fix to work controlnet training
11e8c7d8
Merge branch 'original-u-net' into sdxl
0b730d90
add minimal inference code for sdxl
f7f762c6
update dataset to return size, refactor ctrlnet ds
9e9df2b5
add sdxl fine-tuning and LoRA
747af145
fix warning messages are shown every step
56ca5dfa
Add no_half_vae for SDXL training, add nan check
2c461e4a
fix typos
9ebebb22
update for sdxl
31018d57
update readme
b0dfbe70
update readme
753c63e1
use CLIPTextModelWithProjection
a751dc25
update readme
07d5c710
fix to work train_network with fine-tuning dataset
71a6d49d
fix to work
8521ab79
fix max_token_length not works for sdxl
d395bc06
update transfomer to fix sdxl text model with bf16
38e21f5c
fix to work with dreambooth ds without toml
227a62e4
add feature to sample images during sdxl training
64cf9228
remove debug code
97611e89
update readme
5863676c
add min/max_timestep
ea182461
add error message for old npz
2febbfe4
update readme
6aa62b9b
fix dtype for vae
3b35547d
remove debug print
ce46aa0c
remove debug print
3060eb5b
fix to work sdxl state dict without logit_scale
3d0375da
fix to work with .ckpt from comfyui
4a34e580
Update README.md
288efddf
support sdxl in prepare scipt
cc3d40ca
Merge branch 'sdxl' of https://github.com/kohya-ss/sd-scripts into sdxl
73ab1102
update readme
c1d62383
Fix TE key names for SD1/2 LoRA are invalid
66c03be4
support avif
d599394f
fix wrapper tokenizer not work for weighted prompt
fe7ede5a
add generation script
1d25703a
update readme
8371a7a3
Update sdxl_train.py
5f348579
Update train_util.py for full_bf16 support
d9749597
Merge branch 'sdxl' of https://github.com/kohya-ss/sd-scripts into sdxl
7502f662
Merge pull request #626 from ddPn08/sdxl
256ff5b5
Merge pull request #628 from KohakuBlueleaf/full_bf16
3579b457
support multi gpu in caching text encoder outputs
0416f26a
fix pad token is not handled
a380502c
fix conditioning
77ec70d1
fix uncond/cond order
c2ceb6de
update readme
5c80117f
don't hold latent on memory for finetuning dataset
b6e328ea
support textual inversion training
f54b784d
Fix to show template type
68ca0ea9
make tracker init_kwargs configurable
b841dd78
add no_half_vae option
2e67d74d
fix NaN in sampling image
814996b1
fix gradient accumulation doesn't work
3c67e595
remove unnecessary code
8df94856
support diffusers format for SDXL
8fa5fb28
fix typo in sdxl_train_textual_inversion
a7ce2633
Cast weights to correct precision before transferring them to GPU
cdffd19f
fix sampling gen fails in lora training
3bb80ebf
change tokenizer from open clip to transformers
b4a3824c
fix tokenizer 2 is not same as open clip tokenizer
9de357e3
fix sampling in multi GPU training
81fa5483
refactor caching latents (flip in same npz, etc)
94c151ae
update readme
62dd99be
add caching to disk for text encoder outputs
516f64f4
update readme
3db97f88
fix scheduler steps with gradient accumulation
41d19571
add lora for Diffusers
39e62b94
fix typos
1d4672d7
add convenience function to merge LoRA
7e20c6d1
vram leak fix
8a073ee4
enable different prompt for text encoders
3d66a234
make crop top/left same as stabilityai's prep
0ec71660
add zero_terminal_snr option
6d2d8dfd
Merge pull request #645 from Ttl/prepare_order
7875ca8f
enable full bf16 trainint in train_network
225e8718
Merge pull request #641 from kaibioinfo/patch-1
771f33d1
fix invalid args checking in sdxl TI training
fc276a51
fix original w/h prompt opt shows wrong number
86a8cbd0
make to work with PyTorch 1.12
acf16c06
Merge branch 'main' into sdxl
c45d2f21
Merge pull request #630 from ddPn08/sdxl
73a08c0b
fix to work text encoder only network with bf16
8ba02ac8
add invisible watermark to req.txt
