[FRONTEND][ONNX] Make bias input optional in LayerNormalization

This change updates the LayerNormalization converter to support ONNX
models where the optional bias input is not provided.
When bias is missing, an empty bias tensor is generated.

This behavior aligns with the ONNX’s spec for LayerNormalization (opset 17+) where
the bias input is officially optional.

Author: vacu9708

python/tvm/relax/frontend/onnx/onnx_frontend.py

@@ -2489,6 +2489,10 @@ def _impl_v17(cls, bb, inputs, attr, params):
         axis = attr.get("axis", -1)
         epsilon = attr.get("epsilon", 1e-05)
 
+        if bias is None:
+            seq_len = data.struct_info.shape[1].value
+            bias = relax.const([0.0] * seq_len, dtype="float32")
+
         output = relax.op.nn.layer_norm(data, scale, bias, axis, epsilon)
         # Onnx layernorm has 3 outputs but only the first is used.
         # We construct two empty constants for this.

tests/python/relax/test_frontend_onnx.py

@@ -1303,6 +1303,24 @@ def test_layer_norm():
     model = helper.make_model(graph, producer_name="layer_norm_test")
     check_correctness(model)
 
+    # Test case with no bias that is an optional input
+    layer_norm_node = helper.make_node("LayerNormalization", ["a", "b"], ["d"], epsilon=1e-12)
+
+    graph = helper.make_graph(
+        [layer_norm_node],
+        "layer_norm_test",
+        inputs=[
+            helper.make_tensor_value_info("a", TensorProto.FLOAT, [32, 32]),
+            helper.make_tensor_value_info("b", TensorProto.FLOAT, [32]),
+        ],
+        outputs=[
+            helper.make_tensor_value_info("d", TensorProto.FLOAT, [32, 32]),
+        ],
+    )
+
+    model = helper.make_model(graph, producer_name="layer_norm_test")
+    check_correctness(model)
+
 
 # TODO Enable dynamism
 @pytest.mark.parametrize("dynamic", [False])