Hasham Vakani ⚡ 15547a8720 fix: Add NVIDIA Blackwell (RTX 50xx, sm_120) support ... - Use bfloat16 dtype for UNet on Blackwell GPUs (compute major >= 12) which have native bf16 tensor core support - Skip manual_cast for bfloat16 weights to avoid unnecessary casting - Fix numpy TypeError with bfloat16 tensors in patch.py and ip_adapter.py by converting to float32 before .numpy() calls Tested on RTX 5070 (sm_120, CUDA 12.8) with PyTorch nightly (cu128). Generates images at ~3.2 it/s including Image Prompt (IP-Adapter) mode. Fixes #3862, #4123, #4141		2026-03-04 05:43:44 +05:00
..
BLIP	Use weights_only for loading (#3427)	2024-08-03 12:33:01 +02:00
GroundingDINO	Sync branch 'mashb1t_main' with develop_upstream	2024-07-14 20:28:38 +02:00
facexlib	Use weights_only for loading (#3427)	2024-08-03 12:33:01 +02:00
safety_checker	feat: add nsfw image censoring via config and checkbox (#958)	2024-05-18 15:50:28 +02:00
sam	Sync branch 'mashb1t_main' with develop_upstream	2024-07-14 20:28:38 +02:00
censor.py	Sync branch 'mashb1t_main' with develop_upstream	2024-07-14 20:28:38 +02:00
expansion.py	feat: add early return for prompt expansion when no new tokens should be added	2024-02-19 15:22:10 +01:00
face_crop.py	2.1.826	2023-12-12 11:38:05 -08:00
inpaint_mask.py	Sync branch 'mashb1t_main' with develop_upstream	2024-07-14 20:28:38 +02:00
interrogate.py	use our blip	2023-12-12 21:07:39 -08:00
ip_adapter.py	fix: Add NVIDIA Blackwell (RTX 50xx, sm_120) support	2026-03-04 05:43:44 +05:00
preprocessors.py	feat: advanced params refactoring + prevent users from skipping/stopping other users tasks in queue (#981)	2024-02-24 19:01:06 +01:00
resampler.py	try fix #1761 again (#1680)	2023-12-31 04:38:31 -08:00
vae_interpose.py	feat: update interposer from v3.1 to v4.0 (#2717)	2024-04-06 15:27:35 +02:00
wd14tagger.py	safer link	2023-12-12 21:53:13 -08:00