80 Commits

Author	SHA1	Message	Date
layerdiffusion	e722991752	control rework	2024-08-02 22:17:27 -07:00
layerdiffusion	6ed95d66e5	rework several component patcher ... backend is 65% finished	2024-08-02 14:56:57 -07:00
layerdiffusion	91543adb90	resolve #926	2024-08-02 13:55:11 -07:00
layerdiffusion	d1b8a2676d	rework lora and patching system ... and dora etc - backend rework is 60% finished And I also removed the webui’s extremely annoying lora filter from model versions.	2024-08-02 13:45:26 -07:00
layerdiffusion	e5860a4999	move stream file	2024-08-02 03:37:22 -07:00
layerdiffusion	318219bc9d	move file	2024-08-02 03:37:20 -07:00
layerdiffusion	018903ce2a	fix kohya-hr	2024-08-02 00:37:14 -07:00
layerdiffusion	61ca3bc34f	fix lora	2024-08-01 23:29:27 -07:00
layerdiffusion	f6981339b0	predictor is a better name	2024-08-01 21:29:34 -07:00
layerdiffusion	bc9977a305	UNet from Scratch ... Now backend rewrite is about 50% finished. Estimated finish is in 72 hours. After that, many newer features will land.	2024-08-01 21:19:41 -07:00
layerdiffusion	b3bd78289e	resolve #915	2024-08-01 14:17:04 -07:00
continue revolution	5192e912ab	GroupNorm patcher (#593) ... * add gn wrapper and corresponding patcher * add gn wrapper and corresponding patcher	2024-08-01 12:53:27 -07:00
layerdiffusion	4d1be42975	Intergrate CLIP	2024-08-01 12:27:20 -07:00
layerdiffusion	0d079a846d	Intergrate Native AutoEncoderKL	2024-07-31 21:10:19 -07:00
layerdiffusion	dc62b0d2d7	begin to use new vae impl	2024-07-30 08:43:14 -06:00
layerdiffusion	abd4d4d83d	unify cast name	2024-07-30 08:42:51 -06:00
layerdiffusion	e26abf87ec	Gradio 4 + WebUI 1.10	2024-07-26 12:02:46 -07:00
lllyasviel	29be1da7cf	pass options to cross attention class	2024-03-08 00:50:29 -08:00
lllyasviel	e48533bdcd	change patcher method	2024-03-07 00:26:17 -08:00
lllyasviel	95bcea72b1	Revert "fix alphas cumprod (#475)" ... This reverts commit 72139b000c.	2024-03-03 22:40:56 -08:00
Chengsong Zhang	72139b000c	fix alphas cumprod (#475)	2024-03-03 20:09:04 -06:00
lllyasviel	ef35383b4a	VAE patcher and more types of unet patches	2024-02-29 22:37:34 -08:00
lllyasviel	056d4d0f91	fix for neveroom extension	2024-02-25 16:47:02 -08:00
lllyasviel	a0c89fae12	fix ci	2024-02-25 11:24:56 -08:00
lllyasviel	5e5b60b5b1	rework lora loading ... and add logs	2024-02-25 11:04:14 -08:00
lllyasviel	437c348926	Add build-in extension "NeverOOM" ... see also discussions	2024-02-24 19:09:06 -08:00
lllyasviel	434ca2169f	Add optimization --cuda-stream ... See also the readme for more details	2024-02-24 14:00:48 -08:00
lllyasviel	0f09d98814	revise memory formulation for special amd like #343	2024-02-24 09:36:52 -08:00
lllyasviel	79bdb78619	less verbose	2024-02-23 22:31:16 -08:00
lllyasviel	d508d8132f	add cmd flag hints	2024-02-23 20:06:08 -08:00
lllyasviel	88f395091b	add two optimizations ... --pin-shared-memory and --cuda-malloc See also the updates in Readme for more details	2024-02-23 18:39:32 -08:00
lllyasviel	54c89503eb	Disable pin page ... This is an emergency fix GTX 1060/1050/1066 either does not have shared GPU page vram or have less than 2GB shared page vram - pinning any tensors larger than that will crash Solution is still under investigation.	2024-02-23 16:43:08 -08:00
lllyasviel	bde779a526	apply_token_merging	2024-02-23 15:43:27 -08:00
lllyasviel	2a7fb1be24	less verbose	2024-02-23 14:22:37 -08:00
lllyasviel	ad0ce480f9	always print time	2024-02-23 13:02:30 -08:00
lllyasviel	df12dde12e	Rework unload system ... Previous repeated loading (on cn or other extensions) is fixed. ControlNet saves about 0.7 to 1.1 seconds on my two device when batch count > 1. 8GB VRAM can use SDXL at resolution 6144x6144 now, out of the box, without tiled diffusion or other things. (the max resolution on Automatic1111 txt2img UI is 2048 but one can highres fix to try 6144 or even 8192)	2024-02-23 12:58:09 -08:00
lllyasviel	19473b1a26	fix ci	2024-02-23 09:44:08 -08:00
lllyasviel	26c325296e	rework memory computation for async loader (#377)	2024-02-23 09:24:39 -08:00
lllyasviel	eacb14e115	tune threshold based on more test devices ... for async mover	2024-02-22 17:18:31 -08:00
lllyasviel	8283774b86	revise caster	2024-02-22 10:24:27 -08:00
lllyasviel	6ebef20db3	avoid potential OOM caused by computation being slower than mover ... avoid OOM (or shared vram invoking) caused by computation being slower than mover (GPU filled with loaded but uncomputed tensors), by setting the max async overhead to 512MB	2024-02-22 08:24:23 -08:00
lllyasviel	167dbc6411	safe value for new memory peak	2024-02-22 06:31:21 -08:00
lllyasviel	539bc5035d	safe cleanup to avoid potential problems	2024-02-22 01:28:38 -08:00
lllyasviel	4080e25805	add todo note	2024-02-22 00:47:26 -08:00
lllyasviel	846fdc3341	also implement async offload to control-lora ... controlnet, t2iadapters, etc	2024-02-22 00:20:35 -08:00
lllyasviel	638ee43bf1	Merge upstream PR 14855	2024-02-21 23:59:40 -08:00
lllyasviel	3cdae09639	try solve saturation problems for instant id in #155	2024-02-11 20:53:17 -08:00
lllyasviel	237f80681a	add more notes to clip codes	2024-02-11 20:28:02 -08:00
lllyasviel	30dd8af08c	fix error in head info	2024-02-11 20:12:27 -08:00
lllyasviel	3d039591fe	Expand head info in files ... Previously before this commit, credits are already in entry and licenses are already in root. This commit will make info clearer.	2024-02-11 19:55:18 -08:00