34 Commits

Author	SHA1	Message	Date
kijai	729a6485ea	expose sageattn 2.0.0 functions ... _cuda versions seem to be required on RTX 30xx -series GPUs for sageattn + CogVideoX 1.5	2024-12-01 18:45:10 +02:00
kijai	b74aa75026	Don't use autocast with fp/bf16	2024-11-20 14:22:10 +02:00
Dango233	b31a025673	Fix fused sdpa	2024-11-20 17:40:28 +08:00
kijai	b0eabeba24	fix comfy attention output shape	2024-11-19 20:18:13 +02:00
kijai	882faa6dea	add comfyui attention mode	2024-11-19 19:55:51 +02:00
kijai	128f89c4d2	Update workflows, fix controlnet	2024-11-19 15:23:38 +02:00
kijai	a7646c0d6f	refactor ... - unify all pipelines into one - unify transformer model into one - separate VAE - add single file model loading	2024-11-19 03:04:22 +02:00
kijai	6f9e4ff647	Update custom_cogvideox_transformer_3d.py	2024-11-17 22:23:40 +02:00
kijai	e70da23ac2	exclude sageattn from compile	2024-11-17 22:11:32 +02:00
kijai	eebdc412f9	fix sageattention	2024-11-17 21:43:53 +02:00
kijai	bececf0189	some experimental optimizations	2024-11-16 17:32:31 +02:00
kijai	dac6a2a3ac	allow limiting blocks to cache	2024-11-12 08:42:01 +02:00
kijai	0a121dba53	fix FasterCache	2024-11-12 07:39:00 +02:00
kijai	6931576916	update from upstream, ofs embeds	2024-11-11 18:53:12 +02:00
kijai	5f1a917b93	padding fix	2024-11-11 17:29:57 +02:00
kijai	ca63f5dade	update	2024-11-11 01:19:11 +02:00
kijai	fb246f95ef	attention compile works with higher cache_size_limit	2024-11-09 22:56:50 +02:00
kijai	634c22db50	sageattn	2024-11-09 17:05:55 +02:00
kijai	643bbc18c1	i2v	2024-11-09 12:13:52 +02:00
Jukka Seppänen	b563994afc	finally ... works	2024-11-09 04:02:36 +02:00
Jukka Seppänen	9aab678a9e	test	2024-11-09 03:15:21 +02:00
kijai	e783951dad	maybe	2024-11-09 02:24:18 +02:00
kijai	2074ba578e	doesn't work yet	2024-11-08 21:24:20 +02:00
kijai	1cc6e1f070	use diffusers LoRA loading to support fusing for DimensionX LoRAs ... https://github.com/wenqsun/DimensionX	2024-11-08 14:24:32 +02:00
kijai	5b4819ba65	support Tora for Fun -models	2024-10-29 10:44:09 +02:00
kijai	66ba4e1ee7	fix fastercache start step	2024-10-28 22:52:30 +02:00
kijai	e9fc26b5e3	initial experimental FasterCache support for 2b models	2024-10-28 21:02:10 +02:00
kijai	dcca095743	make torch compile work better	2024-10-26 02:33:29 +03:00
kijai	2cc521062f	correct Tora fuser dtype ... I think...	2024-10-22 18:40:31 +03:00
kijai	a654821515	testing Tora for I2V	2024-10-21 22:53:36 +03:00
kijai	256a638ee4	cleanup, bugfixes	2024-10-21 03:24:53 +03:00
kijai	e8bc2fd052	Initial Tora implementation ... https://github.com/alibaba/Tora	2024-10-21 00:11:39 +03:00
kijai	d76229c49b	controlnet support ... https://huggingface.co/TheDenk/cogvideox-2b-controlnet-hed-v1 https://huggingface.co/TheDenk/cogvideox-2b-controlnet-canny-v1	2024-10-08 16:22:07 +03:00
Jukka Seppänen	49451c4a22	support sageattn ... https://github.com/thu-ml/SageAttention	2024-10-06 20:26:03 +03:00