298977887/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2025-02-28 22:51:45 +00:00
Code Issues Packages Projects Releases Wiki Activity

Slightly lower hunyuan video memory usage.

Browse Source
This commit is contained in:
comfyanonymous 2025-01-16 00:23:01 -05:00
parent 25683b5b02
commit 6320d05696
1 changed files with 8 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
comfy/ldm/flux/math.py
View File

@ -5,8 +5,15 @@ from torch import Tensor
from comfy.ldm.modules.attention import optimized_attention from comfy.ldm.modules.attention import optimized_attention
import comfy.model_management import comfy.model_management
def attention(q: Tensor, k: Tensor, v: Tensor, pe: Tensor, mask=None) -> Tensor: def attention(q: Tensor, k: Tensor, v: Tensor, pe: Tensor, mask=None) -> Tensor:
q, k = apply_rope(q, k, pe) q_shape = q.shape
k_shape = k.shape
q = q.float().reshape(*q.shape[:-1], -1, 1, 2)
k = k.float().reshape(*k.shape[:-1], -1, 1, 2)
q = (pe[..., 0] * q[..., 0] + pe[..., 1] * q[..., 1]).reshape(*q_shape).type_as(v)
k = (pe[..., 0] * k[..., 0] + pe[..., 1] * k[..., 1]).reshape(*k_shape).type_as(v)
heads = q.shape[1] heads = q.shape[1]
x = optimized_attention(q, k, v, heads, skip_reshape=True, mask=mask) x = optimized_attention(q, k, v, heads, skip_reshape=True, mask=mask)