32 Commits

Author	SHA1	Message	Date
Sean Sube	15b9b51ed5	fix(api): make sure new SD converter does not produce extra files	2023-12-24 10:47:46 -06:00
Sean Sube	29ccd16437	fix(api): handle HF repos in new SD converter	2023-12-24 07:05:56 -06:00
Sean Sube	da6e813638	rename old converter	2023-12-24 06:24:45 -06:00
Sean Sube	769350115c	pass correct paths to torch extraction	2023-12-24 06:00:44 -06:00
Sean Sube	9ffe266384	fix(api): unload UNet before CNet conversion	2023-12-24 05:32:39 -06:00
Sean Sube	054e5c406a	fully rename old converter to legacy	2023-12-24 05:31:55 -06:00
Sean Sube	9a077fae15	feat(api): replace legacy SD converter with optimum-based one by default	2023-12-23 22:55:52 -06:00
Sean Sube	f6e6e31789	add controlnet support to optimum converter	2023-12-23 22:33:46 -06:00
Sean Sube	ebe813d035	fix(api): change weights filename for new models to match optimum	2023-12-23 22:18:35 -06:00
Sean Sube	14208de393	register optimum-based SD converter	2023-12-23 22:18:09 -06:00
Sean Sube	b6ef00e437	feat(api): add new optimum-based SD converter	2023-12-23 22:09:57 -06:00
Sean Sube	2b8b59a39c	make SD conversion more like SDXL	2023-12-23 21:42:36 -06:00
Sean Sube	2546084208	lint(api): move some noisy log messages to trace level	2023-12-21 21:07:02 -06:00
Sean Sube	e09631ef05	apply lint	2023-12-19 23:01:05 -06:00
Sean Sube	c8a9734acf	dump unet to torch before making cnet	2023-12-19 22:50:18 -06:00
Sean Sube	4a87fb2a31	update forwarding shapes, apply lint	2023-12-19 22:10:37 -06:00
Sean Sube	b4262eb777	remove bulky controlnet wrapper	2023-12-19 20:56:24 -06:00
Sean Sube	ba9982af51	fix(api): various controlnet fixes	2023-12-19 19:19:48 -06:00
Sean Sube	133c4a20bd	fix(api): use correct post-fetch path when converting from checkpoints (#432)	2023-12-16 13:17:24 -06:00
Sean Sube	95d8f4a598	fix(api): make sure all file types are covered (#432)	2023-12-14 20:04:08 -06:00
Sean Sube	4da4cd95a5	fix(api): trim whitespace from model names because it breaks things (#376)	2023-12-10 13:59:47 -06:00
Sean Sube	9c1fcd16fa	fix(api): only fetch diffusion models if they have not already been converted (#398)	2023-12-10 13:52:52 -06:00
Sean Sube	e7da2cf8a6	fix(api): load pretrained VAE from original path	2023-12-08 23:19:52 -06:00
Sean Sube	4fd50984f0	fix(api): correct VAE extension check during conversion	2023-12-08 22:55:51 -06:00
Sean Sube	9b4ae0916b	lint and test fixes	2023-12-03 12:53:50 -06:00
Sean Sube	4a2498ad8d	apply lint	2023-11-08 22:04:15 -06:00
Sean Sube	f4f3bda6f8	fix(api): allow all supported tensors extensions for VAE files	2023-11-08 22:00:56 -06:00
Sean Sube	0ecae65f88	fix(api): set VAE attn processor during conversion	2023-09-24 15:02:21 -05:00
Sean Sube	5d3a7d77a5	fix(api): use Torch pipelines while loading models for conversion	2023-09-24 10:04:21 -05:00
Sean Sube	c99481f484	fix(api): load replacement VAE from single file for SD v1/v2	2023-09-24 09:49:50 -05:00
Sean Sube	8a5e211172	lint(api): deduplicate some imports	2023-09-13 17:27:05 -05:00
Sean Sube	fe68670844	feat(api): add conversion for SDXL models	2023-09-10 11:52:46 -05:00