kohya-ss/sd-scripts issues and pull requests

#1641 - Kohya_ss native trainer colab broken yesterday. Captured before/ after install log. Jax, Key array.

Issue - State: open - Opened by SarahPeterson2854 6 days ago

#1640 - New optimizer:AdEMAMix8bit and PagedAdEMAMix8bit

Pull Request - State: open - Opened by sdbds 6 days ago - 5 comments

#1638 - Was "Validation Loss" ever successfully implemented?

Issue - State: open - Opened by mostlyhuman 6 days ago

#1637 - Update train_network_README-zh.md

Pull Request - State: open - Opened by a67793581 7 days ago

#1636 - Specify blocks to Train w/ Finetuning?

Issue - State: open - Opened by setothegreat 7 days ago - 1 comment

#1635 - Drop text encoder weights

Issue - State: closed - Opened by oovm 9 days ago - 1 comment

#1634 - FLUX LoRA Training - Text Encoder active training indicated when it should be disabled with --network_train_unet_only

Issue - State: open - Opened by Enyakk 9 days ago - 2 comments

#1633 - FLUX: Bleeding?

Issue - State: open - Opened by DarkAlchy 9 days ago

#1632 - fix import and make npz for alpha_mask

Pull Request - State: open - Opened by Maru-mee 9 days ago - 8 comments

#1631 - Recommended settings for ScheduleFree.

Issue - State: closed - Opened by waomodder 9 days ago - 4 comments

#1630 - Issue Reminder

Issue - State: open - Opened by lynrayy 9 days ago

#1629 - Extract Flux Lora not working after 93/304

Issue - State: closed - Opened by DarkViewAI 9 days ago

#1628 - Make timesteps work in the standard way when Huber loss is used

Pull Request - State: closed - Opened by recris 10 days ago - 6 comments

#1627 - When I train with multiple GPUs, this error will be reported. This has been bothering me for a long time. How can I solve it

Issue - State: closed - Opened by jinwei1660 10 days ago - 1 comment

#1626 - how to use flux full-finetune safetensors in diffusers?

Issue - State: open - Opened by neuron-party 10 days ago

#1624 - Strange behavior with --network_args parameter

Issue - State: open - Opened by codecomputation 10 days ago

#1623 - neither 'setup.py' nor 'pyproject.toml' found.

Issue - State: closed - Opened by wzgrx 11 days ago

#1622 - Inverted extraction for FLUX.

Issue - State: open - Opened by DarkAlchy 11 days ago - 3 comments

#1621 - lora to original flux lora

Issue - State: closed - Opened by LixDemon 11 days ago - 1 comment

#1620 - fix: backward compatibility for text_encoder_lr

Pull Request - State: closed - Opened by Akegarasu 11 days ago - 1 comment

#1619 - Retain alpha in `pil_resize` for `--alpha_mask`

Pull Request - State: closed - Opened by emcmanus 11 days ago - 4 comments

#1618 - TypeError: get_cosine_schedule_with_warmup() got an unexpected keyword argument 'num_decay_steps'

Issue - State: open - Opened by AndreR 12 days ago - 3 comments

#1617 - AttributeError: 'LycorisNetworkKohya' object has no attribute 'train_t5xxl'

Issue - State: closed - Opened by sd2530615 12 days ago - 2 comments

#1616 - ValueError: invalid literal for int() with base 10: 'still' when using facecrop

Issue - State: closed - Opened by Seedmanc 12 days ago - 1 comment

#1615 - Bug fix: alpha_mask load

Pull Request - State: closed - Opened by Maru-mee 12 days ago - 5 comments

#1614 - Issue for convert_flux_lora.py: RuntimeError: Some tensors share memory...

Issue - State: open - Opened by terrificdm 12 days ago

#1613 - [Feature Request] Decouple `linear1` and `linear2` Flux layers in network_args

Issue - State: open - Opened by EricBCoding 12 days ago - 5 comments

#1612 - Flux merge tool does not properly handle LoRA keys from AI-Toolkit when merging w/ Flux.dev model - All keys "Unused" / Unchanged model file outputted

Issue - State: open - Opened by CCpt5 13 days ago - 9 comments

#1611 - Does the FLUX/SD3 branch not work with SDXL yet?

Issue - State: closed - Opened by DarkAlchy 13 days ago

#1610 - FLUX LoRA Extraction Resemblance Issue

Issue - State: closed - Opened by FurkanGozukara 13 days ago - 1 comment

#1609 - Either "cuDNN SDPA backward got grad_output.strides() != output.strides()" or OutOfMemory

Issue - State: closed - Opened by Tablaski 14 days ago - 5 comments

#1608 - Flux does not support Prodigy?

