a527d0c795
Code refactor.
2023-11-07 19:33:40 -05:00
2a23ba0b8c
Fix unet ops not entirely on GPU.
2023-11-07 04:30:37 -05:00
c837a173fa
Fix some memory issues in sub quad attention.
2023-10-30 15:30:49 -04:00
125b03eead
Fix some OOM issues with split attention.
2023-10-30 13:14:11 -04:00
6ec3f12c6e
Support SSD1B model and make it easier to support asymmetric unets.
2023-10-27 14:45:15 -04:00
a373367b0c
Fix some OOM issues with split and sub quad attention.
2023-10-25 20:17:28 -04:00
8b65f5de54
attention_basic now works with hypertile.
2023-10-22 03:59:53 -04:00
e6bc42df46
Make sub_quad and split work with hypertile.
2023-10-22 03:51:29 -04:00
9906e3efe3
Make xformers work with hypertile.
2023-10-21 13:23:03 -04:00
d44a2de49f
Make VAE code closer to sgm.
2023-10-17 15:18:51 -04:00
23680a9155
Refactor the attention stuff in the VAE.
2023-10-17 03:19:29 -04:00
bb064c9796
Add a separate optimized_attention_masked function.
2023-10-16 02:31:24 -04:00
9a55dadb4c
Refactor code so model can be a dtype other than fp32 or fp16.
2023-10-13 14:41:17 -04:00
88733c997f
pytorch_attention_enabled can now return True when xformers is enabled.
2023-10-11 21:30:57 -04:00
ac7d8cfa87
Allow attn_mask in attention_pytorch.
2023-10-11 20:38:48 -04:00
1a4bd9e9a6
Refactor the attention functions.
...
There's no reason for the whole CrossAttention object to be repeated when
only the operation in the middle changes.
2023-10-11 20:38:48 -04:00
fff491b032
Model patches can now know which batch is positive and negative.
2023-09-27 12:04:07 -04:00
afa2399f79
Add a way to set output block patches to modify the h and hsp.
2023-09-22 20:26:47 -04:00
1938f5c5fe
Add a force argument to soft_empty_cache to force a cache empty.
2023-09-04 00:58:18 -04:00
2da73b7073
Revert changes in comfy/ldm/modules/diffusionmodules/util.py, which is unused.
2023-09-02 20:07:52 -07:00
4a0c4ce4ef
Some fixes to generalize CUDA specific functionality to Intel or other GPUs.
2023-09-02 18:22:10 -07:00
0e3b641172
Remove xformers related print.
2023-09-01 02:12:03 -04:00
bed116a1f9
Remove optimization that caused border.
2023-08-29 11:21:36 -04:00
1c794a2161
Fallback to slice attention if xformers doesn't support the operation.
2023-08-27 22:24:42 -04:00
d935ba50c4
Make --bf16-vae work on torch 2.0
2023-08-27 21:33:53 -04:00
cf5ae46928
Controlnet/t2iadapter cleanup.
2023-08-22 01:06:26 -04:00
b80c3276dc
Fix issue with gligen.
2023-08-18 16:32:23 -04:00
d6e4b342e6
Support for Control Loras.
...
Control loras are controlnets where some of the weights are stored in
"lora" format: an up and a down low rank matrice that when multiplied
together and added to the unet weight give the controlnet weight.
This allows a much smaller memory footprint depending on the rank of the
matrices.
These controlnets are used just like regular ones.
2023-08-18 11:59:51 -04:00
2b13939044
Remove some useless code.
2023-07-30 14:13:33 -04:00
95d796fc85
Faster VAE loading.
2023-07-29 16:28:30 -04:00
4b957a0010
Initialize the unet directly on the target device.
2023-07-29 14:51:56 -04:00
9ba440995a
It's actually possible to torch.compile the unet now.
2023-07-18 21:36:35 -04:00
ddc6f12ad5
Disable autocast in unet for increased speed.
2023-07-05 21:58:29 -04:00
103c487a89
Cleanup.
2023-07-02 11:58:23 -04:00
78d8035f73
Fix bug with controlnet.
2023-06-24 11:02:38 -04:00
05676942b7
Add some more transformer hooks and move tomesd to comfy_extras.
...
Tomesd now uses q instead of x to decide which tokens to merge because
it seems to give better results.
2023-06-24 03:30:22 -04:00
fa28d7334b
Remove useless code.
2023-06-23 12:35:26 -04:00
f87ec10a97
Support base SDXL and SDXL refiner models.
...
Large refactor of the model detection and loading code.
2023-06-22 13:03:50 -04:00
9fccf4aa03
Add original_shape parameter to transformer patch extra_options.
2023-06-21 13:22:01 -04:00
8883cb0f67
Add a way to set patches that modify the attn2 output.
...
Change the transformer patches function format to be more future proof.
2023-06-18 22:58:22 -04:00
ae43f09ef7
All the unet weights should now be initialized with the right dtype.
2023-06-15 18:42:30 -04:00
7bf89ba923
Initialize more unet weights as the right dtype.
2023-06-15 15:00:10 -04:00
e21d9ad445
Initialize transformer unet block weights in right dtype at the start.
2023-06-15 14:29:26 -04:00
21f04fe632
Disable default weight values in unet conv2d for faster loading.
2023-06-14 19:46:08 -04:00
9d54066ebc
This isn't needed for inference.
2023-06-14 13:05:08 -04:00
6971646b8b
Speed up model loading a bit.
...
Default pytorch Linear initializes the weights which is useless and slow.
2023-06-14 12:09:41 -04:00
274dff3257
Remove more useless files.
2023-06-13 02:22:19 -04:00
f0a2b81cd0
Cleanup: Remove a bunch of useless files.
2023-06-13 02:19:08 -04:00
b8636a44aa
Make scaled_dot_product switch to sliced attention on OOM.
2023-05-20 16:01:02 -04:00
797c4e8d3b
Simplify and improve some vae attention code.
2023-05-20 15:07:21 -04:00
comfyanonymous
Simon Lui