d1864e24
re-organize import
50b53e18
support Diffusers format
7ec9a7af
Merge branch 'main' into sdxl
2e4783bc
init unet with empty weights
bb167f94
fix safetensors error: device invalid
eec6aadd
fix LoRA with text encoder can't merge closes #660
c1d5c24b
fix max mul embeds doesn't work. closes #656
7ae0cde7
fix to show batch size for each dataset refs #637
b1e44e96
format by black
e83ee217
support sdxl
2b969e9c
remove unused func
b78c0e2a
fix pipeline dtype
50544b78
add dtype to u-net loading
96a52d98
Merge branch 'sdxl' into sdxl
272dd993
refactor model loading to catch error
315fbc11
fix mismatch dtype
fdb58b0b
fix typo
1199eacb
Merge branch 'main' into sdxl
4072f723
fix typos for _load_state_dict
d9180c03
Merge pull request #676 from Isotr0py/sdxl
e20b6acf
Add Paged/ adam8bit/lion8bit for Sdxl bitsandbytes 0.39.1 cuda118 on …
9ec70252
Merge pull request #658 from pamparamm/cache_latents_leak_fix
54a4aa22
move releasing cache outside of the loop
e6034b7e
change method name, add comments
b62185b8
refactor optimizer selection for bnb
a296654c
format by black
2a4ae88f
fix ControlNet not working
0eacadfa
fix batch generation not working
8856c19c
arbitrary args for diffusers lr scheduler
496c3f27
remove dependency for albumenations
f61996b4
fix recorded seed in highres fix
7e474d21
add adding v-pred like loss for noise pred
0636399c
fix vae crashes in large reso
89aae3e0
format by black
db80c5a2
fix ControlNet with regional LoRA, add shuffle cap
6b1cf6c4
fix ControlNet with regional LoRA
cf683289
Add workaround for clip's bug for pooled output
c6d52fde
remove debug print
9d7619d1
remove debug print
f3be995c
make bitsandbytes optional
9d855091
Merge pull request #714 from kohya-ss/dev
f4935dd6
fix sdxl_gen_img not working
25d8cd47
Merge pull request #716 from kohya-ss/dev
2dfa26cc
fix training textencoder in sdxl not working
e5f9772a
Merge pull request #720 from kohya-ss/dev
cd54af01
support sai model spec
c142dadb
update sdxl ver in lora metadata from v0-9 to v1-0
92f41f10
fix error in huggingface_path_in_repo=None
3cdbbb43
update readme
04b1defa
change assert to print
78226f85
Merge pull request #722 from kohya-ss/dev
f4a2bc6c
add merge script
7ea38f90
fix crashing in saving lora with clipskip
6f80fe17
modify import #368
b83ce0c3
Merge pull request #733 from kohya-ss/dev
4e339bb1
fix to work when input_ids has multiple EOS tokens
daad50e3
Merge pull request #741 from kohya-ss/dev
12e58ab3
fix sample gen failed in sdxl training
e73d103e
Merge pull request #744 from kohya-ss/dev
bf31f18c
add warning for bucket_reso_steps with SDXL
6889ee2b
revert default noise offset to 0 (None) in sdxl
3307ccb2
suppress waning for scheduler args #748
8415014d
support block lr for U-Net
e2c2689f
Merge pull request #750 from kohya-ss/dev
e32e24ad
update readme
d43fcd63
add original size for negative cond
a90c9c27
add lora_fa experimentally
d713e4c7
update readme
adf99a33
Merge pull request #755 from kohya-ss/dev
9a60b8a0
add lora controlnet temporarily
983698dd
add lora controlnet train/gen temporarily
3f7235c3
change to use_reentrant=False
306ee24c
read dim/rank from weights
afc03af3
add skip input blocks to lora control net
6111151f
Merge branch 'dev' of https://github.com/kohya-ss/sd-scripts into dev
784a90c3
add cond/uncond, update config
5fa473d5
fix error in generation
809fca0b
add comments
1e52fe6e
add readme about controlnet-lora
3e159166
modify to attn1/attn2 only
b5db90c8
Merge branch 'dev' of https://github.com/kohya-ss/sd-scripts into dev
62fa4734
rename and update
fef7eb73
fix typos, update readme
5a86bbc0
Update train_lll_README-ja.md add sample images
