Pattern Optimizer

The pattern optimizer is implemented by class GraphBuilderPatternOptimization. It searches for a specific sequence of nodes in the graph and replaces it by another one without changing the inputs or the long_outputs of the graph. The goal of the optimizer is to make the whole computation graph more efficient. The goal of this implementation is to make this optimization as fast as possible. Assuming the nodes in an onnx graph are ordered in a way every input of a node was created by previous nodes, the optimizer must not require any global reordering. The cost should be in O(N P I) in the worst case where N is the number of nodes, P is the number of patterns, I is the number of iterations.

It is difficult to foresee what a pattern needs in order to rewrite a part of the graph. This API tries to give as much freedom as it can without leaving too much to do to the developper which tries to add a new pattern.

Patterns

Patterns must inherit from PatternOptimization. This class defines two methods.

PatternOptimization.match

def match(
    self,
    g: "GraphBuilderPatternOptimization",
    node: NodeProto,
    matched: List[MatchResult],
) -> Optional[MatchResult]:
  • g is a GraphBuilderPatternOptimization, it holds all the existing nodes, is able to return any information about type, shape, the node before, the node after another one.

  • node: the matching must determine if some nodes around this one are part of set of nodes this pattern optimizer can rewrite. From there, the function explores wherever it needs, checking any condition it needs.

  • matched: usually unused, it returns of nodes already matching a pattern

The method must not modify the graph. The method returns None if no match is found or an instance of class MatchResult. It must contain:

  • a list of nodes involved in the rewriting. It does not mean all of them will be removed but all of them are needed to do the rewriting and must not be impacted by other pattern optimizer.

  • A function doing the rewriting (usually method apply of the pattern class).

  • An existing node where the rewritten nodes can be inserted. Knowing it makes it faster to rewriter. If not specified, the optimizer will automatically determine the position of the new nodes.

Debugging: method none

def none(
    self,
    node: Optional[NodeProto] = None,
    lineno: Optional[int] = None,
    msg: str = "",
):

It may be useful which reason made a pattern matching fail. Instead of returning None, method match can return the following expression:

return self.none(node, inspect.currentframe().f_lineno)

By setting the verbosity (see next Section), the user may then know which lines in the code returned None and which condition failed.

PatternOptimization.apply

@classmethod
def apply(
    cls, g: "GraphBuilder", *nodes: Sequence[NodeProto]
) -> List[NodeProto]:

The method does the rewriting. It assumes it can happen. It takes a list of nodes impacted by the rewriting. It assumes no other pattern optimizer modified them or will modify them. It receives the list of nodes returned by method apply. Since it is a list of argument, method match can include None values. The method returns the new nodes. The optimizer considers that any node given to this function is removed from the graph, and any node returned by it are added. If a received node must be kept, it must be added to the list of returned node.

Optimization Algorithm

It is implemented in method optimize

def optimize(
    self, max_iter=-1, remove_identity: bool = True
) -> List[Dict[str, Any]]:

The algorithm runs multiple iteration until the graph is not evolving or max_iter is reached. By default, it is equal to the number of nodes. An iteration is:

matches = []

builds all successors and predecessors

# Step 1: match

for all patterns P:

    for all nodes n:

        r = p.match(n)
        if r:
            if no node already scheduled to be rewritten by another match:
                matches.append(r)

# Step 2: apply

for all matches r:
    apply the match r

# Step 3: clean

remove unused nodes
remove identity nodes

This algorithm may apply more than one rewriting at each iteration but it guarantees the local structure when applying the rewriting was not altered by another one.

Adding a pattern

See #80 about the addition of a new pattern.

Example

Simple API

We consider the following simple model:

<<<

import torch
from experimental_experiment.helpers import pretty_onnx
from experimental_experiment.xbuilder import OptimizationOptions
from experimental_experiment.torch_interpreter import to_onnx


class MLP(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.layers = torch.nn.Sequential(
            torch.nn.Linear(10, 32),
            torch.nn.ReLU(),
            torch.nn.Linear(32, 1),
        )

    def forward(self, x):
        return self.layers(x)


x = torch.rand(3, 10)
onx = to_onnx(
    MLP(), (x,), input_names=["x"], options=OptimizationOptions(patterns=None)
)
with open("temp_doc_mlp.onnx", "wb") as f:
    f.write(onx.SerializeToString())
print(pretty_onnx(onx))

>>>

    opset: domain='' version=18
    input: name='x' type=dtype('float32') shape=[3, 10]
    init: name='p_layers_0_weight::T10' type=float32 shape=(10, 32)       -- GraphBuilder.constant_folding.from/fold(p_layers_0_weight)##p_layers_0_weight/DynamoInterpret.placeholder.1/P(layers.0.weight)
    init: name='p_layers_2_weight::T10' type=float32 shape=(32, 1)        -- GraphBuilder.constant_folding.from/fold(p_layers_2_weight)##p_layers_2_weight/DynamoInterpret.placeholder.1/P(layers.2.weight)
    init: name='layers.0.bias' type=float32 shape=(32,)                   -- DynamoInterpret.placeholder.1/P(layers.0.bias)
    init: name='layers.2.bias' type=float32 shape=(1,) -- array([0.082], dtype=float32)-- DynamoInterpret.placeholder.1/P(layers.2.bias)
    MatMul(x, p_layers_0_weight::T10) -> _onx_matmul_x
      Add(_onx_matmul_x, layers.0.bias) -> linear
        Relu(linear) -> relu
          MatMul(relu, p_layers_2_weight::T10) -> _onx_matmul_relu
            Add(_onx_matmul_relu, layers.2.bias) -> output_0
    output: name='output_0' type=dtype('float32') shape=[3, 1]

Which we can renders as follows:

digraph{
  nodesep=0.05;
  orientation=portrait;
  size=7;
  ranksep=0.25;

  x [shape=box color=red label="x\nTensorProto.FLOAT\nshape=[3, 10]" fontsize=10];

  output_0 [shape=box color=green label="output_0\nTensorProto.FLOAT\nshape=[3, 1]" fontsize=10];

  p_layers_0_weight____T10 [shape=box label="p_layers_0_weight____T10\nfloat32((10, 32))\n[[ 0.307  0.265 -0.067 -0.04   0.15   0.051 -0.138..." fontsize=10];
  p_layers_2_weight____T10 [shape=box label="p_layers_2_weight____T10\nfloat32((32, 1))\n[[ 0.108]\n [-0.152]\n [ 0.122]\n [-0.128]\n [ 0.144]\n..." fontsize=10];
  layers_0_bias [shape=box label="layers_0_bias\nfloat32((32,))\n[ 0.06  -0.024 -0.302  0.072 -0.292  0.083 -0.159 ..." fontsize=10];
  layers_2_bias [shape=box label="layers_2_bias\nfloat32((1,))\n[0.082]" fontsize=10];

  _onx_matmul_x [shape=box label="_onx_matmul_x" fontsize=10];
  Opset [shape=box style="filled,rounded" color=orange label="MatMul" fontsize=10];
  x -> Opset;
  p_layers_0_weight____T10 -> Opset;
  Opset -> _onx_matmul_x;

  linear [shape=box label="linear" fontsize=10];
  Opset2 [shape=box style="filled,rounded" color=orange label="Add" fontsize=10];
  _onx_matmul_x -> Opset2;
  layers_0_bias -> Opset2;
  Opset2 -> linear;

  relu [shape=box label="relu" fontsize=10];
  relu [shape=box style="filled,rounded" color=orange label="Relu" fontsize=10];
  linear -> relu;
  relu -> relu;

  _onx_matmul_relu [shape=box label="_onx_matmul_relu" fontsize=10];
  Opset3 [shape=box style="filled,rounded" color=orange label="MatMul" fontsize=10];
  relu -> Opset3;
  p_layers_2_weight____T10 -> Opset3;
  Opset3 -> _onx_matmul_relu;

  Opset4 [shape=box style="filled,rounded" color=orange label="Add" fontsize=10];
  _onx_matmul_relu -> Opset4;
  layers_2_bias -> Opset4;
  Opset4 -> output_0;
}

We then apply the optimizations by writing the following code:

<<<

import onnx
from experimental_experiment.helpers import pretty_onnx
from experimental_experiment.xbuilder import GraphBuilder

onx = onnx.load("temp_doc_mlp.onnx")

# The model is placed in a GraphBuilder.
# It creates dictionnaires to store shapes, ranks, types
# to make it easier to the optimizers to find the information
# they need. It still uses NodeProto to store nodes
gr = GraphBuilder(onx, infer_shapes_options=True)

# Let's optimize.
opt_onx = gr.to_onnx(optimize=True)
with open("temp_doc_mlp_opt.onnx", "wb") as f:
    f.write(opt_onx.SerializeToString())
print(pretty_onnx(opt_onx))

>>>

    opset: domain='' version=18
    input: name='x' type=dtype('float32') shape=[3, 10]
    init: name='layers.0.bias' type=float32 shape=(32,)                   -- DynamoInterpret.placeholder.1/P(layers.0.bias)GraphBuilder._update_structures_with_proto.1/from(layers.0.bias)
    init: name='layers.2.bias' type=float32 shape=(1,) -- array([0.082], dtype=float32)-- DynamoInterpret.placeholder.1/P(layers.2.bias)GraphBuilder._update_structures_with_proto.1/from(layers.2.bias)
    init: name='GemmTransposePattern--p_layers_0_weight::T10' type=float32 shape=(32, 10)-- GraphBuilder.constant_folding.from/fold(p_layers_0_weight::T10)##p_layers_0_weight::T10/GraphBuilder._update_structures_with_proto.1/from(p_layers_0_weight::T10)
    init: name='GemmTransposePattern--p_layers_2_weight::T10' type=float32 shape=(1, 32)-- GraphBuilder.constant_folding.from/fold(init7_s2_1_-1,p_layers_2_weight::T10)##p_layers_2_weight::T10/GraphBuilder._update_structures_with_proto.1/from(p_layers_2_weight::T10)##init7_s2_1_-1/TransposeEqualReshapePattern.apply.new_shape
    Gemm(x, GemmTransposePattern--p_layers_0_weight::T10, layers.0.bias, transB=1) -> linear
      Relu(linear) -> relu
        Gemm(relu, GemmTransposePattern--p_layers_2_weight::T10, layers.2.bias, transB=1) -> output_0
    output: name='output_0' type=dtype('float32') shape=[3, 1]

Which renders as follows:

digraph{
  nodesep=0.05;
  orientation=portrait;
  size=7;
  ranksep=0.25;

  x [shape=box color=red label="x\nTensorProto.FLOAT\nshape=[3, 10]" fontsize=10];

  output_0 [shape=box color=green label="output_0\nTensorProto.FLOAT\nshape=[3, 1]" fontsize=10];

  layers_0_bias [shape=box label="layers_0_bias\nfloat32((32,))\n[ 0.06  -0.024 -0.302  0.072 -0.292  0.083 -0.159 ..." fontsize=10];
  layers_2_bias [shape=box label="layers_2_bias\nfloat32((1,))\n[0.082]" fontsize=10];
  GemmTransposePattern__p_layers_0_weight____T10 [shape=box label="GemmTransposePattern__p_layers_0_weight____T10\nfloat32((32, 10))\n[[ 0.307  0.167  0.174 -0.177  0.3    0.262  0.063..." fontsize=10];
  GemmTransposePattern__p_layers_2_weight____T10 [shape=box label="GemmTransposePattern__p_layers_2_weight____T10\nfloat32((1, 32))\n[[ 0.108 -0.152  0.122 -0.128  0.144  0.083  0.059..." fontsize=10];

  linear [shape=box label="linear" fontsize=10];
  GemmTransposePattern__MatMulAddPattern__Opset2 [shape=box style="filled,rounded" color=orange label="Gemm\ntransB=1" fontsize=10];
  x -> GemmTransposePattern__MatMulAddPattern__Opset2;
  GemmTransposePattern__p_layers_0_weight____T10 -> GemmTransposePattern__MatMulAddPattern__Opset2;
  layers_0_bias -> GemmTransposePattern__MatMulAddPattern__Opset2;
  GemmTransposePattern__MatMulAddPattern__Opset2 -> linear;

  relu [shape=box label="relu" fontsize=10];
  relu [shape=box style="filled,rounded" color=orange label="Relu" fontsize=10];
  linear -> relu;
  relu -> relu;

  GemmTransposePattern__MatMulAddPattern__Opset32 [shape=box style="filled,rounded" color=orange label="Gemm\ntransB=1" fontsize=10];
  relu -> GemmTransposePattern__MatMulAddPattern__Opset32;
  GemmTransposePattern__p_layers_2_weight____T10 -> GemmTransposePattern__MatMulAddPattern__Opset32;
  layers_2_bias -> GemmTransposePattern__MatMulAddPattern__Opset32;
  GemmTransposePattern__MatMulAddPattern__Opset32 -> output_0;
}

Verbosity

<<<

import onnx
from experimental_experiment.xbuilder import GraphBuilder

onx = onnx.load("temp_doc_mlp.onnx")

gr = GraphBuilder(onx, infer_shapes_options=True, verbose=1)
opt_onx = gr.to_onnx(optimize=True)

>>>

    [GraphBuilder-XGA._add_shape_information] dynamic shapes replacements={}
    [GraphBuilder-XGA.optimize] start with 5 nodes
    [GraphBuilder-XGA.optimize] #patterns=72
    [GraphBuilder-XGA.optimize] start with subgraphs
    [GraphBuilder-XGA.optimize] done with subgraphs
    [GraphBuilderPatternOptimization-XGA.optimize] start with 5 nodes, 4 initializers, 72 patterns, priorities=[0, 1, 3], max_iter=30
    [GraphBuilderPatternOptimization-XGA.optimize] iteration 0: 5 nodes, priority=0
    [GraphBuilderPatternOptimization-XGA.optimize] increase priority to 1
    [GraphBuilderPatternOptimization-XGA.optimize] iteration 1: 5 nodes, priority=1
    [GraphBuilderPatternOptimization-XGA.optimize] increase priority to 3
    [GraphBuilderPatternOptimization-XGA.optimize] iteration 2: 5 nodes, priority=3
    [GraphBuilderPatternOptimization-XGA.optimize] applies 2 matches, 2*MatMulAddPattern - time=0.001 | max_time=BatchNormalizationTrainingPattern:0.000
    [GraphBuilderPatternOptimization-XGA.optimize] iteration 3: 3 nodes, priority=3
    [GraphBuilderPatternOptimization-XGA.optimize] applies 2 matches, 2*GemmTransposePattern - time=0.000 | max_time=GemmTransposePattern:0.000
    [GraphBuilderPatternOptimization-XGA.optimize] iteration 4: 5 nodes, priority=3
    [GraphBuilderPatternOptimization-XGA.optimize] applies 1 matches, [0]=MatchResult: TransposeEqualReshapePattern replaces ['Transpose'] - time=0.000 | max_time=TransposeMatMulPattern:0.000
    [GraphBuilderPatternOptimization-XGA.optimize] iteration 5: 5 nodes, priority=3
    [GraphBuilderPatternOptimization-XGA.optimize] stops current_priority_index=3, priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-XGA.optimize] done after 6 iterations with 5 nodes in 0.008
    [GraphBuilder-XGA.optimize] done with 3 nodes in 0.010
    [GraphBuilder-XGA.to_onnx] make_model 4 inits 0 params
    [GraphBuilder-XGA.time_evaluation_constants_] 0
    [GraphBuilder-XGA._build_initializers] start with 4 initializers, large_model=False, external_threshold=1024
    [GraphBuilder-XGA._build_initializers] switch low/high order
    [GraphBuilder-XGA._build_initializers] done in 1.2380005500745028e-06s with 4 initializers, 0 large initializers
    [GraphBuilder-XGA._add_shape_information] dynamic shapes replacements={}

With more verbosity:

<<<

import onnx
from experimental_experiment.xbuilder import GraphBuilder

onx = onnx.load("temp_doc_mlp.onnx")

gr = GraphBuilder(onx, infer_shapes_options=True, verbose=11)
opt_onx = gr.to_onnx(optimize=True)

