Add a supports_fp64 function. (#13368)

comfy/model_management.py

@@ -1732,6 +1732,21 @@ def supports_mxfp8_compute(device=None):
 
     return True
 
+def supports_fp64(device=None):
+    if is_device_mps(device):
+        return False
+
+    if is_intel_xpu():
+        return False
+
+    if is_directml_enabled():
+        return False
+
+    if is_ixuca():
+        return False
+
+    return True
+
 def extended_fp16_support():
     # TODO: check why some models work with fp16 on newer torch versions but not on older
     if torch_version_numeric < (2, 7):