d38e560e42
Implement some rethinking about LoRA system
...
1. Add an option to allow users to use UNet in fp8/gguf but lora in fp16.
2. All FP16 loras do not need patch. Others will only patch again when lora weight change.
3. FP8 unet + fp16 lora are available (somewhat only available) in Forge now. This also solves some “LoRA too subtle” problems.
4. Significantly speed up all gguf models (in Async mode) by using independent thread (CUDA stream) to compute and dequant at the same time, even when low-bit weights are already on GPU.
5. View “online lora” as a module similar to ControlLoRA so that it is moved to GPU together with model when sampling, achieving significant speedup and perfect low VRAM management simultaneously.
2024-08-19 04:31:59 -07:00
3bef4e331a
change space path order
2024-08-18 23:55:28 -07:00
60dfcd0464
revise space
2024-08-18 19:25:39 -07:00
128a793265
gradio
2024-08-18 03:57:25 -07:00
72ab92f83e
upload meta files
2024-08-18 00:12:53 -07:00
4bb5613916
remove space path after invoke
2024-08-17 08:52:41 -07:00
fcf71fd9ae
fix space logics
2024-08-17 08:42:49 -07:00
93b40f355e
Forge Space and BiRefNet
2024-08-17 08:29:08 -07:00
447f261154
fix
2024-08-15 01:56:21 -07:00
a5f3a50d3f
Not all AUTOMATIC have beard
2024-08-15 01:25:47 -07:00
32fab6e30d
use better name and change to dropdown list
2024-08-15 01:06:17 -07:00
df0fee9396
maybe solve --vae-path
2024-08-14 17:58:12 -07:00
a985afd857
thread safety
2024-08-13 15:45:45 -07:00
bb58520a4c
completely solve 'NoneType' object is not iterable
2024-08-13 15:36:18 -07:00
61f83dd610
support all flux models
2024-08-13 05:42:17 -07:00
b1f0d8c6d1
default img2img back to square
2024-08-11 18:20:18 -07:00
643a485d1a
Update forge_version.py
2024-08-10 19:58:04 -07:00
f10359989f
fix
2024-08-10 19:40:56 -07:00
cfa5242a75
forge 2.0.0
...
see also discussions
2024-08-10 19:24:19 -07:00
593455c4de
global unload after env var change
2024-08-08 22:34:55 -07:00
3f3cb12f76
more tests
2024-08-08 22:12:14 -07:00
02ffb04649
revise stream
2024-08-08 19:23:23 -07:00
60c5aea11b
revise stream logics
2024-08-08 18:45:36 -07:00
20e1ba4a82
fix
2024-08-08 15:08:20 -07:00
6921420b3f
Load Model only when click Generate
...
#964
2024-08-08 14:51:13 -07:00
91f3f628bd
add note to Tiling
2024-08-08 13:33:03 -07:00
a91a81d8e6
revise structure
2024-08-07 20:44:34 -07:00
015587ca59
Delete unet_patcher.py
2024-08-07 20:36:40 -07:00
002341af5b
fix controlnet
...
#961
2024-08-07 19:21:05 -07:00
b57573c8da
Implement many kernels from scratch
2024-08-06 20:19:03 -07:00
78e25a8bae
Update shared_options.py
2024-08-06 16:49:53 -07:00
71c94799d1
diffusion in fp8 landed
2024-08-06 16:47:39 -07:00
dd8997ee2e
use binder
2024-08-06 15:31:19 -07:00
5f908c2bf3
add forge opts section
2024-08-06 15:11:40 -07:00
b4ca5d7420
model load entry for all model loads from webui
...
including startup, checkpoint selection, and that HTML tab of some preview images
2024-08-06 14:38:40 -07:00
6d789653b9
better model load logic
2024-08-06 14:34:57 -07:00
1e8c0f3436
change name
2024-08-06 14:05:56 -07:00
71eaa5ca12
rework UI so that the toolbar is managed by Forge
2024-08-06 13:54:06 -07:00
4189045323
old backend removal - part 1
2024-08-05 03:15:44 -07:00
0863765173
rework sd1.5 and sdxl from scratch
2024-08-05 03:08:17 -07:00
0091e82f53
remove old stuff
2024-08-04 18:45:31 -07:00
a72154405e
Text Processing Engine is Finished
...
100% reproduce all previous results, including TI embeddings, LoRAs in CLIP, emphasize settings, BREAK, timestep swap scheduling, AB mixture, advanced uncond, etc
Backend is 85% finished
2024-08-04 18:42:51 -07:00
430482d1a0
fix cast
2024-08-03 16:08:22 -07:00
8c087f920e
rename files
2024-08-03 15:54:39 -07:00
a5265df7f4
rename file
2024-08-03 15:51:04 -07:00
3ecdcee5a8
add sampler
2024-08-03 15:46:27 -07:00
b8b29a2372
remove forge_sampler
2024-08-03 15:38:07 -07:00
4add428e25
move to new backend - part 2
2024-08-03 15:10:37 -07:00
bb5083f3c2
rework sample function
2024-08-03 13:27:23 -07:00
0871870342
Update supported_controlnet.py
2024-08-03 12:52:32 -07:00
layerdiffusion