>>>

    [GraphBuilder-VNK._update_structures_with_proto] -- starts with 5 nodes
    [GraphBuilder-VNK.set_shape] p_layers_0_weight::T10:(10, 32)
    [GraphBuilder-VNK.set_rank] p_layers_0_weight::T10:2
    [GraphBuilder-VNK.set_type] p_layers_0_weight::T10:1
    [GraphBuilder-VNK.make_initializer] p_layers_0_weight::T10[1:(10, 32)]
    [GraphBuilder-VNK.update_node_constant] new constant 'p_layers_0_weight::T10', node=None
    [GraphBuilder-VNK.set_shape] p_layers_2_weight::T10:(32, 1)
    [GraphBuilder-VNK.set_rank] p_layers_2_weight::T10:2
    [GraphBuilder-VNK.set_type] p_layers_2_weight::T10:1
    [GraphBuilder-VNK.make_initializer] p_layers_2_weight::T10[1:(32, 1)]
    [GraphBuilder-VNK.update_node_constant] new constant 'p_layers_2_weight::T10', node=None
    [GraphBuilder-VNK.set_shape] layers.0.bias:(32,)
    [GraphBuilder-VNK.set_rank] layers.0.bias:1
    [GraphBuilder-VNK.set_type] layers.0.bias:1
    [GraphBuilder-VNK.make_initializer] layers.0.bias[1:(32,)]
    [GraphBuilder-VNK.update_node_constant] new constant 'layers.0.bias', node=None
    [GraphBuilder-VNK.set_shape] layers.2.bias:(1,)
    [GraphBuilder-VNK.set_rank] layers.2.bias:1
    [GraphBuilder-VNK.set_type] layers.2.bias:1
    [GraphBuilder-VNK.make_initializer] layers.2.bias[1:(1,)]
    [GraphBuilder-VNK.update_node_constant] new constant 'layers.2.bias', node=None
    [GraphBuilder-VNK.set_type] x:1
    [GraphBuilder-VNK.set_shape] x:(3, 10)
    [GraphBuilder-VNK.set_rank] x:2
    [GraphBuilder-VNK.set_type] output_0:1
    [GraphBuilder-VNK.set_shape] output_0:(3, 1)
    [GraphBuilder-VNK.set_rank] output_0:2
    [GraphBuilder-VNK.set_type] _onx_matmul_x:1
    [GraphBuilder-VNK.set_shape] _onx_matmul_x:(3, 32)
    [GraphBuilder-VNK.set_rank] _onx_matmul_x:2
    [GraphBuilder-VNK.set_type] linear:1
    [GraphBuilder-VNK.set_shape] linear:(3, 32)
    [GraphBuilder-VNK.set_rank] linear:2
    [GraphBuilder-VNK.set_type] relu:1
    [GraphBuilder-VNK.set_shape] relu:(3, 32)
    [GraphBuilder-VNK.set_rank] relu:2
    [GraphBuilder-VNK.set_type] _onx_matmul_relu:1
    [GraphBuilder-VNK.set_shape] _onx_matmul_relu:(3, 1)
    [GraphBuilder-VNK.set_rank] _onx_matmul_relu:2
    [GraphBuilder-VNK.set_type] output_0:1
    [GraphBuilder-VNK._update_structures_with_proto] ends with 5 nodes in 0.0008961910007201368
    [GraphBuilder-VNK.constant_folding] -- starts with 4 constants and 5 nodes.
    [GraphBuilder-VNK.constant_folding] cst:: 1 :: layers.0.bias
    [GraphBuilder-VNK.constant_folding] cst:: . :: _onx_matmul_x
    [GraphBuilder-VNK.constant_folding] cst:: . :: x
    [GraphBuilder-VNK.constant_folding] cst:: . :: relu
    [GraphBuilder-VNK.constant_folding] cst:: . :: output_0
    [GraphBuilder-VNK.constant_folding] cst:: 1 :: layers.2.bias
    [GraphBuilder-VNK.constant_folding] cst:: . :: linear
    [GraphBuilder-VNK.constant_folding] cst:: . :: _onx_matmul_relu
    [GraphBuilder-VNK.constant_folding] cst:: 1 :: p_layers_2_weight::T10
    [GraphBuilder-VNK.constant_folding] cst:: 1 :: p_layers_0_weight::T10
    [GraphBuilder-VNK.constant_folding] initializer: p_layers_0_weight::T10
    [GraphBuilder-VNK.constant_folding] initializer: p_layers_2_weight::T10
    [GraphBuilder-VNK.constant_folding] initializer: layers.0.bias
    [GraphBuilder-VNK.constant_folding] initializer: layers.2.bias
    [GraphBuilder-VNK.constant_folding] ends with 4 constants and 5 nodes in 3.7802999941050075e-05 seconds
    [GraphBuilder-VNK._update_shape_types_with_proto] -- starts with 5 nodes and 0 shapes.
    [GraphBuilder._update_shape_types_with_proto] infer shapes
    [GraphBuilder._update_shape_types_with_proto] infer shapes done 0.00022670500038657337 seconds
    [GraphBuilder._update_shape_types_with_proto] _clean_shapes after 0.0002520689995435532 seconds
    [GraphBuilder-VNK._update_shape_types_with_proto] walk through 0 shapes.
    [GraphBuilder-VNK.set_type] _onx_matmul_x:1
    [_update_shape_types_with_proto_one_result] update shape(_onx_matmul_x) with (3, 32)
    [GraphBuilder-VNK.set_type] linear:1
    [_update_shape_types_with_proto_one_result] update shape(linear) with (3, 32)
    [GraphBuilder-VNK.set_type] relu:1
    [_update_shape_types_with_proto_one_result] update shape(relu) with (3, 32)
    [GraphBuilder-VNK.set_type] _onx_matmul_relu:1
    [_update_shape_types_with_proto_one_result] update shape(_onx_matmul_relu) with (3, 1)
    [GraphBuilder-VNK._update_shape_types_with_proto] ends in 7.657599962840322e-05 seconds.
    [GraphBuilder-VNK._add_shape_information] dynamic shapes replacements={}
    [GraphBuilder-VNK.optimize] start with 5 nodes
    [GraphBuilder-VNK.optimize] options=OptimizationOptions(constant_folding={'Concat', 'Add', 'Sub', 'Transpose', 'Div', 'Cast', 'Mul', 'Reshape'}, patterns=[BatchNormalizationPattern(), BatchNormalizationTrainingPattern(), CastLayerNormalizationCastPattern(), CastPattern(), CastCastBinaryPattern(), CastOpCastPattern(), ClipClipPattern(), ComputationCastOpCastPattern(), ConcatEmptyPattern(), ConcatGatherPattern(), ConcatReshapePattern(), ConcatTwiceUnaryPattern(), ConvBiasNullPattern(), DropoutPattern(), ExpandPattern(), ExpandBroadcastPattern(), ExpandSwapPattern(), GeluPattern(), IdentityPattern(), LayerNormalizationPattern(), LayerNormalizationScalePattern(), LeakyReluPattern(), MulMulMulScalarPattern(), ReduceReshapePattern(), ReduceSumNormalizePattern(), ReshapePattern(), ReshapeMatMulReshapePattern(), Reshape2Of3Pattern(), ReshapeReshapeBinaryPattern(), MatMulAddPattern(), GemmTransposePattern(), MatMulReshape2Of3Pattern(), MulMulMatMulPattern(), ShapeBasedReshapeIsSqueezePattern(), ShapeBasedStaticExpandPattern(), ShapeBasedConcatExpandPattern(), ShapeBasedEditDistanceReshapePattern(), ShapeBasedIdentityPattern(), ShapeBasedExpandBroadcastPattern(), ShapeBasedExpandBroadcastMatMulPattern(), ShapeBasedExpandCastWhereSwapPattern(), ShapeBasedExpandSwapPattern(), ShapeBasedMatMulToMulPattern(), ShapeBasedSameChildrenPattern(), ShapeBasedShapeShapeAddPattern(), ReshapeReshapePattern(), RotaryEmbeddingPattern(), SameChildrenPattern(), SequenceConstructAtPattern(), SliceSlicePattern(), SlicesSplitPattern(), SoftmaxCrossEntropyLossCastPattern(), SplitConcatPattern(), SqueezeAddPattern(), SqueezeUnsqueezePattern(), StaticConcatReshapePattern(), Sub1MulPattern(), SwitchOrderBinaryPattern(), SwitchReshapeActivationPattern(), TransposeEqualReshapePattern(), TransposeMatMulPattern(), TransposeReshapeMatMulPattern(), TransposeReshapeTransposePattern(), TransposeTransposePattern(), UnsqueezeEqualPattern(), UnsqueezeUnsqueezePattern(), RotaryConcatPartPattern(), FunctionCausalMaskPattern(), FunctionCausalMaskMulAddPattern(), FunctionCosSinCachePattern(), FunctionHalfRotaryEmbeddingPattern(), RMSNormalizationPattern()], verbose=11)
    -- GRAPH BEFORE OPTIMIZATON --
    
    opset: : 18
    init: p_layers_0_weight::T10: ?: ?                                     -- GraphBuilder._update_structures_with_proto.1/from(p_layers_0_weight::T10)
    init: p_layers_2_weight::T10: ?: ?                                     -- GraphBuilder._update_structures_with_proto.1/from(p_layers_2_weight::T10)
    init: layers.0.bias: ?: ?                                              -- GraphBuilder._update_structures_with_proto.1/from(layers.0.bias)
    init: layers.2.bias: ?: ?                                              -- GraphBuilder._update_structures_with_proto.1/from(layers.2.bias)
    input:: x                                                                       |T1: 3 x 10
    MatMul: x, p_layers_0_weight::T10 -> _onx_matmul_x                              |T1: 3 x 32                   - Opset
    Add: _onx_matmul_x, layers.0.bias -> linear                                     |T1: 3 x 32                   - Opset2
    Relu: linear -> relu                                                            |T1: 3 x 32                   - relu
    MatMul: relu, p_layers_2_weight::T10 -> _onx_matmul_relu                        |T1: 3 x 1                    - Opset3
    Add: _onx_matmul_relu, layers.2.bias -> output_0                                |T1: 3 x 1                    - Opset4
    output:: output_0                                                               |T1: 3 x 1
    -- END --
    [GraphBuilder-VNK.optimize] start with subgraphs
    [GraphBuilder-VNK.optimize] done with subgraphs
    [GraphBuilder-VNK.remove_identity_nodes] -- starts with 5
    [GraphBuilder-VNK.remove_identity_nodes] found 0 replacements
    [GraphBuilder-VNK.remove_identity_nodes] kept 5 nodes
    [GraphBuilder-VNK.remove_identity_nodes] ends with 5 nodes in 2.9308001103345305e-05 seconds
    [GraphBuilder-VNK.constant_folding] -- starts with 4 constants and 5 nodes.
    [GraphBuilder-VNK.constant_folding] cst:: 1 :: layers.0.bias
    [GraphBuilder-VNK.constant_folding] cst:: . :: _onx_matmul_x
    [GraphBuilder-VNK.constant_folding] cst:: . :: x
    [GraphBuilder-VNK.constant_folding] cst:: . :: relu
    [GraphBuilder-VNK.constant_folding] cst:: . :: output_0
    [GraphBuilder-VNK.constant_folding] cst:: 1 :: layers.2.bias
    [GraphBuilder-VNK.constant_folding] cst:: . :: linear
    [GraphBuilder-VNK.constant_folding] cst:: . :: _onx_matmul_relu
    [GraphBuilder-VNK.constant_folding] cst:: 1 :: p_layers_2_weight::T10
    [GraphBuilder-VNK.constant_folding] cst:: 1 :: p_layers_0_weight::T10
    [GraphBuilder-VNK.constant_folding] initializer: p_layers_0_weight::T10
    [GraphBuilder-VNK.constant_folding] initializer: p_layers_2_weight::T10
    [GraphBuilder-VNK.constant_folding] initializer: layers.0.bias
    [GraphBuilder-VNK.constant_folding] initializer: layers.2.bias
    [GraphBuilder-VNK.constant_folding] ends with 4 constants and 5 nodes in 3.101299989793915e-05 seconds
    [GraphBuilderPatternOptimization-VNK.optimize] start with 5 nodes, 4 initializers, 72 patterns, priorities=[0, 1, 3], max_iter=30
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern   1/72 - P0 - BatchNormalizationPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern   2/72 - P0 - BatchNormalizationTrainingPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern   3/72 - P0 - CastPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern   4/72 - P0 - ConcatGatherPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern   5/72 - P0 - ConcatReshapePattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern   6/72 - P0 - ConvBiasNullPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern   7/72 - P0 - ExpandPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern   8/72 - P0 - GeluPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern   9/72 - P0 - IdentityPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  10/72 - P0 - LeakyReluPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  11/72 - P0 - ReshapePattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  12/72 - P0 - ReshapeReshapePattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  13/72 - P0 - SameChildrenPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  14/72 - P0 - ShapeBasedEditDistanceReshapePattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  15/72 - P0 - ShapeBasedIdentityPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  16/72 - P0 - ShapeBasedReshapeIsSqueezePattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  17/72 - P0 - ShapeBasedSameChildrenPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  18/72 - P0 - ShapeBasedShapeShapeAddPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  19/72 - P0 - ShapeBasedStaticExpandPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  20/72 - P0 - SoftmaxCrossEntropyLossCastPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  21/72 - P0 - SqueezeAddPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  22/72 - P0 - SqueezeUnsqueezePattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  23/72 - P0 - StaticConcatReshapePattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  24/72 - P0 - TransposeReshapeTransposePattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  25/72 - P0 - TransposeTransposePattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  26/72 - P0 - UnsqueezeUnsqueezePattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  27/72 - P1 - CastCastBinaryPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  28/72 - P1 - CastLayerNormalizationCastPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  29/72 - P1 - CastOpCastPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  30/72 - P1 - ClipClipPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  31/72 - P1 - ComputationCastOpCastPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  32/72 - P1 - ConcatEmptyPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  33/72 - P1 - ConcatTwiceUnaryPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  34/72 - P1 - DropoutPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  35/72 - P1 - ExpandBroadcastPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  36/72 - P1 - ExpandSwapPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  37/72 - P1 - FunctionCausalMaskMulAddPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  38/72 - P1 - FunctionCausalMaskPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  39/72 - P1 - FunctionCosSinCachePattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  40/72 - P1 - FunctionHalfRotaryEmbeddingPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  41/72 - P1 - GemmTransposePattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  42/72 - P1 - LayerNormalizationPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  43/72 - P1 - LayerNormalizationScalePattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  44/72 - P1 - MatMulReshape2Of3Pattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  45/72 - P1 - MulMulMatMulPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  46/72 - P1 - MulMulMulScalarPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  47/72 - P1 - RMSNormalizationPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  48/72 - P1 - ReduceReshapePattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  49/72 - P1 - ReduceSumNormalizePattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  50/72 - P1 - Reshape2Of3Pattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  51/72 - P1 - ReshapeMatMulReshapePattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  52/72 - P1 - ReshapeReshapeBinaryPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  53/72 - P1 - RotaryConcatPartPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  54/72 - P1 - RotaryEmbeddingPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  55/72 - P1 - SequenceConstructAtPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  56/72 - P1 - ShapeBasedConcatExpandPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  57/72 - P1 - ShapeBasedExpandBroadcastMatMulPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  58/72 - P1 - ShapeBasedExpandBroadcastPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  59/72 - P1 - ShapeBasedExpandCastWhereSwapPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  60/72 - P1 - ShapeBasedExpandSwapPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  61/72 - P1 - ShapeBasedMatMulToMulPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  62/72 - P1 - SliceSlicePattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  63/72 - P1 - SlicesSplitPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  64/72 - P1 - SplitConcatPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  65/72 - P1 - Sub1MulPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  66/72 - P1 - SwitchOrderBinaryPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  67/72 - P1 - SwitchReshapeActivationPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  68/72 - P1 - TransposeEqualReshapePattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  69/72 - P1 - TransposeMatMulPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  70/72 - P1 - TransposeReshapeMatMulPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  71/72 - P1 - UnsqueezeEqualPattern()
    [GraphBuilderPatternOptimization-VNK.optimize] use pattern  72/72 - P3 - MatMulAddPattern()
    -- optimize starts with...
    
