. getting the error 'RuntimeError: "LayerNormKernelImpl" not implemented for 'Half'' when trying to run dreambooth training specifically after '> INFO Computing text embeddings for prompt: subject wearing sks' is printed Any ideas?

@abhishekkrthakur

Жыл бұрын

uff! sorry! thanks for the correction. the error is because either sd pipe or refiner is not moved to cuda

@numera123

10 ай бұрын

was this resolved i still get this error

I can't make it work on Kaggle ! --output parameter doesn't exist --project-name parameter is required I have solved the 2 parameters problems but I have this error : RuntimeError: "LayerNormKernelImpl" not implemented for 'Half'

how we can train a DreamBooth model with multiple "characters"?Please guide

does this run on Apple silicon MacBooks ?

I get below message on cmd when i run the following command, "C:\Windows\system32>autotrain setup --update-torch usage: autotrain [] AutoTrain advanced CLI: error: argument {app}: invalid choice: 'setup' (choose from 'app')" How do i fix this?

@abhishekkrthakur

Жыл бұрын

that means you have an old version of autotrain installed

thanks for this nice tutorial and the autotrain-advanced tool. I am able to reproduce everything you do. One slight question tho; I'm new to ML & SD so I don't understand everything in detail. What we train here; "pytorch_lora_weights.safetensors" is just a LORA weight right? not a full checkpoint. (since the size is also very small) - I've copied this file into "loras" directory in comfy-ui, set the base model as sdxl and load this one as lora, however it doesn't seem to do anything when I prompt "sks man", I just get a random image and hundreds of "lora key not loaded unet.up_blocks.1.attentions.2.transformer_blocks.1..........." errors - is this not supposed to work as standalone lora?

@archonoff

Жыл бұрын

I've got exactly the same behaviour. A bunch on 'lora key not loaded ...' messages and no effenct on the final result. The results were just random people. I also tried to use this lora in a1111, but the same result: just random people. Have you finally find the solution? What was wrong?

@andresmartin8895

Жыл бұрын

I have the same problem

@cemal6950

11 ай бұрын

have u type the in prompt? also use the "sks man"

@trsd8640

11 ай бұрын

Yes, thats right. The Lora did not work anywhere. I try it in automatic1111, comfyUi, Draw Things,.. it does definitiv NOT work.

In the last step, it is showing me an error in the first line, "diffuser" not installed. What to do? I am new to this.

@PingPongReview

10 ай бұрын

Got it, I have to run the first part again

how to load the pytorch_lora_weights.safetensors file to webui?

just tried the hugging face spaces method and keep getting errors. Some around creation of /tmp/training. Not sure what the exact issue is since everything is automated but followed the same setup from the video with four 512x512 training .jpeg images

@abhishekkrthakur

Жыл бұрын

weird. whats the error in the logs?

@pranaviyer6291

Жыл бұрын

@@abhishekkrthakur The main one was no such directory called /tmp/training --> I assume it was attempting to move the created training dataset there and perhaps didn't actually create it?

It is complusory to download dreambooth ,as my laptop is old ??

does the aspect ratio matter? Does the aspect ratio need to be consistent across the training set?

@chrisrosch4731

7 ай бұрын

would like to know too

@chinmoybasak7749

4 ай бұрын

Did you find out?

Even when I use the prompt that I trained the model with I still can't get it to generate what I trained it to generate. Does anyone have an idea why this could be the case?

@JamalanJuda

Жыл бұрын

make sure the prompt you give it when you train is somewhat unique, so in the video he says sks dog, make your prompt something like "photo of hbvkdnvdkvfnj" and then when you prompt it back use "hbvkdnvdkvfnj" as the subject name.

@trsd8640

11 ай бұрын

It simply doesnt works. I tried the generated lora on nearly every ui, but it dont works.

@abhishekkrthakur

11 ай бұрын

you can use the code for inference. ill see if there is a conversion script

now, AttributeError: 'DownBlock2D' object has no attribute 'attentions' what to do?

