diff --git a/build_tools/build_standalone.sh b/build_tools/build_standalone.sh
index 9b245cbb8..5351eb1a0 100755
--- a/build_tools/build_standalone.sh
+++ b/build_tools/build_standalone.sh
@@ -15,18 +15,14 @@ llvm_project_dir="$project_dir/external/llvm-project"
 build_dir="$project_dir/build"
 
 cmake -GNinja -B"$build_dir" "$llvm_project_dir/llvm" \
-  -DCMAKE_BUILD_TYPE=RelWithDebInfo \
-  -DCMAKE_C_FLAGS_RELWITHDEBINFO="-O2 -DNDEBUG -gline-tables-only" \
-  -DCMAKE_CXX_FLAGS_RELWITHDEBINFO="-O2 -DNDEBUG -gline-tables-only" \
+  -DCMAKE_BUILD_TYPE=Release \
   -DCMAKE_C_COMPILER_LAUNCHER=ccache -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \
   -DLLVM_ENABLE_PROJECTS=mlir \
   -DLLVM_EXTERNAL_PROJECTS=torch-mlir \
   -DLLVM_EXTERNAL_TORCH_MLIR_SOURCE_DIR="$project_dir" \
   -DMLIR_ENABLE_BINDINGS_PYTHON=ON \
+  -DLLVM_ENABLE_ASSERTIONS=ON \
   -DLLVM_TARGETS_TO_BUILD=host
 
-#-DLLVM_ENABLE_ASSERTIONS=ON \
-#
-
 cd "$build_dir"
 ninja tools/torch-mlir/all check-torch-mlir-all
diff --git a/e2e_testing/torchscript/basic.py b/e2e_testing/torchscript/basic.py
index 3f6cc7570..2e11946bd 100644
--- a/e2e_testing/torchscript/basic.py
+++ b/e2e_testing/torchscript/basic.py
@@ -30,12 +30,6 @@ def MmModule_basic(module, tu: TestUtils):
     module.forward(tu.rand(4, 4), tu.rand(4, 4))
 
 
-# TODO: Investigate why RefBackend sometimes can't handle two calls in a row in
-# the trace.
-# It actually works, if MmModule_chained is run by itself, but if other tests
-# are mixed with it, it fails with a mysterious-sounding low level ctypes error
-# that exceeds my current ability to debug.
-#
 @register_test_case(module_factory=lambda: MmModule())
 def MmModule_chained(module, tu: TestUtils):
     res = module.forward(tu.rand(4, 4), tu.rand(4, 4))
diff --git a/e2e_testing/torchscript/xfail_sets.py b/e2e_testing/torchscript/xfail_sets.py
index acb0922a5..978744a30 100644
--- a/e2e_testing/torchscript/xfail_sets.py
+++ b/e2e_testing/torchscript/xfail_sets.py
@@ -18,11 +18,7 @@ _common_torch_mlir_lowering_xfails = {
     'QuantizedMLP_basic',
 }
 
-XFAIL_SETS['refbackend'] = _common_torch_mlir_lowering_xfails | {
-    # The first test in the e2e test batch would fail with SystemError: null
-    # argument to internal routine. Might be some issue with refbackend.
-    'MmModule_basic',
-}
+XFAIL_SETS['refbackend'] = _common_torch_mlir_lowering_xfails
 
 XFAIL_SETS['torchscript'] = {}
 
diff --git a/examples/torchfx/examples/example_add_tanh_sigmoid.py b/examples/torchfx/examples/example_add_tanh_sigmoid.py
new file mode 100644
index 000000000..07cbc13d4
--- /dev/null
+++ b/examples/torchfx/examples/example_add_tanh_sigmoid.py
@@ -0,0 +1,57 @@
+# -*- Python -*-
+# This file is licensed under a pytorch-style license
+# See frontends/pytorch/LICENSE for license information.
+
+# From the torch-mlir root, run with:
+# `python -m examples.torchfx.examples.example_add_tanh_sigmoid`
+# (after setting up python environment with write_env_file.sh)
+
+import torch
+from torch.fx.experimental.fx_acc import acc_tracer
+import torch_mlir
+from torch_mlir.dialects.torch import register_dialect
+from torch_mlir.passmanager import PassManager
+from torch_mlir_e2e_test.linalg_on_tensors_backends.refbackend import RefBackendLinalgOnTensorsBackend
+
+from ..builder import build_module
+from ..annotator import annotate_forward_args
+from ..torch_mlir_types import TorchTensorType
+
+
+class MyModule(torch.nn.Module):
+    def __init__(self):
+        super().__init__()
+
+    def forward(self, x, y):
+        return torch.tanh(x) + torch.sigmoid(y)
+
+
+module = MyModule()
+traced_module = acc_tracer.trace(module, [torch.Tensor(2,2),
+                                          torch.Tensor(2,2)])
+
+print("TRACE")
+arg_type = TorchTensorType(shape=[None, None], dtype=torch.float)
+traced_module = annotate_forward_args(traced_module, [arg_type, arg_type])
+print(traced_module.graph)
+torch_mlir_module = build_module(traced_module)
+
+print("\n\nTORCH MLIR")
+torch_mlir_module.dump()
+print(torch_mlir_module.operation.verify())
+
+with torch_mlir_module.context:
+    pm = PassManager.parse('torchscript-to-linalg-on-tensors-backend-pipeline')
+    pm.run(torch_mlir_module)
+
+print("\n\nLOWERED MLIR")
+torch_mlir_module.dump()
+
+backend = RefBackendLinalgOnTensorsBackend()
+compiled = backend.compile(torch_mlir_module)
+jit_module = backend.load(compiled)
+
+print("\n\nRunning Forward Function")
+t = torch.rand((2, 2), dtype=torch.float)
+print("Compiled result:\n", jit_module.forward(t.numpy(), t.numpy()))
+print("\nExpected result:\n", module.forward(t, t))
diff --git a/external/llvm-project b/external/llvm-project
index 6e60bb688..204d301bb 160000
--- a/external/llvm-project
+++ b/external/llvm-project
@@ -1 +1 @@
-Subproject commit 6e60bb6883178cf14e6fd47a6789495636e4322f
+Subproject commit 204d301bb1921431a853c0bfba32007c018df1d5
diff --git a/lib/Conversion/TorchToLinalg/TorchToLinalg.cpp b/lib/Conversion/TorchToLinalg/TorchToLinalg.cpp
index ae8dad4d4..3b6af675a 100644
--- a/lib/Conversion/TorchToLinalg/TorchToLinalg.cpp
+++ b/lib/Conversion/TorchToLinalg/TorchToLinalg.cpp
@@ -186,8 +186,8 @@ static Value getPaddedTensor(Operation *op, OpBuilder &b, Value &input,
   Type ranked4DTensorType = linalg::PadTensorOp::inferResultType(
       input.getType().cast<RankedTensorType>(), paddingInts, paddingInts);
   Value paddedInput = linalg::PadTensorOp::createPadScalarOp(
-      ranked4DTensorType, input, c0, /*low=*/paddings, /*high=*/paddings, loc,
-      b);
+      ranked4DTensorType, input, c0, /*low=*/paddings, /*high=*/paddings,
+      /*packing=*/false, loc, b);
   return paddedInput;
 }