    opset: : 18
    init: p_layers_0_weight::T10: ?: ?                                     -- GraphBuilder._update_structures_with_proto.1/from(p_layers_0_weight::T10)
    init: p_layers_2_weight::T10: ?: ?                                     -- GraphBuilder._update_structures_with_proto.1/from(p_layers_2_weight::T10)
    init: layers.0.bias: ?: ?                                              -- GraphBuilder._update_structures_with_proto.1/from(layers.0.bias)
    init: layers.2.bias: ?: ?                                              -- GraphBuilder._update_structures_with_proto.1/from(layers.2.bias)
    input:: x                                                                       |T1: 3 x 10
    MatMul: x, p_layers_0_weight::T10 -> _onx_matmul_x                              |T1: 3 x 32                   - Opset
    Add: _onx_matmul_x, layers.0.bias -> linear                                     |T1: 3 x 32                   - Opset2
    Relu: linear -> relu                                                            |T1: 3 x 32                   - relu
    MatMul: relu, p_layers_2_weight::T10 -> _onx_matmul_relu                        |T1: 3 x 1                    - Opset3
    Add: _onx_matmul_relu, layers.2.bias -> output_0                                |T1: 3 x 1                    - Opset4
    output:: output_0                                                               |T1: 3 x 1
    -- starts optimization
    [GraphBuilderPatternOptimization-VNK.optimize] iteration 0: 5 nodes, priority=0
    [PatternOptimization.enumerate_matches] start BatchNormalizationPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start BatchNormalizationTrainingPattern with main_opset=18 and min_opset=1
    [GraphBuilderPatternOptimization-VNK.optimize] skips CastLayerNormalizationCastPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [PatternOptimization.enumerate_matches] start CastPattern with main_opset=18 and min_opset=1
    [GraphBuilderPatternOptimization-VNK.optimize] skips CastCastBinaryPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips CastOpCastPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips ClipClipPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips ComputationCastOpCastPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips ConcatEmptyPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [PatternOptimization.enumerate_matches] start ConcatGatherPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConcatReshapePattern with main_opset=18 and min_opset=1
    [GraphBuilderPatternOptimization-VNK.optimize] skips ConcatTwiceUnaryPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [PatternOptimization.enumerate_matches] start ConvBiasNullPattern with main_opset=18 and min_opset=1
    [GraphBuilderPatternOptimization-VNK.optimize] skips DropoutPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [PatternOptimization.enumerate_matches] start ExpandPattern with main_opset=18 and min_opset=1
    [GraphBuilderPatternOptimization-VNK.optimize] skips ExpandBroadcastPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips ExpandSwapPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [PatternOptimization.enumerate_matches] start GeluPattern with main_opset=18 and min_opset=20
    [PatternOptimization.enumerate_matches] start IdentityPattern with main_opset=18 and min_opset=1
    [IdentityPattern.match] NONE - line: 297:experimental_experiment.xoptim.patterns.onnx_any, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [IdentityPattern.match] NONE - line: 339:experimental_experiment.xoptim.patterns.onnx_any, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [GraphBuilderPatternOptimization-VNK.optimize] skips LayerNormalizationPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips LayerNormalizationScalePattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [PatternOptimization.enumerate_matches] start LeakyReluPattern with main_opset=18 and min_opset=6
    [GraphBuilder-KQQ.make_tensor_input] x[0:None] -- marker=_build_pattern1_x
    [GraphBuilder-KQQ.set_type] x:0
    [GraphBuilder-KQQ.set_type] x:-1
    [GraphBuilder-KQQ.make_tensor_input] zero[0:None] -- marker=_build_pattern1_zero
    [GraphBuilder-KQQ.set_type] zero:0
    [GraphBuilder-KQQ.set_type] zero:-1
    [GraphBuilder-KQQ.make_tensor_input] slope[0:None] -- marker=_build_pattern1_slope
    [GraphBuilder-KQQ.set_type] slope:0
    [GraphBuilder-KQQ.set_type] slope:-1
    [GraphBuilder-KQQ.make_node] [TT:-] Greater: ['x', 'zero']->['_onx_greater_x']
    [GraphBuilder-KQQ.set_type] _onx_greater_x:9
    [GraphBuilder-KQQ.make_node] [TT:-] Mul: ['x', 'slope']->['_onx_mul_x']
    [GraphBuilder-KQQ.set_type] _onx_mul_x:-1
    [GraphBuilder-KQQ.make_node] [TTT:-] Where: ['_onx_greater_x', 'x', '_onx_mul_x']->['_onx_where_greater_x']
    [GraphBuilder-KQQ.set_type] _onx_where_greater_x:-1
    [GraphBuilder-KQQ.make_tensor_output] _onx_where_greater_x[0: None]
    [GraphBuilderPatternOptimization-VNK.optimize] skips MulMulMulScalarPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips ReduceReshapePattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips ReduceSumNormalizePattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [PatternOptimization.enumerate_matches] start ReshapePattern with main_opset=18 and min_opset=1
    [GraphBuilderPatternOptimization-VNK.optimize] skips ReshapeMatMulReshapePattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips Reshape2Of3Pattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips ReshapeReshapeBinaryPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips MatMulAddPattern, pattern.priority=3, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips GemmTransposePattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips MatMulReshape2Of3Pattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips MulMulMatMulPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [PatternOptimization.enumerate_matches] start ShapeBasedReshapeIsSqueezePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedStaticExpandPattern with main_opset=18 and min_opset=1
    [GraphBuilderPatternOptimization-VNK.optimize] skips ShapeBasedConcatExpandPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [PatternOptimization.enumerate_matches] start ShapeBasedEditDistanceReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedIdentityPattern with main_opset=18 and min_opset=1
    [GraphBuilderPatternOptimization-VNK.optimize] skips ShapeBasedExpandBroadcastPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips ShapeBasedExpandBroadcastMatMulPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips ShapeBasedExpandCastWhereSwapPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips ShapeBasedExpandSwapPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips ShapeBasedMatMulToMulPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [PatternOptimization.enumerate_matches] start ShapeBasedSameChildrenPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedShapeShapeAddPattern with main_opset=18 and min_opset=1
    [ShapeBasedShapeShapeAddPattern.match] NONE - line: 23:experimental_experiment.xoptim.patterns.onnx_shape, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [ShapeBasedShapeShapeAddPattern.match] NONE - line: 23:experimental_experiment.xoptim.patterns.onnx_shape, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start ReshapeReshapePattern with main_opset=18 and min_opset=1
    [GraphBuilderPatternOptimization-VNK.optimize] skips RotaryEmbeddingPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [PatternOptimization.enumerate_matches] start SameChildrenPattern with main_opset=18 and min_opset=1
    [GraphBuilderPatternOptimization-VNK.optimize] skips SequenceConstructAtPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips SliceSlicePattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips SlicesSplitPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [PatternOptimization.enumerate_matches] start SoftmaxCrossEntropyLossCastPattern with main_opset=18 and min_opset=14
    [GraphBuilder-VMU.make_tensor_input] X[0:None] -- marker=_build_pattern1_X
    [GraphBuilder-VMU.set_type] X:0
    [GraphBuilder-VMU.set_type] X:-1
    [GraphBuilder-VMU.make_tensor_input] indices[0:None] -- marker=_build_pattern1_indices
    [GraphBuilder-VMU.set_type] indices:0
    [GraphBuilder-VMU.set_type] indices:-1
    [GraphBuilder-VMU.make_tensor_input] axis[0:None] -- marker=_build_pattern1_axis
    [GraphBuilder-VMU.set_type] axis:0
    [GraphBuilder-VMU.set_type] axis:-1
    [GraphBuilder-VMU.make_tensor_input] zerof[0:None] -- marker=_build_pattern1_zerof
    [GraphBuilder-VMU.set_type] zerof:0
    [GraphBuilder-VMU.set_type] zerof:-1
    [GraphBuilder-VMU.make_tensor_input] zeroi[0:None] -- marker=_build_pattern1_zeroi
    [GraphBuilder-VMU.set_type] zeroi:0
    [GraphBuilder-VMU.set_type] zeroi:-1
    [GraphBuilder-VMU.make_tensor_input] b[0:None] -- marker=_build_pattern1_b
    [GraphBuilder-VMU.set_type] b:0
    [GraphBuilder-VMU.set_type] b:-1
    [GraphBuilder-VMU.make_node] [TT:-] Equal: ['indices', 'b']->['_onx_equal_indices']
    [GraphBuilder-VMU.set_type] _onx_equal_indices:9
    [GraphBuilder-VMU.make_node] [T:-] Not: ['_onx_equal_indices']->['_onx_not_equal_indices']
    [GraphBuilder-VMU.set_type] _onx_not_equal_indices:9
    [GraphBuilder-VMU.make_node] [TTT:-] Where: ['_onx_not_equal_indices', 'indices', 'zeroi']->['_onx_where_not_equal_indices']
    [GraphBuilder-VMU.set_type] _onx_where_not_equal_indices:-1
    [GraphBuilder-VMU.make_node] [TT:-] Unsqueeze: ['_onx_where_not_equal_indices', 'axis']->['_onx_where_not_equal_indices::UnSq']
    [GraphBuilder-VMU.set_type] _onx_where_not_equal_indices::UnSq:-1
    [GraphBuilder-VMU.make_node] [T:-] LogSoftmax: ['X']->['_onx_logsoftmax_X']
    [GraphBuilder-VMU.set_type] _onx_logsoftmax_X:-1
    [GraphBuilder-VMU.set_type] _onx_gatherelements_logsoftmax_X:-1
    [GraphBuilder-VMU.make_node] [TT:T] GatherElements: ['_onx_logsoftmax_X', '_onx_where_not_equal_indices::UnSq']->['_onx_gatherelements_logsoftmax_X']
    [GraphBuilder-VMU.set_type] _onx_gatherelements_logsoftmax_X:-1
    [GraphBuilder-VMU.make_node] [TT:-] Squeeze: ['_onx_gatherelements_logsoftmax_X', 'axis']->['_onx_gatherelements_logsoftmax_X::Sq']
    [GraphBuilder-VMU.set_type] _onx_gatherelements_logsoftmax_X::Sq:-1
    [GraphBuilder-VMU.make_node] [T:-] Neg: ['_onx_gatherelements_logsoftmax_X::Sq']->['_onx_neg_gatherelements_logsoftmax_X::Sq']
    [GraphBuilder-VMU.set_type] _onx_neg_gatherelements_logsoftmax_X::Sq:-1
    [GraphBuilder-VMU.make_node] [TTT:-] Where: ['_onx_not_equal_indices', '_onx_neg_gatherelements_logsoftmax_X::Sq', 'zerof']->['_onx_where_not_equal_indices2']
    [GraphBuilder-VMU.set_type] _onx_where_not_equal_indices2:-1
    [GraphBuilder-VMU.make_node] [T:-] Cast: ['_onx_not_equal_indices']->['_onx_not_equal_indices::C1']
    [GraphBuilder-VMU.set_type] _onx_not_equal_indices::C1:1
    [GraphBuilder-VMU.make_node] [T:-] ReduceSum: ['_onx_not_equal_indices::C1']->['_onx_reducesum_not_equal_indices::C1']
    [GraphBuilder-VMU.set_type] _onx_reducesum_not_equal_indices::C1:1
    [GraphBuilder-VMU.set_shape] _onx_reducesum_not_equal_indices::C1:()
    [GraphBuilder-VMU.set_rank] _onx_reducesum_not_equal_indices::C1:0
    [GraphBuilder-VMU.make_node] [#:-] Cast: ['_onx_reducesum_not_equal_indices::C1']->['_onx_reducesum_not_equal_indices::C1::C10']
    [GraphBuilder-VMU.set_type] _onx_reducesum_not_equal_indices::C1::C10:10
    [GraphBuilder-VMU.set_shape] _onx_reducesum_not_equal_indices::C1::C10:()
    [GraphBuilder-VMU.set_rank] _onx_reducesum_not_equal_indices::C1::C10:0
    [GraphBuilder-VMU.make_node] [T:-] Cast: ['_onx_where_not_equal_indices2']->['_onx_where_not_equal_indices2::C1']
    [GraphBuilder-VMU.set_type] _onx_where_not_equal_indices2::C1:1
    [GraphBuilder-VMU.make_node] [T:-] ReduceSum: ['_onx_where_not_equal_indices2::C1']->['_onx_reducesum_where_not_equal_indices2::C1']
    [GraphBuilder-VMU.set_type] _onx_reducesum_where_not_equal_indices2::C1:1
    [GraphBuilder-VMU.set_shape] _onx_reducesum_where_not_equal_indices2::C1:()
    [GraphBuilder-VMU.set_rank] _onx_reducesum_where_not_equal_indices2::C1:0
    [GraphBuilder-VMU.make_node] [#:-] Cast: ['_onx_reducesum_where_not_equal_indices2::C1']->['_onx_reducesum_where_not_equal_indices2::C1::C10']
    [GraphBuilder-VMU.set_type] _onx_reducesum_where_not_equal_indices2::C1::C10:10
    [GraphBuilder-VMU.set_shape] _onx_reducesum_where_not_equal_indices2::C1::C10:()
    [GraphBuilder-VMU.set_rank] _onx_reducesum_where_not_equal_indices2::C1::C10:0
    [GraphBuilder-VMU.make_node] [##:-] Div: ['_onx_reducesum_where_not_equal_indices2::C1::C10', '_onx_reducesum_not_equal_indices::C1::C10']->['_onx_div_reducesum_where_not_equal_indices2::C1::C10']
    [GraphBuilder-VMU.set_type] _onx_div_reducesum_where_not_equal_indices2::C1::C10:10
    [GraphBuilder-VMU.set_shape] _onx_div_reducesum_where_not_equal_indices2::C1::C10:()
    [GraphBuilder-VMU.set_rank] _onx_div_reducesum_where_not_equal_indices2::C1::C10:0
    [GraphBuilder-VMU.make_tensor_output] _onx_div_reducesum_where_not_equal_indices2::C1::C10[0: None]
    [GraphBuilderPatternOptimization-VNK.optimize] skips SplitConcatPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [PatternOptimization.enumerate_matches] start SqueezeAddPattern with main_opset=18 and min_opset=1
    [SqueezeAddPattern.match] NONE - line: 211:experimental_experiment.xoptim.patterns.onnx_unsqueeze, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [SqueezeAddPattern.match] NONE - line: 211:experimental_experiment.xoptim.patterns.onnx_unsqueeze, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start SqueezeUnsqueezePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start StaticConcatReshapePattern with main_opset=18 and min_opset=1
    [GraphBuilderPatternOptimization-VNK.optimize] skips Sub1MulPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips SwitchOrderBinaryPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips SwitchReshapeActivationPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips TransposeEqualReshapePattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips TransposeMatMulPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips TransposeReshapeMatMulPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [PatternOptimization.enumerate_matches] start TransposeReshapeTransposePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start TransposeTransposePattern with main_opset=18 and min_opset=1
    [GraphBuilderPatternOptimization-VNK.optimize] skips UnsqueezeEqualPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [PatternOptimization.enumerate_matches] start UnsqueezeUnsqueezePattern with main_opset=18 and min_opset=1
    [GraphBuilderPatternOptimization-VNK.optimize] skips RotaryConcatPartPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips FunctionCausalMaskPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips FunctionCausalMaskMulAddPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips FunctionCosSinCachePattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips FunctionHalfRotaryEmbeddingPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] skips RMSNormalizationPattern, pattern.priority=1, current_priority_index=0, priorities[current_priority_index]=0 priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] done all: -0 +0 nodes
    [GraphBuilder-VNK.remove_identity_nodes] -- starts with 5
    [GraphBuilder-VNK.remove_identity_nodes] found 0 replacements
    [GraphBuilder-VNK.remove_identity_nodes] kept 5 nodes
    [GraphBuilder-VNK.remove_identity_nodes] ends with 5 nodes in 4.004099901067093e-05 seconds
    [GraphBuilderPatternOptimization-VNK.optimize] increase priority to 1
    [GraphBuilderPatternOptimization-VNK.optimize] iteration 1: 5 nodes, priority=1
    [PatternOptimization.enumerate_matches] start BatchNormalizationPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start BatchNormalizationTrainingPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start CastLayerNormalizationCastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start CastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start CastCastBinaryPattern with main_opset=18 and min_opset=1
    [CastCastBinaryPattern.match] NONE - line: 87:experimental_experiment.xoptim.patterns.onnx_cast, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [CastCastBinaryPattern.match] NONE - line: 87:experimental_experiment.xoptim.patterns.onnx_cast, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start CastOpCastPattern with main_opset=18 and min_opset=1
    [CastOpCastPattern.match] NONE - line: 180:experimental_experiment.xoptim.patterns.onnx_cast, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [CastOpCastPattern.match] NONE - line: 177:experimental_experiment.xoptim.patterns.onnx_cast, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start ClipClipPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ComputationCastOpCastPattern with main_opset=18 and min_opset=1
    [ComputationCastOpCastPattern.match] NONE - line: 334:experimental_experiment.xoptim.patterns.onnx_cast, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [ComputationCastOpCastPattern.match] NONE - line: 334:experimental_experiment.xoptim.patterns.onnx_cast, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start ConcatEmptyPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConcatGatherPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConcatReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConcatTwiceUnaryPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConvBiasNullPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start DropoutPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ExpandPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ExpandBroadcastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ExpandSwapPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start GeluPattern with main_opset=18 and min_opset=20
    [PatternOptimization.enumerate_matches] start IdentityPattern with main_opset=18 and min_opset=1
    [IdentityPattern.match] NONE - line: 297:experimental_experiment.xoptim.patterns.onnx_any, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [IdentityPattern.match] NONE - line: 339:experimental_experiment.xoptim.patterns.onnx_any, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start LayerNormalizationPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start LayerNormalizationScalePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start LeakyReluPattern with main_opset=18 and min_opset=6
    [PatternOptimization.enumerate_matches] start MulMulMulScalarPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReduceReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReduceSumNormalizePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReshapeMatMulReshapePattern with main_opset=18 and min_opset=1
    [ReshapeMatMulReshapePattern.match] NONE - line: 780:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset, inputs=x,p_layers_0_weight::T10
    [ReshapeMatMulReshapePattern.match] NONE - line: 780:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset3, inputs=relu,p_layers_2_weight::T10
    [PatternOptimization.enumerate_matches] start Reshape2Of3Pattern with main_opset=18 and min_opset=1
