wassname/denoising-diffusion-pytorch
1
0
Fork 0
mirror of https://github.com/wassname/denoising-diffusion-pytorch.git synced 2026-09-04 16:24:42 +08:00
Code Issues Packages Projects Releases Wiki Activity
180 Commits 2 Branches 127 Tags
patch-1
Commit Graph
139 Commits
Author SHA1 Message Date
wassname edc6d5136a Update denoising_diffusion_pytorch_1d.py 2022-12-22 11:00:56 +08:00
wassname 207d23af86 Update denoising_diffusion_pytorch.py 
Shouldn't eta be 0 for DDIM sampling? In the DDIM paper in section 5.1 they mention the eta=0 is DDIM, and eta=1 approximates the normal DDPM. They also use [eta=0](https://github.com/openai/improved-diffusion/blob/e94489283bb876ac1477d5dd7709bbbd2d9902ce/improved_diffusion/gaussian_diffusion.py#L487) in the openai implementation of DDIM.
 2022-12-22 10:57:30 +08:00
Phil Wang cf4f44909f 0.1.5 2022-12-21 08:35:59 -08:00
Phil Wang a7f2d670bb always set results_folder on Trainer 2022-12-21 08:34:34 -08:00
Phil Wang 6ab29d5cea start saving version of library in the checkpoints, so in case of breaking changes, researchers can return to the right working version 2022-11-28 12:11:11 -08:00
Phil Wang ddc31bc489 trust paper 2022-11-27 10:05:55 -08:00
Phil Wang 60f760ead5 add classifier free guidance example for @ajrheng at https://github.com/lucidrains/denoising-diffusion-pytorch/issues/120 2022-11-11 20:01:42 -08:00
Phil Wang aece9c2a2a help researchers in need of 1d ddpm, ex. https://github.com/lucidrains/denoising-diffusion-pytorch/issues/116 2022-11-05 11:42:00 -07:00
Phil Wang 12079cadee add the v-parameterization from Salimans et al for the discrete case, allow for distillation 2022-10-30 09:53:28 -07:00
Phil Wang 0ffff59ca0 add option for random fourier features, given misinterpretation of Katherine's code, thanks to @tmabraham for addressing this in https://github.com/lucidrains/denoising-diffusion-pytorch/issues/112 2022-10-30 09:32:06 -07:00
Phil Wang aadaa7d288 bring in the continuous time v-parameterized ddpm, validated to work locally, and which will be used for imagen-video replication 2022-10-23 19:16:31 -07:00
Phil Wang 23fd887a5f switch back to regular attention, given @rromb results 2022-10-16 08:43:10 -07:00
Nicolas Pielawski 2ac3f94a80 Added gradient clipping. 2022-10-05 11:29:49 -07:00
Phil Wang 6e8a0f2082 fix auto-conversion of images to mode in dataset 2022-09-20 19:29:35 -07:00
Kashif Rasul d85b8bbe2e fix torch.cumprod 2022-09-16 17:19:00 +02:00
Ryan Burgert 7c0cd05c27 Trainer.load can use GPU's other than cuda:0 2022-09-04 22:11:35 -04:00
Phil Wang 6dda508ff6 in ddim, clip x0 before calculation of predicted noise, thanks to @lukovnikov again for pointing out this inconsistency with glides implementation 2022-09-01 09:34:50 -07:00
Phil Wang 9ec8d27217 0.27.7 2022-08-31 07:36:19 -07:00
lukovnikov e4a4e4acaa Revert "Revert "fix ddim sampling"" 
This reverts commit cd8329cdd7.
 2022-08-31 15:28:35 +02:00
lukovnikov cd8329cdd7 Revert "fix ddim sampling" 
This reverts commit aec2a26984.
 2022-08-31 15:25:51 +02:00
lukovnikov aec2a26984 fix ddim sampling 2022-08-31 15:24:59 +02:00
TheDudeFromCI e46a89e2bc Update step before training checkpoint 2022-08-31 02:19:12 -07:00
Phil Wang 42158d6248 fix ddim, for issue https://github.com/lucidrains/denoising-diffusion-pytorch/issues/81 2022-08-30 20:30:29 -07:00
Phil Wang d9275a744c add weight standardization prior to groupnorm, lessen cosine sim attention scale to 10 for fp16 2022-08-17 11:42:54 -07:00
Phil Wang beb2f2d8dd add self conditioning for elucidated ddpm 2022-08-10 13:15:45 -07:00
Phil Wang f0d59acdfd fix sampling ddpm tqdm 2022-08-10 12:02:56 -07:00
Phil Wang 689593a579 add the new self conditioning technique from hintons group from bit diffusion paper 
0.27.0
 2022-08-10 10:52:07 -07:00
Phil Wang eba44498d1 higher epsilon for fp16 in layernorm 2022-07-29 13:29:52 -07:00
Phil Wang 12f95b33d8 rescale values to prevent linear attention from overflowing in fp16 setting 2022-07-27 12:25:40 -07:00
Phil Wang 6b504c4ae9 fix accelerator prepare bug for dataloader 2022-07-25 08:13:46 -07:00
Phil Wang 37334ae824 fix a bug with ddim and predict x0 objective 2022-07-18 19:04:57 -07:00
Phil Wang 555566c188 take a gamble on cosine sim attention 2022-07-18 11:29:20 -07:00
Phil Wang 2b742dd2cc move accelerator backward outside of autocast context, also calculate total loss correctly across gradient accumulated steps 2022-07-11 21:02:06 -07:00
Phil Wang 1345a8a41d do not noise at the last timestep for ddim 2022-07-09 18:36:45 -07:00
Phil Wang 931a5af2c3 bring in ddim sampling 2022-07-09 16:10:23 -07:00
Phil Wang a0c3443eaa optimizer should be saved and loaded 2022-07-08 17:43:41 -07:00
Phil Wang 662172851b add convert_image_to keyword argument, for forcing images being loaded to be converted to some format, greyscale, rgb, rgba, whatever 2022-07-08 09:22:59 -07:00
Phil Wang 0248b5e4d3 also make sure grad scaler actually exists in the saved pt file 2022-07-06 11:48:04 -07:00
Phil Wang 6b56af08a2 support multi-gpu training using huggingface accelerate, addressing https://github.com/lucidrains/denoising-diffusion-pytorch/pull/54 2022-07-06 11:45:56 -07:00
Phil Wang d4420248f1 tqdm auto instead 2022-06-29 19:48:49 -07:00
Phil Wang a536e5bee9 reorganize elucidating code 2022-06-29 08:49:38 -07:00
Phil Wang 1b85379d3a add clamping option to elucidated diffusion 2022-06-29 08:32:10 -07:00
Arya Aftab d97bc0278c Solve issue #26 2022-06-29 11:42:54 +04:30
Phil Wang 8408775cfc fix bug in elucidating sampling 2022-06-28 17:52:02 -07:00
Phil Wang 86fcb6785b release elucidating diffusion 2022-06-28 17:39:48 -07:00
Phil Wang 5db64fec4b refactor sigmas and gamma generation 2022-06-28 17:26:58 -07:00
Phil Wang b87ea27781 fix off by one 2022-06-28 16:05:21 -07:00
Phil Wang a8403b83fe no clamping when training from sigmas drawn from log normal distribution, clamp final images being sampled 2022-06-28 16:00:16 -07:00
Phil Wang f4b1d7a67c complete a first pass of elucidated ddpm 2022-06-28 15:15:21 -07:00
Phil Wang 618493714f clamp the sigma coming out of the log normal distribution 2022-06-28 14:34:39 -07:00