Skip to content

[core] improve VAE encode/decode framewise batching #9684

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Oct 16, 2024
Merged

[core] improve VAE encode/decode framewise batching #9684

Changes from 2 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
9b76a78
update
a-r-r-o-w Oct 15, 2024
7a0f667
Merge branch 'main' into cog-vae-num-batches
sayakpaul Oct 16, 2024
8d9f7da
apply suggestions from review
a-r-r-o-w Oct 16, 2024
4bb567c
Merge branch 'main' into cog-vae-num-batches
a-r-r-o-w Oct 16, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 5 additions & 3 deletions src/diffusers/models/autoencoders/autoencoder_kl_cogvideox.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1182,7 +1182,8 @@ def _encode(self, x: torch.Tensor) -> torch.Tensor:

frame_batch_size = self.num_sample_frames_batch_size
# Note: We expect the number of frames to be either `1` or `frame_batch_size * k` or `frame_batch_size * k + 1` for some k.
num_batches = num_frames // frame_batch_size if num_frames > 1 else 1
# So, it is okay to not round up as the extra remaining frame is handled in the loop
num_batches = max(num_frames // frame_batch_size, 1)
conv_cache = None
enc = []

Expand Down Expand Up @@ -1330,7 +1331,8 @@ def tiled_encode(self, x: torch.Tensor) -> torch.Tensor:
row = []
for j in range(0, width, overlap_width):
# Note: We expect the number of frames to be either `1` or `frame_batch_size * k` or `frame_batch_size * k + 1` for some k.
num_batches = num_frames // frame_batch_size if num_frames > 1 else 1
# So, it is okay to not round up as the extra remaining frame is handled in the loop
num_batches = max(num_frames // frame_batch_size, 1)
conv_cache = None
time = []

Expand Down Expand Up @@ -1409,7 +1411,7 @@ def tiled_decode(self, z: torch.Tensor, return_dict: bool = True) -> Union[Decod
for i in range(0, height, overlap_height):
row = []
for j in range(0, width, overlap_width):
num_batches = num_frames // frame_batch_size
num_batches = max(num_frames // frame_batch_size, 1)
conv_cache = None
time = []

Expand Down
Loading