@PingPongReview

10 ай бұрын

this is the code i was using from diffusers import DiffusionPipeline, StableDiffusionXLImg2ImgPipeline import torch prj_path = "/content/my_first_project" model = "stabilityai/stable-diffusion-xl-base-1.0" pipe = DiffusionPipeline.from_pretrained( model, torch_dtype=torch.float16, ) pipe.to("cuda") pipe.load_lora_weights(prj_path, weight_name="pytorch_lora_weights.safetensors") refiner = StableDiffusionXLImg2ImgPipeline.from_pretrained( "stabilityai/stable-diffusion-xl-refiner-1.0", torch_dtype=torch.float16, ) refiner.to("cuda") prompt = "Linkedin photo of me" seed = 42 generator = torch.Generator("cuda").manual_seed(seed) image = pipe(prompt=prompt, generator=generator).images[0] image = refiner(prompt=prompt, generator=generator, image=image).images[0] image.save(f"generated_image.png")

Hello. Is it possible to use the pytorch_lora_weights.safetensors file inside Webui ?

@abhishekkrthakur

Жыл бұрын

i guess so. but since i dont use any webui, i dont know. afaik, most webui support LoRA weights

any tips on using this method (via hugginface spaces) to train for faces? would it need significantly more reference images and steps to produce a reliable output? also do these .safetensor files act the same as any other lora when used in a SDXL workflow utilizing comfyui for example?

@abhishekkrthakur

Жыл бұрын

Using face images should be fine. Unless i dont understand the question 🤔. yes the safetensors should be compatible anywhere

i failed to run on free colab with 8bit and resolution 512 options because of lack of GPU i guess.....

@abhishekkrthakur

Жыл бұрын

you need to attach a gpu to colab. free works with 512 resolution

Hi everyone! I have a noobie question. AutoTrain advanced CLI: error: unrecognized arguments: --output model/ How to fix this error and specify the output model?

Could I have to know how many Graphic Card Memory and System Memory for SDXL with StableDiffusion WebUI?

@abhishekkrthakur

Жыл бұрын

16gb for this tutorial

Receiving the below error while running in colab. Have used the same autotrain command given. Please assist on the error. File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/normalization.py", line 190, in forward return F.layer_norm( File "/usr/local/lib/python3.10/dist-packages/torch/nn/functional.py", line 2515, in layer_norm return torch.layer_norm(input, normalized_shape, weight, bias, eps, torch.backends.cudnn.enabled) RuntimeError: "LayerNormKernelImpl" not implemented for 'Half'

@benjaminm.1053

Жыл бұрын

Are you sure you attached a GPU to your google colab Notebook?

@abhishekkrthakur

Жыл бұрын

if you did attach a gpu, you need to run the setup command. please see description

How do I run inference on huggingface spaces?

@nidesharumugam5488

6 ай бұрын

Hi, did you find a way to deploy it ?

i trained on huggingfaces it went trough, but i only got a file with 23.4mb i thought it would be much bigger

@abhishekkrthakur

Жыл бұрын

its LoRA

@andreasfischer8916

Жыл бұрын

@@abhishekkrthakur ok thank you

fyi: the model training took around 20mins with 16GB gpu memory.

holy shit. can you do another one in a1111 or something?

Thanks a lot for this tutorial! I kept running into the import error that says I can't import 'randn_tensor' from 'diffusers.utils' in colab. Does anyone know how I could fix this?

@abhishekkrthakur

10 ай бұрын

fixed now

@davidyang1322

10 ай бұрын

Awesome!! Thank you for the quick fix!@@abhishekkrthakur

@davidyang1322

10 ай бұрын

Hi again! I was still getting the CUDA out of memory error when using free colab, even after adding the checkpointing_steps and changing res + using 8bit adam adapter, could you please adjust the colab link so that it avoids these errors? maybe it could be that I added the code to the wrong line? Thx in advance@@abhishekkrthakur

@abhishekkrthakur

10 ай бұрын

@@davidyang1322 colab link as it is works perfectly fine.

@abhishekkrthakur

10 ай бұрын

colab.research.google.com/github/huggingface/autotrain-advanced/blob/main/colabs/AutoTrain_Dreambooth.ipynb use this

Same error always! usage: autotrain [] AutoTrain advanced CLI: error: unrecognized arguments: --output output/

@abhishekkrthakur

8 ай бұрын

args have changed after video, use --help to see new args

Image folder name - Kutta 😂😂.. Nice😁

@abhishekkrthakur

10 ай бұрын

i made it too visible and you are the first person who observed this 🤣

@shinchan9075

10 ай бұрын

@@abhishekkrthakur yes😁

Does this work with a 4060TI 16GB VRAM?

