kohya-ss/sd-scripts

Issues

• Cannot apply trained SD 3.5 LoRA in Diffusers

  #1743 opened a month ago
  4

• not find sd3.5 training script and instruction in README

  #1742 opened a month ago
  1

• Problem Training LoRa????

  #1740 opened a month ago
  0

• A massive NPZ file when using alpha mask

  #1739 opened a month ago
  1

• save_last_n_steps is ignored SD 3 FLUX branch

  #1738 opened a month ago
  2

• (sd3-Flux) returned non-zero exit status 3221225477. 13:02:14-584959 INFO Training has ended

  #1737 opened a month ago
  5

• Unable to Detect Images in Kohya_SS Fine-Tuning on RunPod - Need Assistance

  #1736 opened a month ago
  5

• Odd technical problem.

  #1735 opened a month ago
  1

• [Feature Request]Use Gzip for compress latent file size

  #1734 opened a month ago
  3

• flux dev finetune, training loss drops from 0.5 and stays at around 0.34 for thousands of steps, no further changes

  #1733 opened a month ago
  3

• sd3.5 training supports toml configuration and accelerate multi-card training

  #1732 opened a month ago
  0

• Improvment to lora.py and flux lora.py

  #1730 opened a month ago
  2

• 'Max Token Length' setting for flux

  #1729 opened a month ago
  1

• Feature Request: Save N epoch and N steps, comma seperated

  #1727 opened a month ago
  3

• 35D Train Lora Error when sample。

  #1726 opened a month ago
  2

• Error when train sd35 lora

  #1725 opened a month ago
  2

• SD3.5のLora学習エラー

  #1724 opened a month ago
  7

• Flux how to use the lora model with T5xxl trained

  #1723 opened a month ago
  3

• When uploading SDXL LoRA merged by sd-scripts on CivitAI, they reports "hash corrected"

  #1722 opened a month ago
  7

• Training LoRA for Flux Multi GPU not working with more than 2 GPUs

  #1721 opened a month ago
  4

• Enabling dim_from_weights or loraplus_unet_lr_ratio will cause the error: "train_blocks must be single for split mode" (content updated).

  #1720 opened a month ago
  3

• What is the default scheduler used by flux lora inference? I don’t seem to see it in the optional parameters.

  #1718 opened a month ago
  1

• Meaningless Loss in tensorboard.

  #1712 opened a month ago
  1

• "cat_cuda" not implemented for 'Float8_e4m3fn'

  #1711 opened a month ago
  5

• Training with latent cache

  #1709 opened 2 months ago
  1

• --resume decreases training speed by 4-10x

  #1708 opened 2 months ago
  4

• "mixed_precision fp16" not working for Flux

  #1707 opened 2 months ago
  3

• Specify the resolution of sample image

  #1706 opened 2 months ago
  1

• ERROR for FLUX lora training with finetune

  #1705 opened 2 months ago
  0

• Is there a way to set the maximum number of save-state saves created?

  #1704 opened 2 months ago
  4

• Flux De-distilled Training Samples Garbled

  #1703 opened 2 months ago
  1

• Flux Dedistilled / fluxdev2pro support ?

  #1702 opened 2 months ago
  13

• ValueError：Cannot load xxxx from flux/vae because the following keys are missing

  #1701 opened 2 months ago
  4

• OOM using AdamW8bit since recent update

  #1700 opened 2 months ago
  6

• The results of flux_minimal_inference are much worse than webui forge

  #1699 opened 2 months ago
  1

• [Feature] svd_merge_lora ver.flux

  #1698 opened 2 months ago
  0

• Flux AdamWScheduleFree on 24GB

  #1697 opened 2 months ago
  4

• AttributeError: 'bool' object has no attribute 'num_processes'

  #1696 opened 2 months ago
  6

• images sampled during LORA FLUX schnell training and after training are noisy

  #1695 opened 2 months ago
  5

• Support REPA

  #1694 opened 2 months ago
  2

• Odd crashes due to TCP pools in sd-scripts with multiprocessing.

  #1693 opened 2 months ago
  3

• Cannot resume finetuning flux checkpoint - resume option cannot find input path, but input path is correct ###kohya_ss GUI release v24.2.0###

  #1692 opened 2 months ago
  12

• fatal: repository 'https://github.com/kohya-ss/sd-script/' not found

  #1691 opened 2 months ago
  1

• OSError: Can't load tokenizer for 'google/t5-v1_1-xxl'

  #1689 opened 2 months ago
  2

• Best Code for Full SDXL finetuning?

  #1688 opened 2 months ago
  2

• Blurry images by LORA schnell

  #1687 opened 2 months ago
  1

• LoRa Flux never converge and when used in inference has basically no effect

  #1685 opened 2 months ago
  1

• T5XXL tokenizer deadlocks when num workers>0

  #1684 opened 2 months ago
  0

• error while training lycoris for flux.1

  #1683 opened 2 months ago
  3

• GGUF finetune?

  #1682 opened 2 months ago
  0