JaehaerysNL/StableDiffusion
1
0
Fork 0
mirror of https://github.com/Stability-AI/stablediffusion.git synced 2024-12-22 15:44:58 +00:00
Code Issues Projects Releases Packages Wiki Activity

Merge pull request #89 from Stability-AI/dango.patch.atten_overflow

Browse source 
* Force cast to fp32 to avoid atten layer overflow
This commit is contained in:
Robin Rombach 2022-12-07 14:54:35 +01:00 committed by GitHub
commit 8bde0cf64f
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 12 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
ldm/modules/attention.py
View file

@ -16,6 +16,9 @@ try:
except:
XFORMERS_IS_AVAILBLE = False
# CrossAttn precision handling
import os
_ATTN_PRECISION = os.environ.get("ATTN_PRECISION", "fp32")
def exists(val):
return val is not None
@ -167,7 +170,14 @@ class CrossAttention(nn.Module):
q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q, k, v))
# force cast to fp32 to avoid overflowing
if _ATTN_PRECISION =="fp32":
with torch.autocast(enabled=False, device_type = 'cuda'):
q, k = q.float(), k.float()
sim = einsum('b i d, b j d -> b i j', q, k) * self.scale
else:
sim = einsum('b i d, b j d -> b i j', q, k) * self.scale
del q, k
if exists(mask):