fix enable memory efficient attention on ROCm

while calling CK implementation
huggingface · Jan 15, 2025 · 36d34fd · 36d34fd
1 parent b0c8973
commit 36d34fd
Showing 1 changed file with 6 additions and 5 deletions.
diff --git a/src/diffusers/models/attention_processor.py b/src/diffusers/models/attention_processor.py
@@ -399,11 +399,12 @@ def set_use_memory_efficient_attention_xformers(
             else:
                 try:
                     # Make sure we can run the memory efficient attention
-                    _ = xformers.ops.memory_efficient_attention(
-                        torch.randn((1, 2, 40), device="cuda"),
-                        torch.randn((1, 2, 40), device="cuda"),
-                        torch.randn((1, 2, 40), device="cuda"),
-                    )
+                    dtype = None
+                    if attention_op is not None:
+                        op_fw, op_bw = attention_op
+                        dtype = list(op_fw.SUPPORTED_DTYPES)[0]
+                    q = torch.randn((1, 2, 40), device="cuda", dtype=dtype)
+                    _ = xformers.ops.memory_efficient_attention(q, q, q)
                 except Exception as e:
                     raise e