AI/diffusers
1
0
Fork 0
mirror of https://github.com/huggingface/diffusers.git synced 2026-01-27 17:22:53 +03:00
Code Activity

Fix dimensionalities in apply_rotary_emb functions' comments (#11717)

Browse Source 
Fix dimensionality in `apply_rotary_emb` functions' comments.
This commit is contained in:
Tolga Cangöz
2025-06-22 01:09:28 +03:00
committed by GitHub
parent 0874dd04dc
commit 7fc53b5d66
2 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/diffusers/models/embeddings.py
View File

@@ -1199,11 +1199,11 @@ def apply_rotary_emb(
if use_real_unbind_dim == -1:
# Used for flux, cogvideox, hunyuan-dit
x_real, x_imag = x.reshape(*x.shape[:-1], -1, 2).unbind(-1) # [B, S, H, D//2]
x_real, x_imag = x.reshape(*x.shape[:-1], -1, 2).unbind(-1) # [B, H, S, D//2]
x_rotated = torch.stack([-x_imag, x_real], dim=-1).flatten(3)
elif use_real_unbind_dim == -2:
# Used for Stable Audio, OmniGen, CogView4 and Cosmos
x_real, x_imag = x.reshape(*x.shape[:-1], 2, -1).unbind(-2) # [B, S, H, D//2]
x_real, x_imag = x.reshape(*x.shape[:-1], 2, -1).unbind(-2) # [B, H, S, D//2]
x_rotated = torch.cat([-x_imag, x_real], dim=-1)
else:
raise ValueError(f"`use_real_unbind_dim={use_real_unbind_dim}` but should be -1 or -2.")

2
src/diffusers/models/transformers/transformer_ltx.py
View File

@@ -481,7 +481,7 @@ class LTXVideoTransformer3DModel(ModelMixin, ConfigMixin, FromOriginalModelMixin
def apply_rotary_emb(x, freqs):
cos, sin = freqs
x_real, x_imag = x.unflatten(2, (-1, 2)).unbind(-1) # [B, S, H, D // 2]
x_real, x_imag = x.unflatten(2, (-1, 2)).unbind(-1) # [B, S, C // 2]
x_rotated = torch.stack([-x_imag, x_real], dim=-1).flatten(2)
out = (x.float() * cos + x_rotated.float() * sin).to(x.dtype)
return out