@abhishekkrthakur

Жыл бұрын

it should barely fit in 16GB ram. dont forget to enable 8bit adam

@DrHanes

Жыл бұрын

@@abhishekkrthakur hmm barely dosent sound good, and with two 4060TI = 32GB VRAM? Does it support multi GPU?

Looked very easy but no way to use captions?

Hey I would like to train , hindu gods and goddess , could you please suggest me , how can i do that , I am newbie in AI Thank you

@pipmuzik

11 ай бұрын

@abhishekkrthakur

hf with the beefiest gpu FileNotFoundError: [Errno 2] No such file or directory: '/tmp/training'

@abhishekkrthakur

Жыл бұрын

did it happen in first run? can you factory reset the gpu and try again?

@grebenshyo

Жыл бұрын

@@abhishekkrthakur yes, in the first run. thanks for the tip, i'm thinking i might have missed to dial down the batch size. rerunning now. will try if it doesn't work 🙌🏾

@grebenshyo

Жыл бұрын

nope. batch size wasn't the issue... need to figure out how that's done and will try

@abhishekkrthakur

Жыл бұрын

@@grebenshyo in settings

@grebenshyo

Жыл бұрын

@@abhishekkrthakur thank you 👍🏾 i actually lied: was on the entry level gpu tier. switched to the beefier and it seems to work now

Thanks very much for this tutorial! I tried running this code on colab and got this error: Input type (c10::Half) and bias type (float) should be the same

@user-wn3ro1tc6z

Жыл бұрын

Actually, adding this fixed it: !pip install super-gradients==3.1.3

@abhishekkrthakur

Жыл бұрын

did you run autotrain setup command?

@user-wn3ro1tc6z

Жыл бұрын

it turned out that fixing this: !autotrain setup --update-torch (not upgrade) resolved the issue, also. Thanks again! @@abhishekkrthakur

This code doesn't work anymore.

@abhishekkrthakur

9 ай бұрын

use this: colab.research.google.com/github/huggingface/autotrain-advanced/blob/main/colabs/AutoTrain_Dreambooth.ipynb

@Deathless-ye7xb

9 ай бұрын

Thanks, it worked very good!@@abhishekkrthakur, do you know if we can use another model as a base (I mean another sdxl 1.0 model from civitai or huggingface as a base for training our LoRA)? I'll appreciate the response.

@abhishekkrthakur

9 ай бұрын

glad to know. yeah you can use any other compatible model from huggingface. not sure about civitai

@Deathless-ye7xb

9 ай бұрын

It will be really good if you can show us how to train a sdxl 1.0 model (checkpoint/safetensors). I mean sdxl 1.0 fine-tuning, because no one made a video on youtube about it.@@abhishekkrthakur

Define : beginner friendly 🤣

But wait, that's a simple Lora. Not a dreambooth!

@abhishekkrthakur

Жыл бұрын

its not full dreambooth, yes

Wow you just confused me. Are you using Linux? I never had to write code once when using SD so far. I use auto1111 and comfyui and i have trained using a colab as well as my own auto1111 never even seen code.

hi i found your book for free, it is online pdf. i was going through popular github repo when i found ur repo and just for fun tried searching for your book for free. if you want the details please let me know. it is very interesting book or rather research material, i hope you dont blame me for reading it.

@abhishekkrthakur

Жыл бұрын

book is free. link in description 🙂 hope you liked it

@saihanbatbayarsaihan1138

Жыл бұрын

@@abhishekkrthakur oh i was reading your material and it said to notify you if i found it for free. i liked it a lot.

I have tried on Google Colab but I get this, even with --use-8bit-adam and 512 using a Nvidia v100, any clue? torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 26.00 MiB (GPU 0; 15.77 GiB total capacity; 13.58 GiB already allocated; 8.12 MiB free; 14.26 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 Steps: 100% 500/500 [20:20

@suhaaskatikaneni1925

Жыл бұрын

yeah I get same error. did you find fix for this?

@ivangu1446

Жыл бұрын

any solution?

@abhishekkrthakur

Жыл бұрын

set --checkpointing_steps=2000000 saving checkpoints is causing the issue in colab

@benjaminm.1053

Жыл бұрын

@@abhishekkrthakur I also get that error but not during the training, instead afterwards during inference. Any suggestion how to reduce memory consumption? Can I adjust something in your python script? Thx

@riccostapt

11 ай бұрын

@@benjaminm.1053 facing the same issue. Have you found a way to solve it?