Issue - State: closed - Opened by DarkAlchy 14 days ago - 2 comments

#1607 - Flux fine-tune text encoder rate is too high; need to be able to set separately

Issue - State: open - Opened by BelieveDiffusion 14 days ago - 2 comments

#1606 - Update submodules as well since MPS changes were needed

Pull Request - State: open - Opened by JoeyOverby 15 days ago

#1605 - sd3 schedule free opt

Pull Request - State: closed - Opened by kohya-ss 15 days ago - 6 comments
Labels: help wanted

#1603 - flux lora training does not start,error: returned non-zero exit status 1

Issue - State: closed - Opened by zxf5843029 16 days ago - 4 comments

#1602 - unexpected keyword argument 'num_decay_steps' when using cosine or linear scheduler (sd3 branch)

Issue - State: closed - Opened by Enyakk 16 days ago - 1 comment

#1601 - poolFD failed error when caching latents for flux finetuning

Issue - State: open - Opened by huxian0402 16 days ago - 4 comments

#1600 - New ScheduleFree support for Flux

Pull Request - State: closed - Opened by sdbds 16 days ago - 2 comments

#1599 - 上手く学習が完了しません

Issue - State: closed - Opened by bell123405 17 days ago - 4 comments

#1598 - flux with flow matching schedule

Issue - State: closed - Opened by Jimzhou82sub 17 days ago

#1597 - Error loading fbgemm.dll

Issue - State: open - Opened by AndreR 17 days ago - 1 comment

#1596 - How to convert a safetensors(finetuned sdxl model) to diffusers dir format?

Issue - State: open - Opened by laolongboy 18 days ago

#1595 - FLUX training for 8GB VRAM?

Issue - State: open - Opened by cocktailpeanut 19 days ago - 12 comments

#1594 - What does split_mode actually do?

Issue - State: closed - Opened by WainWong 19 days ago - 2 comments

#1592 - Critical typo fix

Pull Request - State: closed - Opened by cocktailpeanut 20 days ago

#1591 - error when use deepspeed for FLUX.1 fine-tuning

Issue - State: open - Opened by huxian0402 20 days ago

#1590 - What fp8_base does？

Issue - State: closed - Opened by WainWong 20 days ago - 1 comment

#1589 - flux lora training does not start

Issue - State: open - Opened by nim00e 21 days ago - 3 comments

#1588 - Flux textual_inversion

Issue - State: open - Opened by aiXander 21 days ago - 8 comments

#1587 - flux full fine-tuning output bad result

Issue - State: closed - Opened by huxian0402 21 days ago - 2 comments

#1586 - (sd3-Flux_LoRA) training stuck

Issue - State: closed - Opened by pranauv1 21 days ago - 6 comments

#1585 - Unused keys when merge flux-dev with hyper-sd lora

Issue - State: closed - Opened by huxian0402 21 days ago - 1 comment

#1584 - *Possible important fixes for some issues with flux

Issue - State: open - Opened by DriveHabits 21 days ago

#1583 - [Feature request] --caption_dropout_rate when cache te output

Issue - State: open - Opened by gesen2egee 21 days ago
Labels: enhancement

#1582 - Can not load in ComfyUI and --network_args "split_qkv=True" does not work

Issue - State: open - Opened by JohnsonXi 22 days ago - 2 comments

#1581 - flux lora training issue

Issue - State: open - Opened by wanglaofei 22 days ago - 1 comment

#1580 - Support : OFT merge to base model

Pull Request - State: closed - Opened by Maru-mee 22 days ago - 14 comments

#1579 - why my flux dreambooth version can't inference right image but a noise Map

Issue - State: open - Opened by godwenbin 22 days ago - 1 comment

#1578 - Multi-GPU training sample generation problem - FLUX

Issue - State: open - Opened by FurkanGozukara 22 days ago

#1577 - Random crashes: 'NoneType' object has no attribute 'ndim'

Issue - State: open - Opened by sampbarrow 23 days ago

#1576 - Improve wandb logging

Pull Request - State: closed - Opened by p1atdev 24 days ago - 2 comments

#1575 - Support Lora Block Weight (LBW) to svd_merge_lora.py

Pull Request - State: closed - Opened by terracottahaniwa 24 days ago - 33 comments

