diff --git a/torch_ttnn/passes/lowering/to_tt_pass.py b/torch_ttnn/passes/lowering/to_tt_pass.py
index d2e3d8503..7e28d9f8d 100644
--- a/torch_ttnn/passes/lowering/to_tt_pass.py
+++ b/torch_ttnn/passes/lowering/to_tt_pass.py
@@ -418,21 +418,19 @@ def lower_binary_eltwise(fn, args):
             if node.target in TTNN_POINTWISE_UNARY_OPS:
                 code = TTNN_POINTWISE_UNARY_OPS[node.target]
 
-            def unsqueeze_to_2d(code, args=args, kwargs=kwargs):
-                ndims = len(node.meta["val"].size())
+            # NOTE(jdh8): Workaround for tenstorrent/tt-metal#12671
+            # Passing a tensor shaped `(N,)` to the kernel results in `(1, N)`.
+            # Reshape the tensor back to get the correct shape.
+            def reshape_1d(code, args=args, kwargs=kwargs):
+                shape = node.meta["val"].size()
                 result = g.call_function(code, args, kwargs)
-
-                if ndims == 1:
-                    result = g.call_function(ttnn.to_layout, (result, TtnnRowMajorLayout()))
-                    result = g.call_function(ttnn.squeeze, (result, 0))
-
-                return result
+                return result if len(shape) > 1 else g.call_function(ttnn.reshape, (result, shape))
 
             if node.target in TTNN_POINTWISE_UNARY_OPS:
-                return unsqueeze_to_2d(TTNN_POINTWISE_UNARY_OPS[node.target])
+                return reshape_1d(TTNN_POINTWISE_UNARY_OPS[node.target])
 
             if node.target == torch.ops.aten.round.default:
-                return unsqueeze_to_2d(ttnn.round, (args[0],), {"decimals": 0})
+                return reshape_1d(ttnn.round, (args[0],), {"decimals": 0})
 
             if node.target == torch.ops.aten.clone.default:
                 arg_metadata = node.meta["val"]