Skip to content
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

refactor rotary embedding 3: so it is not on cpu #9307

Merged
merged 2 commits into from
Aug 29, 2024
Merged

refactor rotary embedding 3: so it is not on cpu #9307

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
1ff6c42
change get_1d_rotary to accept pos as torch tensors
yiyixuxu Aug 28, 2024
aa0c11f
Merge branch 'main' into fix-torch-rope
yiyixuxu Aug 29, 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
11 changes: 7 additions & 4 deletions src/diffusers/models/embeddings.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -545,11 +545,14 @@ def get_1d_rotary_pos_embed(
assert dim % 2 == 0

if isinstance(pos, int):
pos = np.arange(pos)
pos = torch.arange(pos)
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This should also be passed a device argument to allocate it on the GPU. If this isn't on the GPU, then neither will the following Tensors.

if isinstance(pos, np.ndarray):
pos = torch.from_numpy(pos) # type: ignore # [S]

theta = theta * ntk_factor
freqs = 1.0 / (theta ** (torch.arange(0, dim, 2, dtype=freqs_dtype)[: (dim // 2)] / dim)) / linear_factor # [D/2]
t = torch.from_numpy(pos).to(freqs.device) # type: ignore # [S]
freqs = torch.outer(t, freqs) # type: ignore # [S, D/2]
freqs = freqs.to(pos.device)
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'd expect this to cause a sync as well since by default arange allocates on the CPU. One way to mitigate could be to
a) use pin_memory() on freqs ahead of time and set non_blocking=True
b) do arange on the GPU right away (i.e. torch.arange([...], device=pos.device)).

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ohhh let's do torch.arange([...], device=pos.device)

freqs = torch.outer(pos, freqs) # type: ignore # [S, D/2]
if use_real and repeat_interleave_real:
# flux, hunyuan-dit, cogvideox
freqs_cos = freqs.cos().repeat_interleave(2, dim=1).float() # [S, D]
Expand Down Expand Up @@ -626,7 +629,7 @@ def forward(self, ids: torch.Tensor) -> torch.Tensor:
n_axes = ids.shape[-1]
cos_out = []
sin_out = []
pos = ids.squeeze().float().cpu().numpy()
pos = ids.squeeze().float()
is_mps = ids.device.type == "mps"
freqs_dtype = torch.float32 if is_mps else torch.float64
for i in range(n_axes):
Expand Down
Loading