    [Reshape2Of3Pattern.match] NONE - line: 306:experimental_experiment.xoptim.patterns.onnx_reshape, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [Reshape2Of3Pattern.match] NONE - line: 306:experimental_experiment.xoptim.patterns.onnx_reshape, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start ReshapeReshapeBinaryPattern with main_opset=18 and min_opset=1
    [ReshapeReshapeBinaryPattern.match] NONE - line: 486:experimental_experiment.xoptim.patterns.onnx_reshape, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [ReshapeReshapeBinaryPattern.match] NONE - line: 486:experimental_experiment.xoptim.patterns.onnx_reshape, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [GraphBuilderPatternOptimization-VNK.optimize] skips MatMulAddPattern, pattern.priority=3, current_priority_index=1, priorities[current_priority_index]=1 priorities=[0, 1, 3]
    [PatternOptimization.enumerate_matches] start GemmTransposePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start MatMulReshape2Of3Pattern with main_opset=18 and min_opset=1
    [MatMulReshape2Of3Pattern.match] NONE - line: 400:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset, inputs=x,p_layers_0_weight::T10
    [MatMulReshape2Of3Pattern.match] NONE - line: 400:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset3, inputs=relu,p_layers_2_weight::T10
    [PatternOptimization.enumerate_matches] start MulMulMatMulPattern with main_opset=18 and min_opset=1
    [MulMulMatMulPattern.match] NONE - line: 716:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset, inputs=x,p_layers_0_weight::T10
    [MulMulMatMulPattern.match] NONE - line: 716:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset3, inputs=relu,p_layers_2_weight::T10
    [PatternOptimization.enumerate_matches] start ShapeBasedReshapeIsSqueezePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedStaticExpandPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedConcatExpandPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedEditDistanceReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedIdentityPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandBroadcastPattern with main_opset=18 and min_opset=1
    [ShapeBasedExpandBroadcastPattern.match] NONE - line: 232:experimental_experiment.xoptim.patterns.onnx_expand, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [ShapeBasedExpandBroadcastPattern.match] NONE - line: 232:experimental_experiment.xoptim.patterns.onnx_expand, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandBroadcastMatMulPattern with main_opset=18 and min_opset=1
    [ShapeBasedExpandBroadcastMatMulPattern.match] NONE - line: 710:experimental_experiment.xoptim.patterns.onnx_expand, op_type=MatMul, name=Opset, inputs=x,p_layers_0_weight::T10
    [ShapeBasedExpandBroadcastMatMulPattern.match] NONE - line: 710:experimental_experiment.xoptim.patterns.onnx_expand, op_type=MatMul, name=Opset3, inputs=relu,p_layers_2_weight::T10
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandCastWhereSwapPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandSwapPattern with main_opset=18 and min_opset=1
    [ShapeBasedExpandSwapPattern.match] NONE - line: 560:experimental_experiment.xoptim.patterns.onnx_expand, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [ShapeBasedExpandSwapPattern.match] NONE - line: 560:experimental_experiment.xoptim.patterns.onnx_expand, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start ShapeBasedMatMulToMulPattern with main_opset=18 and min_opset=1
    [ShapeBasedMatMulToMulPattern.match] NONE - line: 1255:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset, inputs=x,p_layers_0_weight::T10
    [ShapeBasedMatMulToMulPattern.match] NONE - line: 1255:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset3, inputs=relu,p_layers_2_weight::T10
    [PatternOptimization.enumerate_matches] start ShapeBasedSameChildrenPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedShapeShapeAddPattern with main_opset=18 and min_opset=1
    [ShapeBasedShapeShapeAddPattern.match] NONE - line: 23:experimental_experiment.xoptim.patterns.onnx_shape, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [ShapeBasedShapeShapeAddPattern.match] NONE - line: 23:experimental_experiment.xoptim.patterns.onnx_shape, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start ReshapeReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start RotaryEmbeddingPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SameChildrenPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SequenceConstructAtPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SliceSlicePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SlicesSplitPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SoftmaxCrossEntropyLossCastPattern with main_opset=18 and min_opset=14
    [PatternOptimization.enumerate_matches] start SplitConcatPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SqueezeAddPattern with main_opset=18 and min_opset=1
    [SqueezeAddPattern.match] NONE - line: 211:experimental_experiment.xoptim.patterns.onnx_unsqueeze, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [SqueezeAddPattern.match] NONE - line: 211:experimental_experiment.xoptim.patterns.onnx_unsqueeze, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start SqueezeUnsqueezePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start StaticConcatReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start Sub1MulPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SwitchOrderBinaryPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SwitchReshapeActivationPattern with main_opset=18 and min_opset=1
    [SwitchReshapeActivationPattern.match] NONE - line: 1178:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Relu, name=relu, inputs=linear
    [PatternOptimization.enumerate_matches] start TransposeEqualReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start TransposeMatMulPattern with main_opset=18 and min_opset=1
    [TransposeMatMulPattern.match] NONE - line: 890:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset, inputs=x,p_layers_0_weight::T10
    [TransposeMatMulPattern.match] NONE - line: 890:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset3, inputs=relu,p_layers_2_weight::T10
    [PatternOptimization.enumerate_matches] start TransposeReshapeMatMulPattern with main_opset=18 and min_opset=1
    [TransposeReshapeMatMulPattern.match] NONE - line: 1033:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset, inputs=x,p_layers_0_weight::T10
    [TransposeReshapeMatMulPattern.match] NONE - line: 1033:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset3, inputs=relu,p_layers_2_weight::T10
    [PatternOptimization.enumerate_matches] start TransposeReshapeTransposePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start TransposeTransposePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start UnsqueezeEqualPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start UnsqueezeUnsqueezePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start RotaryConcatPartPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start FunctionCausalMaskPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start FunctionCausalMaskMulAddPattern with main_opset=18 and min_opset=1
    [FunctionCausalMaskMulAddPattern.match] NONE - line: 1119:experimental_experiment.xoptim.patterns.onnx_rotary, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [FunctionCausalMaskMulAddPattern.match] NONE - line: 1119:experimental_experiment.xoptim.patterns.onnx_rotary, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start FunctionCosSinCachePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start FunctionHalfRotaryEmbeddingPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start RMSNormalizationPattern with main_opset=18 and min_opset=1
    [GraphBuilderPatternOptimization-VNK.optimize] done all: -0 +0 nodes
    [GraphBuilder-VNK.remove_identity_nodes] -- starts with 5
    [GraphBuilder-VNK.remove_identity_nodes] found 0 replacements
    [GraphBuilder-VNK.remove_identity_nodes] kept 5 nodes
    [GraphBuilder-VNK.remove_identity_nodes] ends with 5 nodes in 4.6549999751732685e-05 seconds
    [GraphBuilderPatternOptimization-VNK.optimize] increase priority to 3
    [GraphBuilderPatternOptimization-VNK.optimize] iteration 2: 5 nodes, priority=3
    [PatternOptimization.enumerate_matches] start BatchNormalizationPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start BatchNormalizationTrainingPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start CastLayerNormalizationCastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start CastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start CastCastBinaryPattern with main_opset=18 and min_opset=1
    [CastCastBinaryPattern.match] NONE - line: 87:experimental_experiment.xoptim.patterns.onnx_cast, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [CastCastBinaryPattern.match] NONE - line: 87:experimental_experiment.xoptim.patterns.onnx_cast, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start CastOpCastPattern with main_opset=18 and min_opset=1
    [CastOpCastPattern.match] NONE - line: 180:experimental_experiment.xoptim.patterns.onnx_cast, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [CastOpCastPattern.match] NONE - line: 177:experimental_experiment.xoptim.patterns.onnx_cast, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start ClipClipPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ComputationCastOpCastPattern with main_opset=18 and min_opset=1
    [ComputationCastOpCastPattern.match] NONE - line: 334:experimental_experiment.xoptim.patterns.onnx_cast, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [ComputationCastOpCastPattern.match] NONE - line: 334:experimental_experiment.xoptim.patterns.onnx_cast, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start ConcatEmptyPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConcatGatherPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConcatReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConcatTwiceUnaryPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConvBiasNullPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start DropoutPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ExpandPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ExpandBroadcastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ExpandSwapPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start GeluPattern with main_opset=18 and min_opset=20
    [PatternOptimization.enumerate_matches] start IdentityPattern with main_opset=18 and min_opset=1
    [IdentityPattern.match] NONE - line: 297:experimental_experiment.xoptim.patterns.onnx_any, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [IdentityPattern.match] NONE - line: 339:experimental_experiment.xoptim.patterns.onnx_any, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start LayerNormalizationPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start LayerNormalizationScalePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start LeakyReluPattern with main_opset=18 and min_opset=6
    [PatternOptimization.enumerate_matches] start MulMulMulScalarPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReduceReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReduceSumNormalizePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReshapeMatMulReshapePattern with main_opset=18 and min_opset=1
    [ReshapeMatMulReshapePattern.match] NONE - line: 780:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset, inputs=x,p_layers_0_weight::T10
    [ReshapeMatMulReshapePattern.match] NONE - line: 780:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset3, inputs=relu,p_layers_2_weight::T10
    [PatternOptimization.enumerate_matches] start Reshape2Of3Pattern with main_opset=18 and min_opset=1
    [Reshape2Of3Pattern.match] NONE - line: 306:experimental_experiment.xoptim.patterns.onnx_reshape, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [Reshape2Of3Pattern.match] NONE - line: 306:experimental_experiment.xoptim.patterns.onnx_reshape, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start ReshapeReshapeBinaryPattern with main_opset=18 and min_opset=1
    [ReshapeReshapeBinaryPattern.match] NONE - line: 486:experimental_experiment.xoptim.patterns.onnx_reshape, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [ReshapeReshapeBinaryPattern.match] NONE - line: 486:experimental_experiment.xoptim.patterns.onnx_reshape, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start MatMulAddPattern with main_opset=18 and min_opset=1
    [MatchResult.match] MATCH MatMulAddPattern with 2 nodes and types ['MatMul', 'Add']
    [GraphBuilderPatternOptimization-VNK.optimize] match=MatchResult: MatMulAddPattern replaces ['MatMul', 'Add']
    [MatchResult.match] MATCH MatMulAddPattern with 2 nodes and types ['MatMul', 'Add']
    [GraphBuilderPatternOptimization-VNK.optimize] match=MatchResult: MatMulAddPattern replaces ['MatMul', 'Add']
    [PatternOptimization.enumerate_matches] start GemmTransposePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start MatMulReshape2Of3Pattern with main_opset=18 and min_opset=1
    [MatMulReshape2Of3Pattern.match] NONE - line: 400:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset, inputs=x,p_layers_0_weight::T10
    [MatMulReshape2Of3Pattern.match] NONE - line: 400:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset3, inputs=relu,p_layers_2_weight::T10
    [PatternOptimization.enumerate_matches] start MulMulMatMulPattern with main_opset=18 and min_opset=1
    [MulMulMatMulPattern.match] NONE - line: 716:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset, inputs=x,p_layers_0_weight::T10
    [MulMulMatMulPattern.match] NONE - line: 716:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset3, inputs=relu,p_layers_2_weight::T10
    [PatternOptimization.enumerate_matches] start ShapeBasedReshapeIsSqueezePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedStaticExpandPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedConcatExpandPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedEditDistanceReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedIdentityPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandBroadcastPattern with main_opset=18 and min_opset=1
    [ShapeBasedExpandBroadcastPattern.match] NONE - line: 232:experimental_experiment.xoptim.patterns.onnx_expand, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [ShapeBasedExpandBroadcastPattern.match] NONE - line: 232:experimental_experiment.xoptim.patterns.onnx_expand, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandBroadcastMatMulPattern with main_opset=18 and min_opset=1
    [ShapeBasedExpandBroadcastMatMulPattern.match] NONE - line: 710:experimental_experiment.xoptim.patterns.onnx_expand, op_type=MatMul, name=Opset, inputs=x,p_layers_0_weight::T10
    [ShapeBasedExpandBroadcastMatMulPattern.match] NONE - line: 710:experimental_experiment.xoptim.patterns.onnx_expand, op_type=MatMul, name=Opset3, inputs=relu,p_layers_2_weight::T10
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandCastWhereSwapPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandSwapPattern with main_opset=18 and min_opset=1
    [ShapeBasedExpandSwapPattern.match] NONE - line: 560:experimental_experiment.xoptim.patterns.onnx_expand, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [ShapeBasedExpandSwapPattern.match] NONE - line: 560:experimental_experiment.xoptim.patterns.onnx_expand, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start ShapeBasedMatMulToMulPattern with main_opset=18 and min_opset=1
    [ShapeBasedMatMulToMulPattern.match] NONE - line: 1255:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset, inputs=x,p_layers_0_weight::T10
    [ShapeBasedMatMulToMulPattern.match] NONE - line: 1255:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset3, inputs=relu,p_layers_2_weight::T10
    [PatternOptimization.enumerate_matches] start ShapeBasedSameChildrenPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedShapeShapeAddPattern with main_opset=18 and min_opset=1
    [ShapeBasedShapeShapeAddPattern.match] NONE - line: 23:experimental_experiment.xoptim.patterns.onnx_shape, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [ShapeBasedShapeShapeAddPattern.match] NONE - line: 23:experimental_experiment.xoptim.patterns.onnx_shape, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start ReshapeReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start RotaryEmbeddingPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SameChildrenPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SequenceConstructAtPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SliceSlicePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SlicesSplitPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SoftmaxCrossEntropyLossCastPattern with main_opset=18 and min_opset=14
    [PatternOptimization.enumerate_matches] start SplitConcatPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SqueezeAddPattern with main_opset=18 and min_opset=1
    [SqueezeAddPattern.match] NONE - line: 211:experimental_experiment.xoptim.patterns.onnx_unsqueeze, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [SqueezeAddPattern.match] NONE - line: 211:experimental_experiment.xoptim.patterns.onnx_unsqueeze, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start SqueezeUnsqueezePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start StaticConcatReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start Sub1MulPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SwitchOrderBinaryPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SwitchReshapeActivationPattern with main_opset=18 and min_opset=1
    [SwitchReshapeActivationPattern.match] NONE - line: 1178:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Relu, name=relu, inputs=linear
    [PatternOptimization.enumerate_matches] start TransposeEqualReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start TransposeMatMulPattern with main_opset=18 and min_opset=1
    [TransposeMatMulPattern.match] NONE - line: 890:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset, inputs=x,p_layers_0_weight::T10
    [TransposeMatMulPattern.match] NONE - line: 890:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset3, inputs=relu,p_layers_2_weight::T10
    [PatternOptimization.enumerate_matches] start TransposeReshapeMatMulPattern with main_opset=18 and min_opset=1
    [TransposeReshapeMatMulPattern.match] NONE - line: 1033:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset, inputs=x,p_layers_0_weight::T10
    [TransposeReshapeMatMulPattern.match] NONE - line: 1033:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=MatMul, name=Opset3, inputs=relu,p_layers_2_weight::T10
    [PatternOptimization.enumerate_matches] start TransposeReshapeTransposePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start TransposeTransposePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start UnsqueezeEqualPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start UnsqueezeUnsqueezePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start RotaryConcatPartPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start FunctionCausalMaskPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start FunctionCausalMaskMulAddPattern with main_opset=18 and min_opset=1
    [FunctionCausalMaskMulAddPattern.match] NONE - line: 1119:experimental_experiment.xoptim.patterns.onnx_rotary, op_type=Add, name=Opset2, inputs=_onx_matmul_x,layers.0.bias