#1574 - About sdxl_merge_lora.py using GPU for merging

Issue - State: open - Opened by qqyt110 25 days ago

#1573 - Create init.py in networks folder

Pull Request - State: open - Opened by Leommm-byte 25 days ago

#1572 - Flux Lora producing blurry images

Issue - State: open - Opened by nitinh12 25 days ago - 19 comments

#1571 - Flux.1 LoRA training issue

Issue - State: closed - Opened by paoloski97 25 days ago - 5 comments

#1570 - Possible additon of quantization or Nfloat4 for lower vram (Same quality as fp8/bf16)

Issue - State: open - Opened by DriveHabits 25 days ago - 3 comments
Labels: enhancement

#1569 - Flux Image Logging to wandb is Lost in Multi-GPU Mode

Issue - State: open - Opened by tristanwqy 25 days ago
Labels: bug

#1568 - How do I create this JSON file in flux training? Is there an example file?

Issue - State: closed - Opened by pzzmyc 26 days ago

#1567 - Flux Lora Resize takes 25mins and also does not resize at the end

Issue - State: closed - Opened by DarkViewAI 26 days ago

#1566 - Flux Schnell lora size is too small (1.42 MB) and Does not generate images of the character that it was trained on.

Issue - State: open - Opened by Hassan-jr 26 days ago - 3 comments
Labels: enhancement

#1565 - Merge Flux all in one

Issue - State: open - Opened by oovm 26 days ago - 2 comments

#1564 - Flux shift method update

Pull Request - State: open - Opened by sdbds 26 days ago - 3 comments
Labels: help wanted

#1563 - negative number for iteration per second

Issue - State: closed - Opened by alexcong 26 days ago - 2 comments

#1562 - Flux: RuntimeError: Input type (float) and bias type (c10::BFloat16) should be the same

Issue - State: closed - Opened by sampbarrow 26 days ago - 2 comments

#1561 - RuntimeError: "slow_conv2d_cpu" not implemented for 'Half'

Issue - State: closed - Opened by bulolsi 27 days ago - 2 comments

#1560 - CUDA Out of Memory on Flux Minimal Inference code.

Issue - State: closed - Opened by mazharosama 27 days ago - 2 comments

#1559 - How to Properly Resume Training w/ Respect to Schedulers?

Issue - State: open - Opened by setothegreat 28 days ago - 6 comments

#1558 - how is multi gpu loss gathered?

Issue - State: open - Opened by SnapixAI 29 days ago - 1 comment

#1557 - torch not found error

Issue - State: closed - Opened by PheonixAi420 29 days ago - 7 comments

#1556 - FLUX LoRA training getting terminated without any error - 64 GB RAM - video shared

Issue - State: closed - Opened by FurkanGozukara 29 days ago - 11 comments

#1555 - Compatibility of Flux Pruned fp8 Model with sd-scripts for Fine-tuning

Issue - State: closed - Opened by blackmagic24 29 days ago

#1554 - torch 2.3.1+cuda12.1 with flux lora train, but got the train error when use single GPU, pls help

Issue - State: open - Opened by lilyzlt 29 days ago - 1 comment

#1553 - Using FP8 mixed-precision in accelerate give error. Where and how do I install the necessary packages?

Issue - State: open - Opened by Signorlimone 29 days ago - 4 comments

#1551 - Running Flux Lora training on 2 GPUs

Issue - State: open - Opened by innokean 30 days ago - 11 comments

#1550 - Bump crate-ci/typos from 1.19.0 to 1.24.3

Pull Request - State: closed - Opened by dependabot[bot] about 1 month ago
Labels: dependencies, github_actions

#1547 - Block swapping documentation?

Issue - State: closed - Opened by peteallen about 1 month ago - 4 comments

#1542 - How generate meta_lat.json for flux full fine tuning

Issue - State: open - Opened by BenDes21 about 1 month ago - 1 comment

#1535 - VAE error with FLUX finetuning

Issue - State: closed - Opened by adeysif about 1 month ago - 2 comments

#1531 - TokenizeStrategy strategy is already set

Issue - State: closed - Opened by hablaba about 1 month ago - 1 comment

#1526 - OOM when training flux lora on 8gb vram (4060 mobile)

Issue - State: closed - Opened by oleg996 about 1 month ago - 7 comments

#1522 - Trying to merge Flux Loras

Issue - State: closed - Opened by VadimFedenko about 1 month ago - 5 comments

GitHub / kohya-ss/sd-scripts issues and pull requests