VictorMylle/Kohya-ss-sd-scripts
1
0
Fork 1
mirror of https://github.com/kohya-ss/sd-scripts.git synced 2026-04-06 21:52:27 +00:00
Code Issues Packages Projects Releases Wiki Activity

update README

Browse Source
This commit is contained in:
Kohya S
2024-08-12 08:19:45 +09:00
parent d25ae361d0
commit 9e09a69df1
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
README.md
View File

@@ -10,10 +10,10 @@ Aug 10, 2024: LoRA key prefix is changed to `lora_unet` from `lora_flex` to mak
Please update PyTorch to 2.4.0. We have tested with PyTorch 2.4.0 with CUDA 12.4. We also updated `accelerate` to 0.33.0 just to be safe.
We have added a new training script for LoRA training. The script is `flux_train_network.py`. See `--help` for options. Sample command is below. It will work with 24GB VRAM GPUs.
We have added a new training script for LoRA training. The script is `flux_train_network.py`. See `--help` for options. Sample command is below, settings are based on [AI Toolkit by Ostris](https://github.com/ostris/ai-toolkit). It will work with 24GB VRAM GPUs.
```
accelerate launch --mixed_precision bf16 --num_cpu_threads_per_process 1 flux_train_network.py --pretrained_model_name_or_path flux1-dev.sft --clip_l sd3/clip_l.safetensors --t5xxl sd3/t5xxl_fp16.safetensors --ae ae.sft --cache_latents_to_disk --save_model_as safetensors --sdpa --persistent_data_loader_workers --max_data_loader_n_workers 2 --seed 42 --gradient_checkpointing --mixed_precision bf16 --save_precision bf16 --network_module networks.lora_flux --network_dim 4 --optimizer_type adamw8bit --learning_rate 1e-4 --network_train_unet_only --cache_text_encoder_outputs --cache_text_encoder_outputs_to_disk --fp8_base --highvram --max_train_epochs 4 --save_every_n_epochs 1 --dataset_config dataset_1024_bs2.toml --output_dir path/to/output/dir --output_name flux-lora-name --timestep_sampling sigma --model_prediction_type raw --discrete_flow_shift 1.0 --loss_type l2
accelerate launch --mixed_precision bf16 --num_cpu_threads_per_process 1 flux_train_network.py --pretrained_model_name_or_path flux1-dev.sft --clip_l sd3/clip_l.safetensors --t5xxl sd3/t5xxl_fp16.safetensors --ae ae.sft --cache_latents_to_disk --save_model_as safetensors --sdpa --persistent_data_loader_workers --max_data_loader_n_workers 2 --seed 42 --gradient_checkpointing --mixed_precision bf16 --save_precision bf16 --network_module networks.lora_flux --network_dim 4 --optimizer_type adamw8bit --learning_rate 1e-4 --network_train_unet_only --cache_text_encoder_outputs --cache_text_encoder_outputs_to_disk --fp8_base --highvram --max_train_epochs 4 --save_every_n_epochs 1 --dataset_config dataset_1024_bs2.toml --output_dir path/to/output/dir --output_name flux-lora-name --timestep_sampling sigmoid --model_prediction_type raw --guidance_scale 1.0 --loss_type l2
```
LoRAs for Text Encoders are not tested yet.
@@ -29,7 +29,7 @@ We have added some new options (Aug 10, 2024): `--time_sampling`, `--sigmoid_sca
In our experiments, `--timestep_sampling sigma --model_prediction_type raw --discrete_flow_shift 1.0` with `--loss_type l2` seems to work better than the default (SD3) settings. The multiplier of LoRA should be adjusted.
additional note (Aug 11): A quick check shows that the settings in [AI Toolkit by Ostris](https://github.com/ostris/ai-toolkit) seems to be equivalent to `--timestep_sampling sigmoid --model_prediction_type raw --guidance_scale 1.0` (with the default `l2` loss_type).
additional note (Aug 11): A quick check shows that the settings in [AI Toolkit by Ostris](https://github.com/ostris/ai-toolkit) seems to be equivalent to `--timestep_sampling sigmoid --model_prediction_type raw --guidance_scale 1.0` (with the default `l2` loss_type). This seems to be a good starting point. Thanks to Ostris for the great work!
Other settings may work better, so please try different settings.