Move scale multiplication to the front
This commit is contained in:
parent
8ff6f09320
commit
d63dbb3acc
|
@ -20,7 +20,7 @@ def split_cross_attention_forward_v1(self, x, context=None, mask=None):
|
|||
|
||||
q = self.to_q(x)
|
||||
context = default(context, x)
|
||||
k = self.to_k(context)
|
||||
k = self.to_k(context) * self.scale
|
||||
v = self.to_v(context)
|
||||
del context, x
|
||||
|
||||
|
@ -85,7 +85,7 @@ def split_cross_attention_forward(self, x, context=None, mask=None):
|
|||
slice_size = q.shape[1] // steps if (q.shape[1] % steps) == 0 else q.shape[1]
|
||||
for i in range(0, q.shape[1], slice_size):
|
||||
end = i + slice_size
|
||||
s1 = einsum('b i d, b j d -> b i j', q[:, i:end], k) * self.scale
|
||||
s1 = einsum('b i d, b j d -> b i j', q[:, i:end], k)
|
||||
|
||||
s2 = s1.softmax(dim=-1, dtype=q.dtype)
|
||||
del s1
|
||||
|
|
Loading…
Reference in New Issue