Remove the xfails after Brian's multiple fake mode PR

test/conftest.py

@@ -6,7 +6,9 @@ def x_fail_activation_hooks(request):
     use_activation_hooks = request.getfixturevalue("use_activation_hooks")
     if use_activation_hooks:
         request.node.add_marker(
-            pytest.mark.xfail(reason="use_activation_hooks is not supported for AOT")
+            pytest.mark.xfail(
+                reason="use_activation_hooks is not supported for AOT", strict=True
+            )
         )
 
 
@@ -16,5 +18,7 @@ def x_fail_activation_hooks_with_delayed(request):
     use_activation_hooks = request.getfixturevalue("use_activation_hooks")
     if use_activation_hooks and linear_type == linear_type.DELAYED:
         request.node.add_marker(
-            pytest.mark.xfail(reason="use_activation_hooks is not supported for AOT")
+            pytest.mark.xfail(
+                reason="use_activation_hooks is not supported for AOT", strict=True
+            )
         )

test/test_compile.py

@@ -141,7 +141,6 @@ def forward(self, x):
                 return x_hp
             return x_fp8
 
-    @pytest.mark.xfail(reason="TODO: Fix this test, see TODO in MockLinear")
     def test_float8_with_graph_break_in_the_middle(self):
         """Test that having Float8Tensor object at the boundary of a subgraph"""
         cnts = CompileCounterWithBackend("inductor")
@@ -174,7 +173,6 @@ def to_float(x):
         )
         torch.testing.assert_close(y2_eager, y2_compiled)
 
-    @pytest.mark.xfail(reason="TODO: Fix this test, see TODO in MockLinear")
     def test_float8_graph_output(self):
         """Test that having Float8Tensor object as a graph output works"""
         cnts = CompileCounterWithBackend("inductor")