Remove strictness in export calls

Summary: The new PT default has been to do non-strict export for a little while now. We have explicit strict because the PT compiler team preserved the current behavior at the time, but we should be able to remove it and match the default. It might also make more models go through.

Differential Revision: D73859317

Author: mcremon-meta

backends/cadence/aot/compiler.py

@@ -90,7 +90,7 @@ def prepare_and_convert_pt2(
     remove_decompositions(decomp_table, ops_to_keep)
     # Export with dynamo
     model_gm = (
-        torch.export.export_for_training(model, inputs, strict=True)
+        torch.export.export_for_training(model, inputs)
         .run_decompositions(decomp_table)
         .module()
     )
@@ -192,7 +192,7 @@ def export_program(
     torch._C._set_mkldnn_enabled(False)
 
     # Export the model and return it.
-    expo_program = export(model, inputs, strict=True)
+    expo_program = export(model, inputs)
 
     return expo_program