Add better debug print for failed prepare (#3950)

Add better debug print for failed prepare (#3950)

Summary:
Pull Request resolved: #3950

If you pass a "raw" graph_module from an exported module, prepare will fail.  It is not immediately clear why.

This PR makes it more clear.

Differential Revision: D94379595

Author: JakeStevens

torchao/quantization/pt2e/utils.py

@@ -595,7 +595,13 @@ def _is_connected(source: torch.fx.Node, dest: torch.fx.Node) -> bool:
 def _get_tensor_constant_from_node(node, m):
     if node is None:
         return None
-    assert node.op == "get_attr"
+    if node.op == "placeholder":
+        raise ValueError(
+            "Expected get_attr node, got placeholder. "
+            "If working with an ExportedProgram, make sure you are using .module()"
+        )
+    if node.op != "get_attr":
+        raise ValueError(f"Expected get_attr node, got {node.op}")
     target_atoms = node.target.split(".")
     attr_itr = m
     for i, atom in enumerate(target_atoms):