Comments (5)
There are 3 learning rates to configure.
The first one is simply called "Learning rate" but it can be overridden by the other two: "Text Encoder learning rate (optional)" and "Unet learning rate (optional)"
If you want to disable TE training, then ignore the first Learning rate, set the learning rates for Text Encoder to 0 and Unet to whatever value you need.
You can verify that it's training unet-only by looking for the line network_train_unet_only = true
in the config_lora-xxxxx.toml file.
Another way to verify is through the log output. When you are training unet only, the log should only say enable
for the Unet but not for the text encoder
INFO create LoRA network. base dim (rank): 64, alpha: 32
INFO neuron dropout: p=0.1, rank dropout: p=0.1, module dropout: p=0.1
INFO create LoRA for Text Encoder 1:
INFO create LoRA for Text Encoder 2:
INFO create LoRA for Text Encoder: 264 modules.
INFO create LoRA for U-Net: 722 modules.
INFO enable LoRA for U-Net: 722 modules
prepare optimizer, data loader etc.
from kohya_ss.
Oh gosh, the devil in the details. I didn't understand the gui's top LR entry. When does one use the same values for Unet and te. But makes perfect sense to only use the bottom values
Thanks for clear explanation. I can see now that the te is correctly behaving on the graphs thank you very much for your kind response
from kohya_ss.
There are 3 learning rates to configure.
The first one is simply called "Learning rate" but it can be overridden by the other two: "Text Encoder learning rate (optional)" and "Unet learning rate (optional)"
If you want to disable TE training, then ignore the first Learning rate, set the learning rates for Text Encoder to 0 and Unet to whatever value you need.
You can verify that it's training unet-only by looking for the line
network_train_unet_only = true
in the config_lora-xxxxx.toml file.Another way to verify is through the log output. When you are training unet only, the log should only say
enable
for the Unet but not for the text encoderINFO create LoRA network. base dim (rank): 64, alpha: 32 INFO neuron dropout: p=0.1, rank dropout: p=0.1, module dropout: p=0.1 INFO create LoRA for Text Encoder 1: INFO create LoRA for Text Encoder 2: INFO create LoRA for Text Encoder: 264 modules. INFO create LoRA for U-Net: 722 modules. INFO enable LoRA for U-Net: 722 modules prepare optimizer, data loader etc.
@b-fission btw, do you do any consulting?
from kohya_ss.
No, I'm just dabbling with the tools here occasionally.
from kohya_ss.
@b-fission thanks no problem
from kohya_ss.
Related Issues (20)
- How to train LORA with multiple GPUs HOT 5
- Stable Diffusion 3 HOT 25
- Cannot start DoRa training due to error HOT 6
- controlnet lllite-RuntimeError: Sizes of tensors must match except in dimension 2. Expected size 0 but got size 16 for tensor number 1 in the list.
- How to extract LyCORIS keep small size by good effect.
- voluptuous.error.MultipleInvalid: extra keys not allowed @ data['settings'] why? HOT 1
- Is there a way to train SD3 with LoRA in free Colab? HOT 1
- bug in apply settings Noise offset type HOT 3
- bug Adaptive noise scale from JSON not load or not update GUI HOT 2
- how to uninstall kohya_ss completely. HOT 3
- change command Print for easy formatting and run in terminal HOT 3
- Feature Request: model name = WANDB-run name HOT 4
- dreamboth checkpoint save error HOT 3
- Is there a way to hard code the config file in Lora training? HOT 12
- Dreambooth extration with DoRA HOT 1
- Finetuning SDXL with TE2 set to 0, but TE2 is still training HOT 9
- Is SD3 training available? HOT 14
- Python3.10 required but Debian "Bookworm" 12.5 only has Python3.11 HOT 2
- ImportError: cannot import name 'train_util' from 'library' HOT 4
- Multi Concept, Multi Resolution, Multi Regularzation image training?
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from kohya_ss.