AI/diffusers
1
0
Fork 0
mirror of https://github.com/huggingface/diffusers.git synced 2026-01-29 07:22:12 +03:00
Code Activity

Support for Offset Noise in examples (#2753)

Browse Source 
* add noise offset

* make style
This commit is contained in:
Haofan Wang
2023-03-23 12:06:17 +08:00
committed by GitHub
parent 92e1164e2e
commit e0d8c9ef83
2 changed files with 14 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
examples/text_to_image/train_text_to_image.py
View File

@@ -297,6 +297,7 @@ def parse_args():
parser.add_argument(
"--enable_xformers_memory_efficient_attention", action="store_true", help="Whether or not to use xformers."
)
parser.add_argument("--noise_offset", type=float, default=0, help="The scale of noise offset.")
args = parser.parse_args()
env_local_rank = int(os.environ.get("LOCAL_RANK", -1))
@@ -705,6 +706,12 @@ def main():
# Sample noise that we'll add to the latents
noise = torch.randn_like(latents)
if args.noise_offset:
# https://www.crosslabs.org//blog/diffusion-with-offset-noise
noise += args.noise_offset * torch.randn(
(latents.shape[0], latents.shape[1], 1, 1), device=latents.device
)
bsz = latents.shape[0]
# Sample a random timestep for each image
timesteps = torch.randint(0, noise_scheduler.num_train_timesteps, (bsz,), device=latents.device)

7
examples/text_to_image/train_text_to_image_lora.py
View File

@@ -333,6 +333,7 @@ def parse_args():
parser.add_argument(
"--enable_xformers_memory_efficient_attention", action="store_true", help="Whether or not to use xformers."
)
parser.add_argument("--noise_offset", type=float, default=0, help="The scale of noise offset.")
args = parser.parse_args()
env_local_rank = int(os.environ.get("LOCAL_RANK", -1))
@@ -718,6 +719,12 @@ def main():
# Sample noise that we'll add to the latents
noise = torch.randn_like(latents)
if args.noise_offset:
# https://www.crosslabs.org//blog/diffusion-with-offset-noise
noise += args.noise_offset * torch.randn(
(latents.shape[0], latents.shape[1], 1, 1), device=latents.device
)
bsz = latents.shape[0]
# Sample a random timestep for each image
timesteps = torch.randint(0, noise_scheduler.num_train_timesteps, (bsz,), device=latents.device)