diff --git a/ldm/modules/attention.py b/ldm/modules/attention.py
index 622c6fc..509cd87 100644
--- a/ldm/modules/attention.py
+++ b/ldm/modules/attention.py
@@ -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
@@ -168,8 +171,11 @@ 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
-        with torch.autocast(enabled=False, device_type = 'cuda'):
-            q, k = q.float(), k.float()
+        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