Add MatMulNBits shape infer to SymbolicShapeInference (#21246)

### Description
Support MatMulNBits shape infer in SymbolicShapeInference

MatMulNBits's B input is rank-2, so implicit merge does not apply.

### Motivation and Context
[Issue with performing shape inference using symbolic_shape_infer.py
with Phi-3 ONNX Models · Issue #21194 · microsoft/onnxruntime
(github.com)](#21194)

onnxruntime/python/tools/symbolic_shape_infer.py

@@ -206,10 +206,9 @@ def __init__(self, int_max, auto_merge, guess_output_rank, verbose, prefix=""):
             "GemmFloat8": self._infer_GemmFloat8,
             "GroupNorm": self._infer_GroupNorm,
             "GroupQueryAttention": self._infer_GroupQueryAttention,
-            "SparseAttention": self._infer_SparseAttention,
-            "SkipGroupNorm": self._infer_SkipGroupNorm,
             "LayerNormalization": self._infer_LayerNormalization,
             "LongformerAttention": self._infer_LongformerAttention,
+            "MatMulNBits": self._infer_MatMulNBits,
             "MultiHeadAttention": self._infer_MultiHeadAttention,
             "NhwcConv": self._infer_NhwcConv,
             "PackedAttention": self._infer_PackedAttention,
@@ -223,8 +222,10 @@ def __init__(self, int_max, auto_merge, guess_output_rank, verbose, prefix=""):
             "RestorePadding": self._infer_RestorePadding,
             "RotaryEmbedding": self._infer_RotaryEmbedding,
             "SimplifiedLayerNormalization": self._infer_LayerNormalization,
+            "SkipGroupNorm": self._infer_SkipGroupNorm,
             "SkipLayerNormalization": self._infer_SkipLayerNormalization,
             "SkipSimplifiedLayerNormalization": self._infer_SkipLayerNormalization,
+            "SparseAttention": self._infer_SparseAttention,
         }
         self.aten_op_dispatcher_ = {
             "embedding": self._infer_Gather,
@@ -1256,6 +1257,25 @@ def _infer_MatMul(self, node):  # noqa: N802
     def _infer_MatMulInteger(self, node):  # noqa: N802
         self._compute_matmul_shape(node, onnx.TensorProto.INT32)
 
+    def _infer_MatMulNBits(self, node):  # noqa: N802
+        lhs_shape = self._get_shape(node, 0)
+        rhs_shape = [get_attribute(node, "K"), get_attribute(node, "N")]
+        lhs_rank = len(lhs_shape)
+        assert lhs_rank > 0
+        if lhs_rank == 1:
+            new_shape = rhs_shape[1:]
+        else:
+            new_shape = lhs_shape[:-1] + rhs_shape[1:]
+        # merge reduce dim
+        self._check_merged_dims(
+            [lhs_shape[-1], rhs_shape[0]],
+            allow_broadcast=False,
+        )
+        # infer output_dtype from input type when not specified
+        output_dtype = self.known_vi_[node.input[0]].type.tensor_type.elem_type
+        vi = self.known_vi_[node.output[0]]
+        vi.CopyFrom(helper.make_tensor_value_info(node.output[0], output_dtype, new_shape))
+
     def _infer_NonMaxSuppression(self, node):  # noqa: N802
         selected = str(self._new_symbolic_dim_from_output(node))
         vi = self.known_vi_[node.output[0]]

onnxruntime/test/python/onnxruntime_test_python_symbolic_shape_infer.py

@@ -594,6 +594,55 @@ def test_dequantize_linear_ms_domain(self):
         ]
         self._check_shapes(graph, inferred.graph, expected_shapes)
 
+    def test_matmulnbits(self):
+        """
+        Test ORT MatMulNBits op.
+        Check that the output shape is propagated from the inputs and that the output data
+        type comes from the first input.
+        """
+        b_np = numpy.random.randint(0, 255, (4, 1, 8), numpy.uint8)
+        b = numpy_helper.from_array(b_np, name="b")
+        scale_np = numpy.random.rand(4).astype(numpy.float32)
+        scale = numpy_helper.from_array(scale_np, name="scale")
+        zero_point_np = numpy.random.randint(0, 255, (4), numpy.uint8)
+        zero_point = numpy_helper.from_array(zero_point_np, name="zero_point")
+
+        initializers = [b, scale, zero_point]
+
+        kwargs = {"K": 10, "N": 4, "block_size": 16}
+
+        nodes = [
+            helper.make_node(
+                "MatMulNBits",
+                inputs=[
+                    "input_f32",
+                    "b",
+                    "scale",
+                    "zero_point",
+                ],
+                outputs=["output_f32"],
+                **kwargs,
+            ),
+        ]
+
+        inputs = [
+            helper.make_tensor_value_info("input_f32", TensorProto.FLOAT, ["x", 2, 3, 10]),
+        ]
+
+        outputs = [
+            helper.make_tensor_value_info("output_f32", TensorProto.UNDEFINED, None),
+        ]
+
+        graph = helper.make_graph(nodes, "MatMulNBits_Test", inputs, outputs, initializers)
+        model = helper.make_model(graph)
+
+        inferred = SymbolicShapeInference.infer_shapes(model, auto_merge=True)
+
+        expected_shapes = [
+            helper.make_tensor_value_info("output_f32", TensorProto.FLOAT, ["x", 2, 3, 4]),
+        ]
+        self._check_shapes(graph, inferred.graph, expected_shapes)
+
 
 class TestSymbolicShapeInferenceForSlice(unittest.TestCase):
     def check_slice_of_concat(self, input_dims, start, end, step, expected_output_dim):