kohya-ss/sd-scripts

Issues

• SD3.5LのLoraで複数概念が学習できない。

  #1802 opened 12 days ago by waomodder
  9

• New format for caching latent and Text Encoder outputs

  #1750 opened a month ago by kohya-ss
  19

• how to use flux loras with text encoder keys with diffusers?

  #1803 opened 9 days ago by neuron-party
  0

• New format for dataset metadata

  #1752 opened a month ago by kohya-ss
  6

• is pytorch 2.4.1 working?

  #1788 opened 12 days ago by RaySteve312
  2

• [FLUX] shuttle-3-diffusion (finetune/lora)

  #1785 opened 17 days ago by kunibald413
  1

• Flux conditional flow implementation

  #1783 opened 18 days ago by aleksandrinvictor
  1

• Issue on training Flux1 LORA, and Fine-Tuning pretrained model.

  #1767 opened a month ago by Hibiki82
  5

• zero3 train flux, accelerate unwrap_model got tensor(size(0))

  #1782 opened 19 days ago by Reginald-L
  1

• "Is there an XL version of extract_lora_from_models.py?"

  #1781 opened 12 days ago by qqyt110
  1

• Flux training methods

  #1801 opened 12 days ago by wanglaofei
  0

• example script for sdxl_train_control_net.py

  #1800 opened 12 days ago by blacklig
  0

• Prodigy + ScheduleFree

  #1796 opened 13 days ago by hablaba
  4

• Feature: model merger support

  #1798 opened 13 days ago by kalle07
  0

• what is the default value of d_coef if not specificed? (for prodigy optimizer)

  #1797 opened 13 days ago by Hugs288
  2

• Multi-gpus(RTX3090-24GB) Flux finetuning

  #1791 opened 15 days ago by wanglaofei
  2

• Koyha Issue

  #1786 opened 17 days ago by Ivanoah
  1

• How does cosine lr scheduler work when the prodigy optimizer is used ?

  #1787 opened 16 days ago by sipie800
  1

• Crashed randomly in the middle of cooking

  #1795 opened 14 days ago by AbstractEyes
  0

• Can't load tokenizer for 'openai/clip-vit-large-patch14'

  #1792 opened 14 days ago by hotdogarea
  2

• Runtimeerror when dataset is too large

  #1790 opened 15 days ago by JEFFSEVENTHSENSE
  2

• Plans to support FLUX inpainting model training?

  #1794 opened 14 days ago by huxian0402
  0

• Request: Add Support for V-Prediction SDXL Model in LoRA Training

  #1772 opened 23 days ago by Wenaka2004
  4

• Intermittent problems when generating sample images. AttributeError: 'NoneType' object has no attribute 'to'

  #1759 opened 15 days ago by ReppiksProductions
  5

• AssertionError: g_tokens must be None if l_tokens is None

  #1793 opened 15 days ago by betterftr
  0

• faster-block-swap works amazing but it has problems for 12 GB and below GPUs

  #1764 opened 16 days ago by FurkanGozukara
  5

• Has SD3 implemented different shifts for different resolutions?

  #1762 opened a month ago by gesen2egee
  1

• error when trying to use --blockwise_fused_optimizers with SD3.5 L finetune training.

  #1755 opened 16 days ago by Ice-YY
  2

• UserWarning: cuDNN SDPA backward got grad_output.

  #1757 opened a month ago by kitamo1981
  2

• give different prompts to different text models (for 3.x and flux)

  #1780 opened 20 days ago by yoinked-h
  0

• When using masked loss, the console output is confusing as hell.

  #1773 opened 23 days ago by iqddd
  1

• --gpu_ids 0 lose efficacy after pull

  #1778 opened 21 days ago by jinwei1660
  0

• High VRAM usage with Blocks to swap on ROCM

  #1776 opened 22 days ago by csf3o
  2

• Python 3.12 support

  #1777 opened 21 days ago by ChristianMayer
  2

• FSDP support

  #1775 opened 22 days ago by ljleb
  1

• any plans to support training for OpenFlux?

  #1766 opened a month ago by huxian0402
  1

• How to generate Flux images in a CLI for testing LORAs? Can I use gen_img_diffusers.py for this?

  #1761 opened a month ago by ReppiksProductions
  9

• BUG: Cache validation does not check latent size for SD/SDXL (sd3 branch)

  #1771 opened 25 days ago by feffy380
  0

• SD3.5 finetuning crashes due to dtype difference in vae call when not caching latents (potentially DDP related?)

  #1758 opened a month ago by yoinked-h
  3

• Which branch to use for Flux and SD3.5 training

  #1754 opened a month ago by ddpasa
  3

• [feature request]: Recursive loading of image folder

  #1760 opened a month ago by TeKett
  3

• Merge of Flux lycoris.kohya LoRAs fails with KeyError: 'lora_te1_text_model_encoder_layers_0_mlp_fc1.hada_w1_'

  #1753 opened a month ago by envy-ai
  1

• OOM (RAM) when saving checkpoint during full finetuning

  #1749 opened a month ago by aiXander
  1

• Possible to Separate Learning Rates for Text-Encoder in SDXL LoRA Model Training?

  #1748 opened a month ago by Gtonero
  1

• Preset has error, SDXL - LoRA AI_Now prodigy 1.0 bucket reso steps

  #1763 opened a month ago by revolvedai
  0

• Merge Flux and SD3.5 Lycoris models to the full models support?

  #1756 opened a month ago by DarkViewAI
  0

• using DDP with sd3.5 training crashes when logging device

  #1751 opened a month ago by yoinked-h
  1

• Support LoRA train for Flux Schnnell or openFlux?

  #1745 opened a month ago by 4d4m0r
  1

• Using TI Embeddings during training

  #1746 opened a month ago by DaStapo
  0

• --always_first_tags not working (possibly)

  #1744 opened a month ago by dpons222
  0