Commit Graph

253 Commits

Author SHA1 Message Date
Nuullll
a183de04e3 Execute model_loaded_callback after moving to target device 2024-01-06 20:03:33 +08:00
AUTOMATIC1111
267fd5d76b
Merge pull request #14145 from drhead/zero-terminal-snr
Implement zero terminal SNR noise schedule option
2024-01-01 14:45:12 +03:00
Kohaku-Blueleaf
672dc4efa8 Fix forced reload 2023-12-06 15:16:10 +08:00
drhead
78acdcf677 fix variable 2023-12-02 14:09:18 -05:00
drhead
dc1adeecdd Create alphas_cumprod_original on full precision path 2023-12-02 14:06:56 -05:00
Kohaku-Blueleaf
50a21cb09f Ensure the cached weight will not be affected 2023-12-02 22:06:47 +08:00
Kohaku-Blueleaf
110485d5bb Merge branch 'dev' into test-fp8 2023-12-02 17:00:09 +08:00
MrCheeze
6080045b2a Add support for SD 2.1 Turbo, by converting the state dict from SGM to LDM on load 2023-12-01 22:58:05 -05:00
drhead
b25c126ccd
Protect alphas_cumprod from downcasting 2023-11-29 17:38:53 -05:00
Kohaku-Blueleaf
40ac134c55 Fix pre-fp8 2023-11-25 12:35:09 +08:00
Kohaku-Blueleaf
370a77f8e7 Option for using fp16 weight when apply lora 2023-11-21 19:59:34 +08:00
Kohaku-Blueleaf
598da5cd49 Use options instead of cmd_args 2023-11-19 15:50:06 +08:00
Kohaku-Blueleaf
cd12256575 Merge branch 'dev' into test-fp8 2023-11-16 21:53:13 +08:00
AUTOMATIC1111
6ad666e479 more changes for #13865: fix formatting, rename the function, add comment and add a readme entry 2023-11-05 19:46:20 +03:00
AUTOMATIC1111
80d639a440 linter 2023-11-05 19:32:21 +03:00
AUTOMATIC1111
ff805d8d0e
Merge branch 'dev' into master 2023-11-05 19:30:57 +03:00
Ritesh Gangnani
44c5097375 Use devices.torch_gc() instead of empty_cache() 2023-11-05 20:31:57 +05:30
Ritesh Gangnani
ff1609f91e Add SSD-1B as a supported model 2023-11-05 19:13:49 +05:30
Kohaku-Blueleaf
d4d3134f6d ManualCast for 10/16 series gpu 2023-10-28 15:24:26 +08:00
Kohaku-Blueleaf
dda067f64d ignore mps for fp8 2023-10-25 19:53:22 +08:00
Kohaku-Blueleaf
bf5067f50c Fix alphas cumprod 2023-10-25 12:54:28 +08:00
Kohaku-Blueleaf
4830b25136 Fix alphas_cumprod dtype 2023-10-25 11:53:37 +08:00
Kohaku-Blueleaf
1df6c8bfec fp8 for TE 2023-10-25 11:36:43 +08:00
Kohaku-Blueleaf
9c1eba2af3 Fix lint 2023-10-24 02:11:27 +08:00
Kohaku-Blueleaf
eaa9f5162f Add CPU fp8 support
Since norm layer need fp32, I only convert the linear operation layer(conv2d/linear)

And TE have some pytorch function not support bf16 amp in CPU. I add a condition to indicate if the autocast is for unet.
2023-10-24 01:49:05 +08:00
Kohaku-Blueleaf
5f9ddfa46f Add sdxl only arg 2023-10-19 23:57:22 +08:00
Kohaku-Blueleaf
7c128bbdac Add fp8 for sd unet 2023-10-19 13:56:17 +08:00
AUTOMATIC1111
282903bb67 repair unload sd checkpoint button 2023-10-15 09:41:02 +03:00
AUTOMATIC1111
0619df9835 use shallow copy for #13535 2023-10-14 08:01:04 +03:00
AUTOMATIC1111
7cc96429f2
Merge pull request #13535 from chu8129/dev
fix: checkpoints_loaded:{checkpoint:state_dict}, model.load_state_dict issue in dict value empty
2023-10-14 08:00:04 +03:00
wangqiuwen
770ee23f18 reverst 2023-10-07 15:38:50 +08:00
wangqiuwen
76010a51ef up 2023-10-07 15:36:01 +08:00
AUTOMATIC1111
951842d785
Merge pull request #13139 from AUTOMATIC1111/ckpt-dir-path-separator
fix `--ckpt-dir` path separator and option use `short name` for checkpoint dropdown
2023-09-30 10:02:28 +03:00
AUTOMATIC1111
87b50397a6 add missing import, simplify code, use patches module for #13276 2023-09-30 09:11:31 +03:00
AUTOMATIC1111
e309583f29
Merge pull request #13276 from woweenie/patch-1
patch DDPM.register_betas so that users can put given_betas in model yaml
2023-09-30 09:01:12 +03:00
王秋文/qwwang
8e355fbd75 fix 2023-09-18 16:45:42 +08:00
woweenie
d9d94141dc
patch DDPM.register_betas so that users can put given_betas in model yaml 2023-09-15 18:59:44 +02:00
qiuwen.wang
813535d38b
use dict[key]=model; did not update orderdict order, should use move to end 2023-09-15 18:23:23 +08:00
w-e-w
e4726cccf9 parsing string to path 2023-09-08 09:46:34 +09:00
AUTOMATIC1111
503bd3fc0f keep order in list of checkpoints when loading model that doesn't have a checksum 2023-08-30 08:54:41 +03:00
AUTOMATIC1111
f874b1bcad keep order in list of checkpoints when loading model that doesn't have a checksum 2023-08-30 08:54:31 +03:00
AUTOMATIC1111
0232a987bb set devices.dtype_unet correctly 2023-08-23 07:10:43 +03:00
AUTOMATIC1111
016554e437 add --medvram-sdxl 2023-08-22 18:49:08 +03:00
Uminosachi
be301f224d Fix for consistency with shared.opts.sd_vae of UI 2023-08-21 11:28:53 +09:00
Uminosachi
549b0fc526 Change where VAE state are stored in model 2023-08-20 23:06:51 +09:00
Uminosachi
af5d2e8e5f Change to access sd_model attribute with dot 2023-08-20 20:08:22 +09:00
Uminosachi
5159edbf0e Store base_vae and loaded_vae_file in sd_model 2023-08-20 19:44:37 +09:00
Uminosachi
042e1d5d0b Fix SD VAE switch error after model reuse 2023-08-20 15:00:14 +09:00
AUTOMATIC1111
0dc74545c0 resolve the issue with loading fp16 checkpoints while using --no-half 2023-08-17 07:54:07 +03:00
AUTOMATIC1111
eaba3d7349 send weights to target device instead of CPU memory 2023-08-16 12:11:01 +03:00