    [FunctionCausalMaskMulAddPattern.match] NONE - line: 1119:experimental_experiment.xoptim.patterns.onnx_rotary, op_type=Add, name=Opset4, inputs=_onx_matmul_relu,layers.2.bias
    [PatternOptimization.enumerate_matches] start FunctionCosSinCachePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start FunctionHalfRotaryEmbeddingPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start RMSNormalizationPattern with main_opset=18 and min_opset=1
    [GraphBuilderPatternOptimization-VNK.optimize] applies 2 matches, 2*MatMulAddPattern - time=0.001 | max_time=IdentityPattern:0.000
    [GraphBuilderPatternOptimization-VNK.optimize] apply MatchResult: MatMulAddPattern replaces ['MatMul', 'Add'], inputs: ['x', 'p_layers_0_weight::T10', '_onx_matmul_x', 'layers.0.bias'], outputs: ['_onx_matmul_x', 'linear']
    [GraphBuilderPatternOptimization-VNK.apply_match] MatchResult: MatMulAddPattern replaces ['MatMul', 'Add']
      - MatMul: ['x', 'p_layers_0_weight::T10'] -> ['_onx_matmul_x']
      - Add: ['_onx_matmul_x', 'layers.0.bias'] -> ['linear']
      + Gemm: ['x', 'p_layers_0_weight::T10', 'layers.0.bias'] -> ['linear']
    [GraphBuilder-VNK.set_type] linear:1
    [GraphBuilderPatternOptimization-VNK.apply_match] MatchResult: MatMulAddPattern replaces ['MatMul', 'Add'] applied.
    [GraphBuilderPatternOptimization-VNK.optimize] - add ['Gemm']
    [GraphBuilderPatternOptimization-VNK.optimize] done MatchResult: MatMulAddPattern replaces ['MatMul', 'Add']: -2 +1 nodes
    [GraphBuilderPatternOptimization-VNK.optimize] removed outputs {'_onx_matmul_x'}
    [GraphBuilderPatternOptimization-VNK.optimize] apply MatchResult: MatMulAddPattern replaces ['MatMul', 'Add'], inputs: ['relu', 'p_layers_2_weight::T10', '_onx_matmul_relu', 'layers.2.bias'], outputs: ['_onx_matmul_relu', 'output_0']
    [GraphBuilderPatternOptimization-VNK.apply_match] MatchResult: MatMulAddPattern replaces ['MatMul', 'Add']
      - MatMul: ['relu', 'p_layers_2_weight::T10'] -> ['_onx_matmul_relu']
      - Add: ['_onx_matmul_relu', 'layers.2.bias'] -> ['output_0']
      + Gemm: ['relu', 'p_layers_2_weight::T10', 'layers.2.bias'] -> ['output_0']
    [GraphBuilder-VNK.set_type] output_0:1
    [GraphBuilderPatternOptimization-VNK.apply_match] MatchResult: MatMulAddPattern replaces ['MatMul', 'Add'] applied.
    [GraphBuilderPatternOptimization-VNK.optimize] - add ['Gemm']
    [GraphBuilderPatternOptimization-VNK.optimize] done MatchResult: MatMulAddPattern replaces ['MatMul', 'Add']: -2 +1 nodes
    [GraphBuilderPatternOptimization-VNK.optimize] removed outputs {'_onx_matmul_relu'}
    [GraphBuilderPatternOptimization-VNK.optimize] done all: -4 +2 nodes
    [GraphBuilder-VNK.remove_identity_nodes] -- starts with 3
    [GraphBuilder-VNK.remove_identity_nodes] found 0 replacements
    [GraphBuilder-VNK.remove_identity_nodes] kept 3 nodes
    [GraphBuilder-VNK.remove_identity_nodes] ends with 3 nodes in 3.006999941135291e-05 seconds
    [GraphBuilderPatternOptimization-VNK.optimize] iteration 3: 3 nodes, priority=3
    [PatternOptimization.enumerate_matches] start BatchNormalizationPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start BatchNormalizationTrainingPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start CastLayerNormalizationCastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start CastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start CastCastBinaryPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start CastOpCastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ClipClipPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ComputationCastOpCastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConcatEmptyPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConcatGatherPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConcatReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConcatTwiceUnaryPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConvBiasNullPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start DropoutPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ExpandPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ExpandBroadcastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ExpandSwapPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start GeluPattern with main_opset=18 and min_opset=20
    [PatternOptimization.enumerate_matches] start IdentityPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start LayerNormalizationPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start LayerNormalizationScalePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start LeakyReluPattern with main_opset=18 and min_opset=6
    [PatternOptimization.enumerate_matches] start MulMulMulScalarPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReduceReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReduceSumNormalizePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReshapeMatMulReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start Reshape2Of3Pattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReshapeReshapeBinaryPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start MatMulAddPattern with main_opset=18 and min_opset=1
    [MatMulAddPattern.match] NONE - line: 58:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Gemm, name=MatMulAddPattern--Opset, inputs=x,p_layers_0_weight::T10,layers.0.bias
    [MatMulAddPattern.match] NONE - line: 55:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Gemm, name=MatMulAddPattern--Opset3, inputs=relu,p_layers_2_weight::T10,layers.2.bias
    [PatternOptimization.enumerate_matches] start GemmTransposePattern with main_opset=18 and min_opset=1
    [MatchResult.match] MATCH GemmTransposePattern with 1 nodes and types ['Gemm']
    [GraphBuilderPatternOptimization-VNK.optimize] match=MatchResult: GemmTransposePattern replaces ['Gemm']
    [MatchResult.match] MATCH GemmTransposePattern with 1 nodes and types ['Gemm']
    [GraphBuilderPatternOptimization-VNK.optimize] match=MatchResult: GemmTransposePattern replaces ['Gemm']
    [PatternOptimization.enumerate_matches] start MatMulReshape2Of3Pattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start MulMulMatMulPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedReshapeIsSqueezePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedStaticExpandPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedConcatExpandPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedEditDistanceReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedIdentityPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandBroadcastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandBroadcastMatMulPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandCastWhereSwapPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandSwapPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedMatMulToMulPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedSameChildrenPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedShapeShapeAddPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReshapeReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start RotaryEmbeddingPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SameChildrenPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SequenceConstructAtPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SliceSlicePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SlicesSplitPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SoftmaxCrossEntropyLossCastPattern with main_opset=18 and min_opset=14
    [PatternOptimization.enumerate_matches] start SplitConcatPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SqueezeAddPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SqueezeUnsqueezePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start StaticConcatReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start Sub1MulPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SwitchOrderBinaryPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SwitchReshapeActivationPattern with main_opset=18 and min_opset=1
    [SwitchReshapeActivationPattern.match] NONE - line: 1178:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Relu, name=relu, inputs=linear
    [PatternOptimization.enumerate_matches] start TransposeEqualReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start TransposeMatMulPattern with main_opset=18 and min_opset=1
    [TransposeMatMulPattern.match] NONE - line: 890:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Gemm, name=MatMulAddPattern--Opset, inputs=x,p_layers_0_weight::T10,layers.0.bias
    [TransposeMatMulPattern.match] NONE - line: 890:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Gemm, name=MatMulAddPattern--Opset3, inputs=relu,p_layers_2_weight::T10,layers.2.bias
    [PatternOptimization.enumerate_matches] start TransposeReshapeMatMulPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start TransposeReshapeTransposePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start TransposeTransposePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start UnsqueezeEqualPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start UnsqueezeUnsqueezePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start RotaryConcatPartPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start FunctionCausalMaskPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start FunctionCausalMaskMulAddPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start FunctionCosSinCachePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start FunctionHalfRotaryEmbeddingPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start RMSNormalizationPattern with main_opset=18 and min_opset=1
    [GraphBuilderPatternOptimization-VNK.optimize] applies 2 matches, 2*GemmTransposePattern - time=0.001 | max_time=GemmTransposePattern:0.000
    [GraphBuilderPatternOptimization-VNK.optimize] apply MatchResult: GemmTransposePattern replaces ['Gemm'], inputs: ['x', 'p_layers_0_weight::T10', 'layers.0.bias'], outputs: ['linear']
    [GraphBuilder-VNK.update_node_constant] new constant 'GemmTransposePattern--p_layers_0_weight::T10', node=Transpose
    [GraphBuilderPatternOptimization-VNK.apply_match] MatchResult: GemmTransposePattern replaces ['Gemm']
      - Gemm: ['x', 'p_layers_0_weight::T10', 'layers.0.bias'] -> ['linear']
      + Transpose: ['p_layers_0_weight::T10'] -> ['GemmTransposePattern--p_layers_0_weight::T10']
      + Gemm: ['x', 'GemmTransposePattern--p_layers_0_weight::T10', 'layers.0.bias'] -> ['linear']
    [GraphBuilder-VNK.update_node_constant] new constant 'GemmTransposePattern--p_layers_0_weight::T10', node=Transpose
    [GraphBuilder-VNK.set_type] GemmTransposePattern--p_layers_0_weight::T10:1
    [GraphBuilder-VNK.set_shape] GemmTransposePattern--p_layers_0_weight::T10:(32, 10)
    [GraphBuilder-VNK.set_rank] GemmTransposePattern--p_layers_0_weight::T10:2
    [GraphBuilder-VNK.set_type] linear:1
    [GraphBuilderPatternOptimization-VNK.apply_match] MatchResult: GemmTransposePattern replaces ['Gemm'] applied.
    [GraphBuilderPatternOptimization-VNK.optimize] - add ['Transpose', 'Gemm']
    [GraphBuilderPatternOptimization-VNK.optimize] done MatchResult: GemmTransposePattern replaces ['Gemm']: -1 +2 nodes
    [GraphBuilderPatternOptimization-VNK.optimize] apply MatchResult: GemmTransposePattern replaces ['Gemm'], inputs: ['relu', 'p_layers_2_weight::T10', 'layers.2.bias'], outputs: ['output_0']
    [GraphBuilder-VNK.update_node_constant] new constant 'GemmTransposePattern--p_layers_2_weight::T10', node=Transpose
    [GraphBuilderPatternOptimization-VNK.apply_match] MatchResult: GemmTransposePattern replaces ['Gemm']
      - Gemm: ['relu', 'p_layers_2_weight::T10', 'layers.2.bias'] -> ['output_0']
      + Transpose: ['p_layers_2_weight::T10'] -> ['GemmTransposePattern--p_layers_2_weight::T10']
      + Gemm: ['relu', 'GemmTransposePattern--p_layers_2_weight::T10', 'layers.2.bias'] -> ['output_0']
    [GraphBuilder-VNK.update_node_constant] new constant 'GemmTransposePattern--p_layers_2_weight::T10', node=Transpose
    [GraphBuilder-VNK.set_type] GemmTransposePattern--p_layers_2_weight::T10:1
    [GraphBuilder-VNK.set_shape] GemmTransposePattern--p_layers_2_weight::T10:(1, 32)
    [GraphBuilder-VNK.set_rank] GemmTransposePattern--p_layers_2_weight::T10:2
    [GraphBuilder-VNK.set_type] output_0:1
    [GraphBuilderPatternOptimization-VNK.apply_match] MatchResult: GemmTransposePattern replaces ['Gemm'] applied.
    [GraphBuilderPatternOptimization-VNK.optimize] - add ['Transpose', 'Gemm']
    [GraphBuilderPatternOptimization-VNK.optimize] done MatchResult: GemmTransposePattern replaces ['Gemm']: -1 +2 nodes
    [GraphBuilderPatternOptimization-VNK.optimize] done all: -2 +4 nodes
    [GraphBuilderPatternOptimization-VNK.optimize] iteration 4: 5 nodes, priority=3
    [PatternOptimization.enumerate_matches] start BatchNormalizationPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start BatchNormalizationTrainingPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start CastLayerNormalizationCastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start CastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start CastCastBinaryPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start CastOpCastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ClipClipPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ComputationCastOpCastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConcatEmptyPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConcatGatherPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConcatReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConcatTwiceUnaryPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConvBiasNullPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start DropoutPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ExpandPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ExpandBroadcastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ExpandSwapPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start GeluPattern with main_opset=18 and min_opset=20
    [PatternOptimization.enumerate_matches] start IdentityPattern with main_opset=18 and min_opset=1
    [IdentityPattern.match] NONE - line: 258:experimental_experiment.xoptim.patterns.onnx_any, op_type=Transpose, name=GemmTransposePattern--MatMulAddPattern--Opset, inputs=p_layers_0_weight::T10
    [IdentityPattern.match] NONE - line: 258:experimental_experiment.xoptim.patterns.onnx_any, op_type=Transpose, name=GemmTransposePattern--MatMulAddPattern--Opset3, inputs=p_layers_2_weight::T10
    [PatternOptimization.enumerate_matches] start LayerNormalizationPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start LayerNormalizationScalePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start LeakyReluPattern with main_opset=18 and min_opset=6
    [PatternOptimization.enumerate_matches] start MulMulMulScalarPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReduceReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReduceSumNormalizePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReshapeMatMulReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start Reshape2Of3Pattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReshapeReshapeBinaryPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start MatMulAddPattern with main_opset=18 and min_opset=1
    [MatMulAddPattern.match] NONE - line: 58:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Gemm, name=GemmTransposePattern--MatMulAddPattern--Opset2, inputs=x,GemmTransposePattern--p_layers_0_weight::T10,layers.0.bias
    [MatMulAddPattern.match] NONE - line: 55:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Gemm, name=GemmTransposePattern--MatMulAddPattern--Opset32, inputs=relu,GemmTransposePattern--p_layers_2_weight::T10,layers.2.bias
    [PatternOptimization.enumerate_matches] start GemmTransposePattern with main_opset=18 and min_opset=1
    [GemmTransposePattern.match] NONE - line: 307:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Gemm, name=GemmTransposePattern--MatMulAddPattern--Opset2, inputs=x,GemmTransposePattern--p_layers_0_weight::T10,layers.0.bias
    [GemmTransposePattern.match] NONE - line: 307:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Gemm, name=GemmTransposePattern--MatMulAddPattern--Opset32, inputs=relu,GemmTransposePattern--p_layers_2_weight::T10,layers.2.bias
    [PatternOptimization.enumerate_matches] start MatMulReshape2Of3Pattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start MulMulMatMulPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedReshapeIsSqueezePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedStaticExpandPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedConcatExpandPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedEditDistanceReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedIdentityPattern with main_opset=18 and min_opset=1
    [ShapeBasedIdentityPattern.match] NONE - line: 401:experimental_experiment.xoptim.patterns.onnx_any, op_type=Transpose, name=GemmTransposePattern--MatMulAddPattern--Opset, inputs=p_layers_0_weight::T10
    [ShapeBasedIdentityPattern.match] NONE - line: 401:experimental_experiment.xoptim.patterns.onnx_any, op_type=Transpose, name=GemmTransposePattern--MatMulAddPattern--Opset3, inputs=p_layers_2_weight::T10
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandBroadcastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandBroadcastMatMulPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandCastWhereSwapPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandSwapPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedMatMulToMulPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedSameChildrenPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedShapeShapeAddPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReshapeReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start RotaryEmbeddingPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SameChildrenPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SequenceConstructAtPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SliceSlicePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SlicesSplitPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SoftmaxCrossEntropyLossCastPattern with main_opset=18 and min_opset=14
    [PatternOptimization.enumerate_matches] start SplitConcatPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SqueezeAddPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SqueezeUnsqueezePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start StaticConcatReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start Sub1MulPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SwitchOrderBinaryPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SwitchReshapeActivationPattern with main_opset=18 and min_opset=1
    [SwitchReshapeActivationPattern.match] NONE - line: 1178:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Relu, name=relu, inputs=linear
    [PatternOptimization.enumerate_matches] start TransposeEqualReshapePattern with main_opset=18 and min_opset=1
    [TransposeEqualReshapePattern.match] NONE - line: 342:experimental_experiment.xoptim.patterns.onnx_transpose, op_type=Transpose, name=GemmTransposePattern--MatMulAddPattern--Opset, inputs=p_layers_0_weight::T10
    [MatchResult.match] MATCH TransposeEqualReshapePattern with 1 nodes and types ['Transpose']
    [GraphBuilderPatternOptimization-VNK.optimize] match=MatchResult: TransposeEqualReshapePattern replaces ['Transpose']
    [PatternOptimization.enumerate_matches] start TransposeMatMulPattern with main_opset=18 and min_opset=1
    [TransposeMatMulPattern.match] NONE - line: 928:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Gemm, name=GemmTransposePattern--MatMulAddPattern--Opset2, inputs=x,GemmTransposePattern--p_layers_0_weight::T10,layers.0.bias
    [TransposeMatMulPattern.match] NONE - line: 928:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Gemm, name=GemmTransposePattern--MatMulAddPattern--Opset32, inputs=relu,GemmTransposePattern--p_layers_2_weight::T10,layers.2.bias
    [PatternOptimization.enumerate_matches] start TransposeReshapeMatMulPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start TransposeReshapeTransposePattern with main_opset=18 and min_opset=1
    [TransposeReshapeTransposePattern.match] NONE - line: 140:experimental_experiment.xoptim.patterns.onnx_transpose, op_type=Transpose, name=GemmTransposePattern--MatMulAddPattern--Opset, inputs=p_layers_0_weight::T10
    [TransposeReshapeTransposePattern.match] NONE - line: 140:experimental_experiment.xoptim.patterns.onnx_transpose, op_type=Transpose, name=GemmTransposePattern--MatMulAddPattern--Opset3, inputs=p_layers_2_weight::T10
    [PatternOptimization.enumerate_matches] start TransposeTransposePattern with main_opset=18 and min_opset=1
    [TransposeTransposePattern.match] NONE - line: 51:experimental_experiment.xoptim.patterns.onnx_transpose, op_type=Transpose, name=GemmTransposePattern--MatMulAddPattern--Opset, inputs=p_layers_0_weight::T10
    [TransposeTransposePattern.match] NONE - line: 51:experimental_experiment.xoptim.patterns.onnx_transpose, op_type=Transpose, name=GemmTransposePattern--MatMulAddPattern--Opset3, inputs=p_layers_2_weight::T10
    [PatternOptimization.enumerate_matches] start UnsqueezeEqualPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start UnsqueezeUnsqueezePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start RotaryConcatPartPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start FunctionCausalMaskPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start FunctionCausalMaskMulAddPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start FunctionCosSinCachePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start FunctionHalfRotaryEmbeddingPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start RMSNormalizationPattern with main_opset=18 and min_opset=1
    [GraphBuilderPatternOptimization-VNK.optimize] applies 1 matches, [0]=MatchResult: TransposeEqualReshapePattern replaces ['Transpose'] - time=0.001 | max_time=Reshape2Of3Pattern:0.000
    [GraphBuilderPatternOptimization-VNK.optimize] apply MatchResult: TransposeEqualReshapePattern replaces ['Transpose'], inputs: ['p_layers_2_weight::T10'], outputs: ['GemmTransposePattern--p_layers_2_weight::T10']
    [GraphBuilder-VNK.set_shape] init7_s2_1_-1:(2,)
    [GraphBuilder-VNK.set_rank] init7_s2_1_-1:1
    [GraphBuilder-VNK.set_type] init7_s2_1_-1:7
    [GraphBuilder-VNK.make_initializer] init7_s2_1_-1[7:(2,)]
    [GraphBuilder-VNK.update_node_constant] new constant 'init7_s2_1_-1', node=None
    [GraphBuilder-VNK.update_node_constant] new constant 'GemmTransposePattern--p_layers_2_weight::T10', node=Reshape
    [GraphBuilderPatternOptimization-VNK.apply_match] MatchResult: TransposeEqualReshapePattern replaces ['Transpose']
      - Transpose: ['p_layers_2_weight::T10'] -> ['GemmTransposePattern--p_layers_2_weight::T10']
      + Reshape: ['p_layers_2_weight::T10', 'init7_s2_1_-1'] -> ['GemmTransposePattern--p_layers_2_weight::T10']
    [GraphBuilder-VNK.update_node_constant] new constant 'GemmTransposePattern--p_layers_2_weight::T10', node=Reshape
    [GraphBuilder-VNK.set_type] GemmTransposePattern--p_layers_2_weight::T10:1
    [GraphBuilder-VNK.set_type] GemmTransposePattern--p_layers_2_weight::T10:1
    [GraphBuilderPatternOptimization-VNK.apply_match] MatchResult: TransposeEqualReshapePattern replaces ['Transpose'] applied.
    [GraphBuilderPatternOptimization-VNK.optimize] - add ['Reshape']
    [GraphBuilderPatternOptimization-VNK.optimize] done MatchResult: TransposeEqualReshapePattern replaces ['Transpose']: -1 +1 nodes
    [GraphBuilderPatternOptimization-VNK.optimize] done all: -1 +1 nodes
    [GraphBuilderPatternOptimization-VNK.optimize] iteration 5: 5 nodes, priority=3
    [PatternOptimization.enumerate_matches] start BatchNormalizationPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start BatchNormalizationTrainingPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start CastLayerNormalizationCastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start CastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start CastCastBinaryPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start CastOpCastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ClipClipPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ComputationCastOpCastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConcatEmptyPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConcatGatherPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConcatReshapePattern with main_opset=18 and min_opset=1
    [ConcatReshapePattern.match] NONE - line: 552:experimental_experiment.xoptim.patterns.onnx_reshape, op_type=Reshape, name=TransposeEqualReshapePattern--B--GemmTransposePattern--MatMulAddPattern--Opset3, inputs=p_layers_2_weight::T10,init7_s2_1_-1
    [PatternOptimization.enumerate_matches] start ConcatTwiceUnaryPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ConvBiasNullPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start DropoutPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ExpandPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ExpandBroadcastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ExpandSwapPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start GeluPattern with main_opset=18 and min_opset=20
    [PatternOptimization.enumerate_matches] start IdentityPattern with main_opset=18 and min_opset=1
    [IdentityPattern.match] NONE - line: 258:experimental_experiment.xoptim.patterns.onnx_any, op_type=Transpose, name=GemmTransposePattern--MatMulAddPattern--Opset, inputs=p_layers_0_weight::T10
    [PatternOptimization.enumerate_matches] start LayerNormalizationPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start LayerNormalizationScalePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start LeakyReluPattern with main_opset=18 and min_opset=6
    [PatternOptimization.enumerate_matches] start MulMulMulScalarPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReduceReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReduceSumNormalizePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReshapePattern with main_opset=18 and min_opset=1
    [ReshapePattern.match] NONE - line: 37:experimental_experiment.xoptim.patterns.onnx_reshape, op_type=Reshape, name=TransposeEqualReshapePattern--B--GemmTransposePattern--MatMulAddPattern--Opset3, inputs=p_layers_2_weight::T10,init7_s2_1_-1
    [PatternOptimization.enumerate_matches] start ReshapeMatMulReshapePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start Reshape2Of3Pattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReshapeReshapeBinaryPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start MatMulAddPattern with main_opset=18 and min_opset=1
    [MatMulAddPattern.match] NONE - line: 58:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Gemm, name=GemmTransposePattern--MatMulAddPattern--Opset2, inputs=x,GemmTransposePattern--p_layers_0_weight::T10,layers.0.bias
    [MatMulAddPattern.match] NONE - line: 55:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Gemm, name=GemmTransposePattern--MatMulAddPattern--Opset32, inputs=relu,GemmTransposePattern--p_layers_2_weight::T10,layers.2.bias
    [PatternOptimization.enumerate_matches] start GemmTransposePattern with main_opset=18 and min_opset=1
    [GemmTransposePattern.match] NONE - line: 307:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Gemm, name=GemmTransposePattern--MatMulAddPattern--Opset2, inputs=x,GemmTransposePattern--p_layers_0_weight::T10,layers.0.bias
    [GemmTransposePattern.match] NONE - line: 307:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Gemm, name=GemmTransposePattern--MatMulAddPattern--Opset32, inputs=relu,GemmTransposePattern--p_layers_2_weight::T10,layers.2.bias
    [PatternOptimization.enumerate_matches] start MatMulReshape2Of3Pattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start MulMulMatMulPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedReshapeIsSqueezePattern with main_opset=18 and min_opset=1
    [ShapeBasedReshapeIsSqueezePattern.match] NONE - line: 977:experimental_experiment.xoptim.patterns.onnx_reshape, op_type=Reshape, name=TransposeEqualReshapePattern--B--GemmTransposePattern--MatMulAddPattern--Opset3, inputs=p_layers_2_weight::T10,init7_s2_1_-1
    [PatternOptimization.enumerate_matches] start ShapeBasedStaticExpandPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedConcatExpandPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedEditDistanceReshapePattern with main_opset=18 and min_opset=1
    [ShapeBasedEditDistanceReshapePattern.match] NONE - line: 886:experimental_experiment.xoptim.patterns.onnx_reshape, op_type=Reshape, name=TransposeEqualReshapePattern--B--GemmTransposePattern--MatMulAddPattern--Opset3, inputs=p_layers_2_weight::T10,init7_s2_1_-1
    [PatternOptimization.enumerate_matches] start ShapeBasedIdentityPattern with main_opset=18 and min_opset=1
    [ShapeBasedIdentityPattern.match] NONE - line: 401:experimental_experiment.xoptim.patterns.onnx_any, op_type=Transpose, name=GemmTransposePattern--MatMulAddPattern--Opset, inputs=p_layers_0_weight::T10
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandBroadcastPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandBroadcastMatMulPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandCastWhereSwapPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedExpandSwapPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedMatMulToMulPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedSameChildrenPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ShapeBasedShapeShapeAddPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start ReshapeReshapePattern with main_opset=18 and min_opset=1
    [ReshapeReshapePattern.match] NONE - line: 166:experimental_experiment.xoptim.patterns.onnx_reshape, op_type=Reshape, name=TransposeEqualReshapePattern--B--GemmTransposePattern--MatMulAddPattern--Opset3, inputs=p_layers_2_weight::T10,init7_s2_1_-1
    [PatternOptimization.enumerate_matches] start RotaryEmbeddingPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SameChildrenPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SequenceConstructAtPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SliceSlicePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SlicesSplitPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SoftmaxCrossEntropyLossCastPattern with main_opset=18 and min_opset=14
    [PatternOptimization.enumerate_matches] start SplitConcatPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SqueezeAddPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SqueezeUnsqueezePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start StaticConcatReshapePattern with main_opset=18 and min_opset=1
    [StaticConcatReshapePattern.match] NONE - line: 664:experimental_experiment.xoptim.patterns.onnx_reshape, op_type=Reshape, name=TransposeEqualReshapePattern--B--GemmTransposePattern--MatMulAddPattern--Opset3, inputs=p_layers_2_weight::T10,init7_s2_1_-1
    [PatternOptimization.enumerate_matches] start Sub1MulPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SwitchOrderBinaryPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start SwitchReshapeActivationPattern with main_opset=18 and min_opset=1
    [SwitchReshapeActivationPattern.match] NONE - line: 1178:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Relu, name=relu, inputs=linear
    [PatternOptimization.enumerate_matches] start TransposeEqualReshapePattern with main_opset=18 and min_opset=1
    [TransposeEqualReshapePattern.match] NONE - line: 342:experimental_experiment.xoptim.patterns.onnx_transpose, op_type=Transpose, name=GemmTransposePattern--MatMulAddPattern--Opset, inputs=p_layers_0_weight::T10
    [PatternOptimization.enumerate_matches] start TransposeMatMulPattern with main_opset=18 and min_opset=1
    [TransposeMatMulPattern.match] NONE - line: 928:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Gemm, name=GemmTransposePattern--MatMulAddPattern--Opset2, inputs=x,GemmTransposePattern--p_layers_0_weight::T10,layers.0.bias
    [TransposeMatMulPattern.match] NONE - line: 890:experimental_experiment.xoptim.patterns.onnx_matmul, op_type=Gemm, name=GemmTransposePattern--MatMulAddPattern--Opset32, inputs=relu,GemmTransposePattern--p_layers_2_weight::T10,layers.2.bias
    [PatternOptimization.enumerate_matches] start TransposeReshapeMatMulPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start TransposeReshapeTransposePattern with main_opset=18 and min_opset=1
    [TransposeReshapeTransposePattern.match] NONE - line: 140:experimental_experiment.xoptim.patterns.onnx_transpose, op_type=Transpose, name=GemmTransposePattern--MatMulAddPattern--Opset, inputs=p_layers_0_weight::T10
    [PatternOptimization.enumerate_matches] start TransposeTransposePattern with main_opset=18 and min_opset=1
    [TransposeTransposePattern.match] NONE - line: 51:experimental_experiment.xoptim.patterns.onnx_transpose, op_type=Transpose, name=GemmTransposePattern--MatMulAddPattern--Opset, inputs=p_layers_0_weight::T10
    [PatternOptimization.enumerate_matches] start UnsqueezeEqualPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start UnsqueezeUnsqueezePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start RotaryConcatPartPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start FunctionCausalMaskPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start FunctionCausalMaskMulAddPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start FunctionCosSinCachePattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start FunctionHalfRotaryEmbeddingPattern with main_opset=18 and min_opset=1
    [PatternOptimization.enumerate_matches] start RMSNormalizationPattern with main_opset=18 and min_opset=1
    [GraphBuilderPatternOptimization-VNK.optimize] done all: -0 +0 nodes
    [GraphBuilderPatternOptimization-VNK.optimize] stops current_priority_index=3, priorities=[0, 1, 3]
    [GraphBuilderPatternOptimization-VNK.optimize] done after 6 iterations with 5 nodes in 0.013
        STAT apply_GemmTransposePattern +4 -2 #it=1 maxmatch=1 i=2 - time=0.0006508419992314884
        STAT apply_MatMulAddPattern +2 -4 #it=1 maxmatch=1 i=2 - time=0.00040602700028102845
        STAT apply_TransposeEqualReshapePattern +1 -1 #it=1 maxmatch=0 i=1 - time=0.0006819099999120226
        STAT build_graph_for_pattern +0 -0 #it=6 maxmatch=0 i=0 - time=0.0002969959987240145
        STAT check_pattern_00 +0 -0 #it=1 maxmatch=0 i=0 - time=2.33369992201915e-05
        STAT check_pattern_A0 +0 -0 #it=3 maxmatch=0 i=0 - time=0.00020871200104011223
        STAT check_pattern_B0 +0 -0 #it=6 maxmatch=0 i=0 - time=0.0002637689995026449
        STAT match_BatchNormalizationPattern +0 -0 #it=6 maxmatch=0 i=0 - time=7.434100007230882e-05
        STAT match_BatchNormalizationTrainingPattern +0 -0 #it=6 maxmatch=0 i=0 - time=4.760699812322855e-05
        STAT match_CastCastBinaryPattern +0 -0 #it=5 maxmatch=0 i=0 - time=0.0001230689995281864
        STAT match_CastLayerNormalizationCastPattern +0 -0 #it=5 maxmatch=0 i=0 - time=3.8672000300721265e-05
        STAT match_CastOpCastPattern +0 -0 #it=5 maxmatch=0 i=0 - time=9.698199755803216e-05
        STAT match_CastPattern +0 -0 #it=6 maxmatch=0 i=0 - time=4.120499943383038e-05
        STAT match_ClipClipPattern +0 -0 #it=5 maxmatch=0 i=0 - time=3.61340007657418e-05
        STAT match_ComputationCastOpCastPattern +0 -0 #it=5 maxmatch=0 i=0 - time=6.632899930991698e-05
        STAT match_ConcatEmptyPattern +0 -0 #it=5 maxmatch=0 i=0 - time=3.260700214013923e-05
        STAT match_ConcatGatherPattern +0 -0 #it=6 maxmatch=0 i=0 - time=4.299100146454293e-05
        STAT match_ConcatReshapePattern +0 -0 #it=6 maxmatch=0 i=0 - time=5.633900218526833e-05
        STAT match_ConcatTwiceUnaryPattern +0 -0 #it=5 maxmatch=0 i=0 - time=3.7579002309939824e-05
        STAT match_ConvBiasNullPattern +0 -0 #it=6 maxmatch=0 i=0 - time=3.6948003980796784e-05
        STAT match_DropoutPattern +0 -0 #it=5 maxmatch=0 i=0 - time=3.2151998311746866e-05
        STAT match_ExpandBroadcastPattern +0 -0 #it=5 maxmatch=0 i=0 - time=3.119100074400194e-05
        STAT match_ExpandPattern +0 -0 #it=6 maxmatch=0 i=0 - time=3.537799602781888e-05
        STAT match_ExpandSwapPattern +0 -0 #it=5 maxmatch=0 i=0 - time=2.934900112450123e-05
        STAT match_FunctionCausalMaskMulAddPattern +0 -0 #it=5 maxmatch=2 i=0 - time=8.815600085654296e-05
        STAT match_FunctionCausalMaskPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.155100057483651e-05
        STAT match_FunctionCosSinCachePattern +0 -0 #it=5 maxmatch=2 i=0 - time=0.0001067709999915678
        STAT match_FunctionHalfRotaryEmbeddingPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.8892001612111926e-05
        STAT match_GeluPattern +0 -0 #it=6 maxmatch=0 i=0 - time=1.3644998034578748e-05
        STAT match_GemmTransposePattern +0 -0 #it=5 maxmatch=2 i=2 - time=0.0001653399995120708
        STAT match_IdentityPattern +0 -0 #it=6 maxmatch=0 i=0 - time=0.0005031620021327399
        STAT match_LayerNormalizationPattern +0 -0 #it=5 maxmatch=0 i=0 - time=3.892699896823615e-05
        STAT match_LayerNormalizationScalePattern +0 -0 #it=5 maxmatch=0 i=0 - time=3.306199869257398e-05
        STAT match_LeakyReluPattern +0 -0 #it=6 maxmatch=0 i=0 - time=0.0012129089955124073
        STAT match_MatMulAddPattern +0 -0 #it=4 maxmatch=2 i=2 - time=0.0002485609984432813
        STAT match_MatMulReshape2Of3Pattern +0 -0 #it=5 maxmatch=2 i=0 - time=0.00010799999836308416
        STAT match_MulMulMatMulPattern +0 -0 #it=5 maxmatch=2 i=0 - time=6.843599840067327e-05
        STAT match_MulMulMulScalarPattern +0 -0 #it=5 maxmatch=0 i=0 - time=6.651900002907496e-05
        STAT match_RMSNormalizationPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.467900023679249e-05
        STAT match_ReduceReshapePattern +0 -0 #it=5 maxmatch=0 i=0 - time=4.031100070278626e-05
        STAT match_ReduceSumNormalizePattern +0 -0 #it=5 maxmatch=0 i=0 - time=3.764500070246868e-05
        STAT match_Reshape2Of3Pattern +0 -0 #it=5 maxmatch=0 i=0 - time=0.0002217780020146165
        STAT match_ReshapeMatMulReshapePattern +0 -0 #it=5 maxmatch=0 i=0 - time=7.608400119352154e-05
        STAT match_ReshapePattern +0 -0 #it=6 maxmatch=0 i=0 - time=7.30459996702848e-05
        STAT match_ReshapeReshapeBinaryPattern +0 -0 #it=5 maxmatch=0 i=0 - time=7.328399988182355e-05
        STAT match_ReshapeReshapePattern +0 -0 #it=6 maxmatch=2 i=0 - time=4.748299943457823e-05
        STAT match_RotaryConcatPartPattern +0 -0 #it=5 maxmatch=2 i=0 - time=4.104999970877543e-05
        STAT match_RotaryEmbeddingPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.122300040558912e-05
        STAT match_SameChildrenPattern +0 -0 #it=6 maxmatch=2 i=0 - time=6.179799856909085e-05
        STAT match_SequenceConstructAtPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.66729982488323e-05
        STAT match_ShapeBasedConcatExpandPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.434999962337315e-05
        STAT match_ShapeBasedEditDistanceReshapePattern +0 -0 #it=6 maxmatch=2 i=0 - time=0.00011381599688320421
        STAT match_ShapeBasedExpandBroadcastMatMulPattern +0 -0 #it=5 maxmatch=2 i=0 - time=7.776400161674246e-05
        STAT match_ShapeBasedExpandBroadcastPattern +0 -0 #it=5 maxmatch=2 i=0 - time=8.976800017990172e-05
        STAT match_ShapeBasedExpandCastWhereSwapPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.33359985233983e-05
        STAT match_ShapeBasedExpandSwapPattern +0 -0 #it=5 maxmatch=2 i=0 - time=7.048799852782395e-05
        STAT match_ShapeBasedIdentityPattern +0 -0 #it=6 maxmatch=2 i=0 - time=6.35130018054042e-05
        STAT match_ShapeBasedMatMulToMulPattern +0 -0 #it=5 maxmatch=2 i=0 - time=7.140300112951081e-05
        STAT match_ShapeBasedReshapeIsSqueezePattern +0 -0 #it=6 maxmatch=2 i=0 - time=6.6841999796452e-05
        STAT match_ShapeBasedSameChildrenPattern +0 -0 #it=6 maxmatch=2 i=0 - time=4.0420003642793745e-05
        STAT match_ShapeBasedShapeShapeAddPattern +0 -0 #it=6 maxmatch=2 i=0 - time=8.378799975616857e-05
        STAT match_ShapeBasedStaticExpandPattern +0 -0 #it=6 maxmatch=2 i=0 - time=3.700900015246589e-05
        STAT match_SliceSlicePattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.178899896738585e-05
        STAT match_SlicesSplitPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.501399805827532e-05
        STAT match_SoftmaxCrossEntropyLossCastPattern +0 -0 #it=6 maxmatch=2 i=0 - time=0.001981257997613284
        STAT match_SplitConcatPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.38849986292189e-05
        STAT match_SqueezeAddPattern +0 -0 #it=6 maxmatch=2 i=0 - time=0.00011344200174789876
        STAT match_SqueezeUnsqueezePattern +0 -0 #it=6 maxmatch=2 i=0 - time=6.647400186921004e-05
        STAT match_StaticConcatReshapePattern +0 -0 #it=6 maxmatch=2 i=0 - time=4.6860999646014534e-05
        STAT match_Sub1MulPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.114200080744922e-05
        STAT match_SwitchOrderBinaryPattern +0 -0 #it=5 maxmatch=2 i=0 - time=5.62149998586392e-05
        STAT match_SwitchReshapeActivationPattern +0 -0 #it=5 maxmatch=2 i=0 - time=9.913300164043903e-05
        STAT match_TransposeEqualReshapePattern +0 -0 #it=5 maxmatch=2 i=1 - time=0.00012121800136810634
        STAT match_TransposeMatMulPattern +0 -0 #it=5 maxmatch=2 i=0 - time=0.0002322609998373082
        STAT match_TransposeReshapeMatMulPattern +0 -0 #it=5 maxmatch=2 i=0 - time=7.186100083345082e-05
        STAT match_TransposeReshapeTransposePattern +0 -0 #it=6 maxmatch=2 i=0 - time=6.346800182654988e-05
        STAT match_TransposeTransposePattern +0 -0 #it=6 maxmatch=2 i=0 - time=5.7784001910476945e-05
        STAT match_UnsqueezeEqualPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.106499934801832e-05
        STAT match_UnsqueezeUnsqueezePattern +0 -0 #it=6 maxmatch=2 i=0 - time=3.584799742384348e-05
        STAT remove_identity_nodes +0 -0 #it=3 maxmatch=0 i=0 - time=0.000595396999415243
        STAT remove_unused +0 -0 #it=6 maxmatch=0 i=0 - time=0.0009860719965217868
    --MODEL: 5 nodes, 1 inputs, 1 outputs, 5 initializers--
             INPUT:   1 x 1t
         INPUT-SEQ:   1 x Falset
            OUTPUT:   1 x 1t
        OUTPUT-SEQ:   1 x Falset
              INIT:   4 x 1t
              INIT:   1 x 7t
              NODE:   2 x Gemm
              NODE:   1 x Relu
              NODE:   1 x Reshape
              NODE:   1 x Transpose
    --MODEL: 5 nodes, 1 inputs, 1 outputs, 5 initializers--DETAILED--
         INPUT:   1 x 1t[3x10]
        OUTPUT:   1 x 1t[3x1]
          INIT:   1 x 1t[10x32]
          INIT:   1 x 1t[1]
          INIT:   1 x 1t[32]
          INIT:   1 x 1t[32x1]
          INIT:   1 x 7t[2]
          NODE:   1 x Gemm -SIG- 1t[3x10], 1t[32x10], 1t[32]
          NODE:   1 x Gemm -SIG- 1t[3x32], 1t[1x32], 1t[1]
          NODE:   1 x Relu -SIG- 1t[3x32]
          NODE:   1 x Reshape -SIG- 1t[32x1], 7t[2]
          NODE:   1 x Transpose -SIG- 1t[10x32]-perm=1;0
    [GraphBuilder-VNK.remove_identity_nodes] -- starts with 5
    [GraphBuilder-VNK.remove_identity_nodes] found 0 replacements
    [GraphBuilder-VNK.remove_identity_nodes] kept 5 nodes
    [GraphBuilder-VNK.remove_identity_nodes] ends with 5 nodes in 3.849400127364788e-05 seconds
    [GraphBuilder-VNK.constant_folding] -- starts with 7 constants and 5 nodes.
    [GraphBuilder-VNK.constant_folding] cst:: 1 :: layers.0.bias
    [GraphBuilder-VNK.constant_folding] cst:: 1 :: GemmTransposePattern--p_layers_2_weight::T10
    [GraphBuilder-VNK.constant_folding] cst:: . :: _onx_matmul_x
    [GraphBuilder-VNK.constant_folding] cst:: . :: x
    [GraphBuilder-VNK.constant_folding] cst:: . :: relu
    [GraphBuilder-VNK.constant_folding] cst:: . :: output_0
    [GraphBuilder-VNK.constant_folding] cst:: 1 :: layers.2.bias
    [GraphBuilder-VNK.constant_folding] cst:: 1 :: init7_s2_1_-1
    [GraphBuilder-VNK.constant_folding] cst:: . :: linear
    [GraphBuilder-VNK.constant_folding] cst:: . :: _onx_matmul_relu
    [GraphBuilder-VNK.constant_folding] cst:: 1 :: GemmTransposePattern--p_layers_0_weight::T10
    [GraphBuilder-VNK.constant_folding] cst:: 1 :: p_layers_2_weight::T10
    [GraphBuilder-VNK.constant_folding] cst:: 1 :: p_layers_0_weight::T10
    [GraphBuilder-VNK.constant_folding] initializer: p_layers_0_weight::T10
    [GraphBuilder-VNK.constant_folding] initializer: p_layers_2_weight::T10
    [GraphBuilder-VNK.constant_folding] initializer: layers.0.bias
    [GraphBuilder-VNK.constant_folding] initializer: layers.2.bias
    [GraphBuilder-VNK.constant_folding] from: Transpose(GemmTransposePattern--p_layers_0_weight::T10)
    [GraphBuilder-VNK.set_type] GemmTransposePattern--p_layers_0_weight::T10:1
    [GraphBuilder-VNK.make_initializer] GemmTransposePattern--p_layers_0_weight::T10[1:(32, 10)]
    [GraphBuilder-VNK.update_node_constant] new constant 'GemmTransposePattern--p_layers_0_weight::T10', node=None
    [GraphBuilder-VNK.constant_folding] fold_constant:Transpose:GemmTransposePattern--p_layers_0_weight::T10[torch.float32:torch.Size([32, 10])]:from:p_layers_0_weight::T10
    [GraphBuilder-VNK.constant_folding] from: Reshape(GemmTransposePattern--p_layers_2_weight::T10)
    [GraphBuilder-VNK.set_type] GemmTransposePattern--p_layers_2_weight::T10:1
    [GraphBuilder-VNK.make_initializer] GemmTransposePattern--p_layers_2_weight::T10[1:(1, 32)]
    [GraphBuilder-VNK.update_node_constant] new constant 'GemmTransposePattern--p_layers_2_weight::T10', node=None
    [GraphBuilder-VNK.constant_folding] fold_constant:Reshape:GemmTransposePattern--p_layers_2_weight::T10[float32:(1, 32)]:from:init7_s2_1_-1,p_layers_2_weight::T10
    [GraphBuilder-VNK.constant_folding] initializer: init7_s2_1_-1
    [GraphBuilder-VNK.update_node_constant] new constant 'GemmTransposePattern--p_layers_0_weight::T10', node=None
    [GraphBuilder-VNK.update_node_constant] new constant 'GemmTransposePattern--p_layers_2_weight::T10', node=None
    [GraphBuilder-VNK.constant_folding] ends with 7 constants and 3 nodes in 0.0006285919989750255 seconds
    [GraphBuilder-VNK.remove_unused] remove_initializer 1:0/7:p_layers_0_weight::T10
    [GraphBuilder-VNK.remove_unused] remove_initializer 2:1/7:p_layers_2_weight::T10
    [GraphBuilder-VNK.remove_unused] remove_initializer 3:4/7:init7_s2_1_-1:int64[(2,)]
    [GraphBuilder-VNK.optimize] done with 3 nodes in 0.017
        STAT apply_GemmTransposePattern +4 -2 #it=1 maxmatch=1 i=2 - time=0.0006508419992314884
        STAT apply_MatMulAddPattern +2 -4 #it=1 maxmatch=1 i=2 - time=0.00040602700028102845
        STAT apply_TransposeEqualReshapePattern +1 -1 #it=1 maxmatch=0 i=1 - time=0.0006819099999120226
        STAT apply_constant_folding +0 -2 #it=2 maxmatch=0 i=0 - time=0.0008475449994875817
        STAT apply_constant_folding__Reshape +0 -0 #it=1 maxmatch=0 i=0 - time=0.0
        STAT apply_constant_folding__Transpose +0 -0 #it=1 maxmatch=0 i=0 - time=0.0
        STAT apply_constant_folding_new_inits +0 -0 #it=2 maxmatch=0 i=0 - time=0.0
        STAT build_graph_for_pattern +0 -0 #it=6 maxmatch=0 i=0 - time=0.0002969959987240145
        STAT check_A-dynamic_dimension_naming +0 -0 #it=0 maxmatch=0 i=0 - time=2.2875001377542503e-05
        STAT check_A-opt-sub +0 -0 #it=0 maxmatch=0 i=0 - time=1.86079996638e-05
        STAT check_B-remove-identity +0 -0 #it=0 maxmatch=0 i=0 - time=2.200900053139776e-05
        STAT check_C-remove-unused +0 -0 #it=0 maxmatch=0 i=0 - time=1.678500120760873e-05
        STAT check_Da-constant-folding +0 -0 #it=0 maxmatch=0 i=0 - time=1.5906000044196844e-05
        STAT check_Db-constant-folding +0 -0 #it=0 maxmatch=0 i=0 - time=2.3223999960464425e-05
        STAT check_Ea-remove-unused +0 -0 #it=0 maxmatch=0 i=0 - time=1.5442999938386492e-05
        STAT check_Eb-remove-unused +0 -0 #it=0 maxmatch=0 i=0 - time=1.9225999494665302e-05
        STAT check_F-patterns +0 -0 #it=0 maxmatch=0 i=0 - time=3.019500036316458e-05
        STAT check_G-remove-identity +0 -0 #it=0 maxmatch=0 i=0 - time=2.3991000489331782e-05
        STAT check_G-remove-unused +0 -0 #it=0 maxmatch=0 i=0 - time=2.622499960125424e-05
        STAT check_H-remove-duplicated-initializer +0 -0 #it=0 maxmatch=0 i=0 - time=1.5530999007751234e-05
        STAT check_H-remove-unused +0 -0 #it=0 maxmatch=0 i=0 - time=1.650000012887176e-05
        STAT check_pattern_00 +0 -0 #it=1 maxmatch=0 i=0 - time=2.33369992201915e-05
        STAT check_pattern_A0 +0 -0 #it=3 maxmatch=0 i=0 - time=0.00020871200104011223
        STAT check_pattern_B0 +0 -0 #it=6 maxmatch=0 i=0 - time=0.0002637689995026449
        STAT dynamic_dimension_naming +0 -0 #it=0 maxmatch=0 i=0 - time=2.8393000320647843e-05
        STAT match_BatchNormalizationPattern +0 -0 #it=6 maxmatch=0 i=0 - time=7.434100007230882e-05
        STAT match_BatchNormalizationTrainingPattern +0 -0 #it=6 maxmatch=0 i=0 - time=4.760699812322855e-05
        STAT match_CastCastBinaryPattern +0 -0 #it=5 maxmatch=0 i=0 - time=0.0001230689995281864
        STAT match_CastLayerNormalizationCastPattern +0 -0 #it=5 maxmatch=0 i=0 - time=3.8672000300721265e-05
        STAT match_CastOpCastPattern +0 -0 #it=5 maxmatch=0 i=0 - time=9.698199755803216e-05
        STAT match_CastPattern +0 -0 #it=6 maxmatch=0 i=0 - time=4.120499943383038e-05
        STAT match_ClipClipPattern +0 -0 #it=5 maxmatch=0 i=0 - time=3.61340007657418e-05
        STAT match_ComputationCastOpCastPattern +0 -0 #it=5 maxmatch=0 i=0 - time=6.632899930991698e-05
        STAT match_ConcatEmptyPattern +0 -0 #it=5 maxmatch=0 i=0 - time=3.260700214013923e-05
        STAT match_ConcatGatherPattern +0 -0 #it=6 maxmatch=0 i=0 - time=4.299100146454293e-05
        STAT match_ConcatReshapePattern +0 -0 #it=6 maxmatch=0 i=0 - time=5.633900218526833e-05
        STAT match_ConcatTwiceUnaryPattern +0 -0 #it=5 maxmatch=0 i=0 - time=3.7579002309939824e-05
        STAT match_ConvBiasNullPattern +0 -0 #it=6 maxmatch=0 i=0 - time=3.6948003980796784e-05
        STAT match_DropoutPattern +0 -0 #it=5 maxmatch=0 i=0 - time=3.2151998311746866e-05
        STAT match_ExpandBroadcastPattern +0 -0 #it=5 maxmatch=0 i=0 - time=3.119100074400194e-05
        STAT match_ExpandPattern +0 -0 #it=6 maxmatch=0 i=0 - time=3.537799602781888e-05
        STAT match_ExpandSwapPattern +0 -0 #it=5 maxmatch=0 i=0 - time=2.934900112450123e-05
        STAT match_FunctionCausalMaskMulAddPattern +0 -0 #it=5 maxmatch=2 i=0 - time=8.815600085654296e-05
        STAT match_FunctionCausalMaskPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.155100057483651e-05
        STAT match_FunctionCosSinCachePattern +0 -0 #it=5 maxmatch=2 i=0 - time=0.0001067709999915678
        STAT match_FunctionHalfRotaryEmbeddingPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.8892001612111926e-05
        STAT match_GeluPattern +0 -0 #it=6 maxmatch=0 i=0 - time=1.3644998034578748e-05
        STAT match_GemmTransposePattern +0 -0 #it=5 maxmatch=2 i=2 - time=0.0001653399995120708
        STAT match_IdentityPattern +0 -0 #it=6 maxmatch=0 i=0 - time=0.0005031620021327399
        STAT match_LayerNormalizationPattern +0 -0 #it=5 maxmatch=0 i=0 - time=3.892699896823615e-05
        STAT match_LayerNormalizationScalePattern +0 -0 #it=5 maxmatch=0 i=0 - time=3.306199869257398e-05
        STAT match_LeakyReluPattern +0 -0 #it=6 maxmatch=0 i=0 - time=0.0012129089955124073
        STAT match_MatMulAddPattern +0 -0 #it=4 maxmatch=2 i=2 - time=0.0002485609984432813
        STAT match_MatMulReshape2Of3Pattern +0 -0 #it=5 maxmatch=2 i=0 - time=0.00010799999836308416
        STAT match_MulMulMatMulPattern +0 -0 #it=5 maxmatch=2 i=0 - time=6.843599840067327e-05
        STAT match_MulMulMulScalarPattern +0 -0 #it=5 maxmatch=0 i=0 - time=6.651900002907496e-05
        STAT match_RMSNormalizationPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.467900023679249e-05
        STAT match_ReduceReshapePattern +0 -0 #it=5 maxmatch=0 i=0 - time=4.031100070278626e-05
        STAT match_ReduceSumNormalizePattern +0 -0 #it=5 maxmatch=0 i=0 - time=3.764500070246868e-05
        STAT match_Reshape2Of3Pattern +0 -0 #it=5 maxmatch=0 i=0 - time=0.0002217780020146165
        STAT match_ReshapeMatMulReshapePattern +0 -0 #it=5 maxmatch=0 i=0 - time=7.608400119352154e-05
        STAT match_ReshapePattern +0 -0 #it=6 maxmatch=0 i=0 - time=7.30459996702848e-05
        STAT match_ReshapeReshapeBinaryPattern +0 -0 #it=5 maxmatch=0 i=0 - time=7.328399988182355e-05
        STAT match_ReshapeReshapePattern +0 -0 #it=6 maxmatch=2 i=0 - time=4.748299943457823e-05
        STAT match_RotaryConcatPartPattern +0 -0 #it=5 maxmatch=2 i=0 - time=4.104999970877543e-05
        STAT match_RotaryEmbeddingPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.122300040558912e-05
        STAT match_SameChildrenPattern +0 -0 #it=6 maxmatch=2 i=0 - time=6.179799856909085e-05
        STAT match_SequenceConstructAtPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.66729982488323e-05
        STAT match_ShapeBasedConcatExpandPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.434999962337315e-05
        STAT match_ShapeBasedEditDistanceReshapePattern +0 -0 #it=6 maxmatch=2 i=0 - time=0.00011381599688320421
        STAT match_ShapeBasedExpandBroadcastMatMulPattern +0 -0 #it=5 maxmatch=2 i=0 - time=7.776400161674246e-05
        STAT match_ShapeBasedExpandBroadcastPattern +0 -0 #it=5 maxmatch=2 i=0 - time=8.976800017990172e-05
        STAT match_ShapeBasedExpandCastWhereSwapPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.33359985233983e-05
        STAT match_ShapeBasedExpandSwapPattern +0 -0 #it=5 maxmatch=2 i=0 - time=7.048799852782395e-05
        STAT match_ShapeBasedIdentityPattern +0 -0 #it=6 maxmatch=2 i=0 - time=6.35130018054042e-05
        STAT match_ShapeBasedMatMulToMulPattern +0 -0 #it=5 maxmatch=2 i=0 - time=7.140300112951081e-05
        STAT match_ShapeBasedReshapeIsSqueezePattern +0 -0 #it=6 maxmatch=2 i=0 - time=6.6841999796452e-05
        STAT match_ShapeBasedSameChildrenPattern +0 -0 #it=6 maxmatch=2 i=0 - time=4.0420003642793745e-05
        STAT match_ShapeBasedShapeShapeAddPattern +0 -0 #it=6 maxmatch=2 i=0 - time=8.378799975616857e-05
        STAT match_ShapeBasedStaticExpandPattern +0 -0 #it=6 maxmatch=2 i=0 - time=3.700900015246589e-05
        STAT match_SliceSlicePattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.178899896738585e-05
        STAT match_SlicesSplitPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.501399805827532e-05
        STAT match_SoftmaxCrossEntropyLossCastPattern +0 -0 #it=6 maxmatch=2 i=0 - time=0.001981257997613284
        STAT match_SplitConcatPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.38849986292189e-05
        STAT match_SqueezeAddPattern +0 -0 #it=6 maxmatch=2 i=0 - time=0.00011344200174789876
        STAT match_SqueezeUnsqueezePattern +0 -0 #it=6 maxmatch=2 i=0 - time=6.647400186921004e-05
        STAT match_StaticConcatReshapePattern +0 -0 #it=6 maxmatch=2 i=0 - time=4.6860999646014534e-05
        STAT match_Sub1MulPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.114200080744922e-05
        STAT match_SwitchOrderBinaryPattern +0 -0 #it=5 maxmatch=2 i=0 - time=5.62149998586392e-05
        STAT match_SwitchReshapeActivationPattern +0 -0 #it=5 maxmatch=2 i=0 - time=9.913300164043903e-05
        STAT match_TransposeEqualReshapePattern +0 -0 #it=5 maxmatch=2 i=1 - time=0.00012121800136810634
        STAT match_TransposeMatMulPattern +0 -0 #it=5 maxmatch=2 i=0 - time=0.0002322609998373082
        STAT match_TransposeReshapeMatMulPattern +0 -0 #it=5 maxmatch=2 i=0 - time=7.186100083345082e-05
        STAT match_TransposeReshapeTransposePattern +0 -0 #it=6 maxmatch=2 i=0 - time=6.346800182654988e-05
        STAT match_TransposeTransposePattern +0 -0 #it=6 maxmatch=2 i=0 - time=5.7784001910476945e-05
        STAT match_UnsqueezeEqualPattern +0 -0 #it=5 maxmatch=2 i=0 - time=3.106499934801832e-05
        STAT match_UnsqueezeUnsqueezePattern +0 -0 #it=6 maxmatch=2 i=0 - time=3.584799742384348e-05
        STAT pattern_optimization +0 -0 #it=0 maxmatch=0 i=0 - time=0.014893425999616738
        STAT remove_duplicated_initializer +0 -0 #it=0 maxmatch=0 i=0 - time=4.3788999391836114e-05
        STAT remove_identity_nodes +0 -0 #it=3 maxmatch=0 i=0 - time=0.0009794929992494872
        STAT remove_unused +0 -0 #it=6 maxmatch=0 i=0 - time=0.0015925719944789307
    --MODEL: 3 nodes, 1 inputs, 1 outputs, 4 initializers--
             INPUT:   1 x 1t
         INPUT-SEQ:   1 x Falset
            OUTPUT:   1 x 1t
        OUTPUT-SEQ:   1 x Falset
              INIT:   4 x 1t
              NODE:   2 x Gemm
              NODE:   1 x Relu
    --MODEL: 3 nodes, 1 inputs, 1 outputs, 4 initializers--DETAILED--
         INPUT:   1 x 1t[3x10]
        OUTPUT:   1 x 1t[3x1]
          INIT:   1 x 1t[1]
          INIT:   1 x 1t[1x32]
          INIT:   1 x 1t[32]
          INIT:   1 x 1t[32x10]
          NODE:   1 x Gemm -SIG- 1t[3x10], 1t[32x10], 1t[32]
          NODE:   1 x Gemm -SIG- 1t[3x32], 1t[1x32], 1t[1]
          NODE:   1 x Relu -SIG- 1t[3x32]
    [GraphBuilder-VNK.to_onnx] make_model 4 inits 0 params
    [GraphBuilder-VNK.time_evaluation_constants_] 0
    [GraphBuilder-VNK._build_initializers] start with 4 initializers, large_model=False, external_threshold=1024
    [GraphBuilder-VNK._build_initializers] switch low/high order
    [GraphBuilder-VNK._build_initializers] TensorProto-layers.0.bias:1[(32,)]
    [GraphBuilder-VNK._build_initializers] TensorProto-layers.2.bias:1[(1,)]
    [GraphBuilder-VNK._build_initializers] <Tensor>-GemmTransposePattern--p_layers_0_weight::T10:torch.float32[torch.Size([32, 10])]
    [proto_from_array] 1[torch.Size([32, 10])]
    [GraphBuilder-VNK._build_initializers] <ndarray>-GemmTransposePattern--p_layers_2_weight::T10:float32[(1, 32)]
    [GraphBuilder-VNK._build_initializers] done in 3.3670003176666796e-06s with 4 initializers, 0 large initializers
    [GraphBuilder-VNK._add_shape_information] dynamic shapes replacements={}

Select the pattern to use

Class OptimizationOptions is used to enable or disable patterns.

<<<

import onnx
from experimental_experiment.xbuilder import GraphBuilder, OptimizationOptions

onx = onnx.load("temp_doc_mlp.onnx")

gr = GraphBuilder(
    onx,
    infer_shapes_options=True,
    optimization_options=OptimizationOptions(
        patterns="TransposeTranspose,TransposeMatMul", verbose=1
    ),
)
opt_onx = gr.to_onnx(optimize=True)

>>>

    [GraphBuilder-LXO.optimize] start with 5 nodes
    [GraphBuilder-LXO.optimize] #patterns=2
    [GraphBuilderPatternOptimization-LXO.optimize] start with 5 nodes, 4 initializers, 2 patterns, priorities=[0, 1], max_iter=20
    [GraphBuilderPatternOptimization-LXO.optimize] iteration 0: 5 nodes, priority=0
    [GraphBuilderPatternOptimization-LXO.optimize] increase priority to 1
    [GraphBuilderPatternOptimization-LXO.optimize] iteration 1: 5 nodes, priority=1
    [GraphBuilderPatternOptimization-LXO.optimize] stops current_priority_index=2, priorities=[0, 1]
    [GraphBuilderPatternOptimization-LXO.optimize] done after 2 iterations with 5 nodes in 0.001
    [GraphBuilder-LXO.optimize] done with 5 nodes in 0.002

There exists some predefined lists of patterns:

  • default: includes all patterns using only standard onnx patterns.

  • onnxruntime: patterns specific to onnxruntime, the final model may be executed by onnxruntime and possibly only onnxruntime as it may introduce patterns from Supported Operators and Data Types.

<<<

import onnx
from experimental_experiment.xbuilder import GraphBuilder, OptimizationOptions

onx = onnx.load("temp_doc_mlp.onnx")

gr = GraphBuilder(
    onx,
    infer_shapes_options=True,
    optimization_options=OptimizationOptions(patterns="default+onnxruntime", verbose=1),
)
opt_onx = gr.to_onnx(optimize=True)

>>>

    [GraphBuilder-ISM.optimize] start with 5 nodes
    [GraphBuilder-ISM.optimize] #patterns=92
    [GraphBuilderPatternOptimization-ISM.optimize] start with 5 nodes, 4 initializers, 92 patterns, priorities=[0, 1, 2, 3], max_iter=40
    [GraphBuilderPatternOptimization-ISM.optimize] iteration 0: 5 nodes, priority=0
    [GraphBuilderPatternOptimization-ISM.optimize] increase priority to 1
    [GraphBuilderPatternOptimization-ISM.optimize] iteration 1: 5 nodes, priority=1
    [GraphBuilderPatternOptimization-ISM.optimize] increase priority to 2
    [GraphBuilderPatternOptimization-ISM.optimize] iteration 2: 5 nodes, priority=2
    [GraphBuilderPatternOptimization-ISM.optimize] increase priority to 3
    [GraphBuilderPatternOptimization-ISM.optimize] iteration 3: 5 nodes, priority=3
    [GraphBuilderPatternOptimization-ISM.optimize] applies 2 matches, 2*MatMulAddPattern - time=0.001 | max_time=IdentityPattern:0.000
    [GraphBuilderPatternOptimization-ISM.optimize] iteration 4: 3 nodes, priority=3
    [GraphBuilderPatternOptimization-ISM.optimize] applies 2 matches, 2*GemmTransposePattern - time=0.001 | max_time=MulMulMatMulPattern:0.000
    [GraphBuilderPatternOptimization-ISM.optimize] iteration 5: 5 nodes, priority=3
    [GraphBuilderPatternOptimization-ISM.optimize] applies 1 matches, [0]=MatchResult: TransposeEqualReshapePattern replaces ['Transpose'] - time=0.001 | max_time=TransposeMatMulPattern:0.000
    [GraphBuilderPatternOptimization-ISM.optimize] iteration 6: 5 nodes, priority=3
    [GraphBuilderPatternOptimization-ISM.optimize] stops current_priority_index=4, priorities=[0, 1, 2, 3]
    [GraphBuilderPatternOptimization-ISM.optimize] done after 7 iterations with 5 nodes in 0.016
    [GraphBuilder-ISM.optimize] done with 3 nodes in 0.018

Statistics

This can be used to see when a pattern is applied and how long it takes.

<<<

import pandas
import onnx
from experimental_experiment.xbuilder import GraphBuilder, OptimizationOptions

onx = onnx.load("temp_doc_mlp.onnx")

gr = GraphBuilder(
    onx,
    infer_shapes_options=True,
    optimization_options=OptimizationOptions(patterns="default"),
)
stat = gr.optimize()

print(pandas.DataFrame(stat))

>>>

                                       pattern  removed  added   time_in  iteration  value  instances  match_index
    0                 dynamic_dimension_naming      0.0    0.0  0.000018        NaN    NaN        NaN          NaN
    1         check_A-dynamic_dimension_naming      NaN    NaN  0.000017        NaN    NaN        NaN          NaN
    2                          check_A-opt-sub      NaN    NaN  0.000011        NaN    NaN        NaN          NaN
    3                    remove_identity_nodes      0.0    0.0  0.000083        NaN    NaN        NaN          NaN
    4                  check_B-remove-identity      NaN    NaN  0.000013        NaN    NaN        NaN          NaN
    ..                                     ...      ...    ...       ...        ...    ...        ...          ...
    445          remove_duplicated_initializer      0.0    0.0  0.000022        NaN    NaN        NaN          NaN
    446  check_H-remove-duplicated-initializer      NaN    NaN  0.000008        NaN    NaN        NaN          NaN
    447                  remove_identity_nodes      0.0    0.0  0.000031        NaN    NaN        NaN          NaN
    448                          remove_unused      0.0    NaN  0.000036        NaN    NaN        NaN          NaN
    449                  check_H-remove-unused      NaN    NaN  0.000009        NaN    NaN        NaN          NaN
    
    [450 rows x 8 columns]

It can be aggregated:

<<<

import pandas
import onnx
from experimental_experiment.xbuilder import GraphBuilder, OptimizationOptions

onx = onnx.load("temp_doc_mlp.onnx")

gr = GraphBuilder(
    onx,
    infer_shapes_options=True,
    optimization_options=OptimizationOptions(patterns="default"),
)
stat = gr.optimize()

df = pandas.DataFrame(stat)
for c in df.columns:
    if "time" not in c and "pattern" not in c:
        df[c] = df[c].fillna(0).astype(int)
aggs = {
    "time_in": "sum",
    "added": "sum",
    "removed": "sum",
    "iteration": "max",
    "match_index": "max",
    "instances": "sum",
}
print(df.groupby("pattern").agg(aggs))

>>>

                                         time_in  added  removed  iteration  match_index  instances
    pattern                                                                                        
    apply_GemmTransposePattern          0.000326      4        2          3            1          2
    apply_MatMulAddPattern              0.000239      2        4          2            1          2
    apply_TransposeEqualReshapePattern  0.000361      1        1          4            0          1
    apply_constant_folding              0.000512      0        2          1            0          0
    apply_constant_folding__Reshape     0.000000      0        0          1            0          0
    ...                                      ...    ...      ...        ...          ...        ...
    match_UnsqueezeUnsqueezePattern     0.000023      0        0          5            2          0
    pattern_optimization                0.011033      0        0          0            0          0
    remove_duplicated_initializer       0.000045      0        0          0            0          0
    remove_identity_nodes               0.000763      0        0          2            0          0
    remove_unused                       0.001222      0        0          5            0          0
    
    [101 rows x 6 columns]

Shape inference

The optimizers require to know the shapes to ensure they can rewrite some nodes and avoid producing a model which does not return the same results. If it is missing, some patterns cannot match for sure and they will not match.

This information can be built by running shape inference on the onnx models. That’s what is done is the previous examples. However, the best case is when this information comes from torch.

Function to_onnx converts a torch model into ONNX. While doing so, it stores the shape information coming from torch. There is no need to run shape inference on the onnx model it generates before optimizing it.

Available Patterns and API

All patterns may be found at .xoptim.patterns and .xoptim.patterns_ort.

When writing a pattern, walking along the graph or checking the shape is very common. Class GraphBuilderPatternOptimization provides the following methods.

Opsets

Patterns must rewrite using the nodes of the opset defined in the model.

Shapes, Types

Constants

  • is_constant: tells if a node is a constant (it may be a constant, an initializer or any value built on other constants)

  • is_constant_scalar: checks a constant is a scalar and compares its value to a number

  • get_computed_constant: returns the constant, computes it is a constant built from other constants

  • get_attribute: returns an attribute of a node

Graph

Nodes

  • make_node: creates a node without adding it to the graph

  • make_node_check_opset: creates a node without adding it to the graph, deals with some constraints related to opset version

Examples or Tools