AI/diffusers
1
0
Fork 0
mirror of https://github.com/huggingface/diffusers.git synced 2026-10-23 09:09:01 +03:00
Code Activity

chore: add one additional copied from in qwenimage 2.1 vae (#14810)

Browse Source 
* add one additional copied from in qwenimage 2.1 vae.

* formattinmg
This commit is contained in:
Sayak Paul
2026-10-01 13:51:28 +05:30
committed by GitHub
parent acbabca385
commit 7997e4eabb
2 changed files with 9 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/diffusers/models/autoencoders/autoencoder_kl_qwenimage21.py
View File

@@ -31,6 +31,7 @@ logger = logging.get_logger(__name__) # pylint: disable=invalid-name
CACHE_T = 2
# Copied from diffusers.models.autoencoders.autoencoder_kl_wan.AvgDown3D with AvgDown3D->QwenImage21AvgDown3D
class QwenImage21AvgDown3D(nn.Module):
def __init__(
self,
@@ -46,7 +47,6 @@ class QwenImage21AvgDown3D(nn.Module):
f"`in_channels` ({in_channels}) times the downsampling factor ({factor}) must be divisible by "
f"`out_channels` ({out_channels})."
)
self.in_channels = in_channels
self.out_channels = out_channels
self.factor_t = factor_t

12
src/diffusers/models/autoencoders/autoencoder_kl_wan.py
View File

@@ -40,14 +40,18 @@ class AvgDown3D(nn.Module):
factor_s=1,
):
super().__init__()
factor = factor_t * factor_s * factor_s
if in_channels * factor % out_channels != 0:
raise ValueError(
f"`in_channels` ({in_channels}) times the downsampling factor ({factor}) must be divisible by "
f"`out_channels` ({out_channels})."
)
self.in_channels = in_channels
self.out_channels = out_channels
self.factor_t = factor_t
self.factor_s = factor_s
self.factor = self.factor_t * self.factor_s * self.factor_s
assert in_channels * self.factor % out_channels == 0
self.group_size = in_channels * self.factor // out_channels
self.factor = factor
self.group_size = in_channels * factor // out_channels
def forward(self, x: torch.Tensor) -> torch.Tensor:
pad_t = (self.factor_t - x.shape[2] % self.factor_t) % self.factor_t