782b11b8
fix a bug x is updated inplace
06461120
Merge branch 'dev' of https://github.com/kohya-ss/sd-scripts into dev
2841927d
fix bucketing doesn't work in controlnet training
e1918928
update lllite doc
bee5c3f1
update lllite doc
b74dfba2
Update train_lllite_README.md
98f8785a
Merge branch 'dev' of https://github.com/kohya-ss/sd-scripts into dev
0c197550
update lllite doc
15b463d1
Merge pull request #768 from kohya-ss/dev
8b2a11fd
Update train_lllite_README.md
edd314cc
update lllite doc
50c6aaae
fix typos (;^ω^)
9947197a
fix debug_dataset for controlnet dataset
1161a5c6
update doc
9be19ad7
Merge pull request #770 from kohya-ss/dev
014c4b47
fix: VRAM memory leak in sdxl_gen_img.py
fbc550d0
add multiplier, steps range
6d5cffae
Merge branch 'sdxl' into dev
927c6876
add about dataset synthesis for LLLite doc
85f1114c
Merge pull request #776 from kohya-ss/dev
71369ac9
add .toml example for lllite doc
ef28a919
Merge branch 'dev' of https://github.com/kohya-ss/sd-scripts into dev
d97a1638
Merge branch 'sdxl' into dev
1f4a495e
Merge pull request #783 from kohya-ss/dev
b44644bc
fix "\\" to "/" for compatiblity
a46a4781
support jpeg xl
5d88351b
alternative impl of ControlNet-LLLite training
526488fe
fix typo
3d12cdc6
Merge pull request #787 from kohya-ss/dev
f64c78b7
intergrate fp16/bf16 to model loading
6155f9c1
delet missed line
2e0942d5
Merge branch 'main' into sdxl
7e850f3b
Merge branch 'sdxl' of https://github.com/kohya-ss/sd-scripts into sdxl
633bb8d3
add input perturbation noise
e0beb6a9
support diffusers' new VAE
c06a8670
Merge pull request #799 from kohya-ss/dev
f77bdf96
Merge pull request #774 from lansing/lansing/sdxl-fix-gen-memleak
64001167
Merge pull request #786 from Isotr0py/jxl
497051c1
Merge branch 'main' into sdxl
f19a48a2
Merge branch 'sdxl' into dev
cd590030
add caption_prefix/suffix to dataset
948cf174
Merge pull request #800 from kohya-ss/dev
903825af
Merge pull request #791 from Isotr0py/dev
f6d417e2
Merge pull request #798 from vvern999/vvern999-patch-1
2eae9b66
fix typos, add comments etc.
0ee75fd7
fix ControlNetDataset not working
360af277
Merge pull request #802 from kohya-ss/dev
4530b96c
update readme
95bc6e87
update readme
5f08a21d
fix typo
867e7d32
update readme (#803)
74561dbd
fix to work regional LoRA
207fc8b2
Merge pull request #804 from kohya-ss/dev
6b3a580e
Add ip_noise_gamma metadata
80aca1cc
fix VAE becomes last one
0ecfd91a
add support model without position_ids
90c47140
get pool from CLIPVisionModel in img2img
d337bbf8
fix to work highres_fix_latents_upscaling
db7a28ac
Intel ARC support with IPEX
b64389c8
Update getDeviceIdListForCard
b99cd2a9
update bitsandbytes for 0.41.1 and fixed bugs with generate_controlne…
d5be8125
Merge pull request #825 from Disty0/dev
624edf42
Merge branch 'dev' into sdxl
d8464310
rename train_lllite_alt to train_lllite
37578552
Merge pull request #808 from rockerBOO/metadata
d39f1a34
fix sai metadata for sdxl closes #824
477b5260
fix to work iter_same_seed
20e929e2
update versions of accelerate and diffusers
7e736da3
update readme
28272de9
Merge pull request #831 from kohya-ss/dev
98615166
Merge branch 'main' into sdxl
1e395ed2
Support concat LoRA
14aa2923
IPEX attention optimizations
209eafb6
Merge pull request #841 from Disty0/dev
8e117f9f
Merge pull request #839 from laksjdjf/sdxl
365a06bd
Merge pull request #844 from kohya-ss/dev
35a1d68e
update doc
6bd6cd9c
Fix to work training U-Net only LoRA for SD1/2
81419f7f
fix placing of requires_grad_ of U-Net
4cc91960
Merge pull request #846 from kohya-ss/dev
93155242
update readme
c9184892
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub