ArmNN
 25.02
All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Pages
LstmLayer.cpp
Go to the documentation of this file.
1 //
2 // Copyright © 2017-2024 Arm Ltd and Contributors. All rights reserved.
3 // SPDX-License-Identifier: MIT
4 //
5 #include "LstmLayer.hpp"
6 
7 #include "LayerCloneBase.hpp"
8 
9 #include <armnn/LstmParams.hpp>
10 #include <armnn/TypesUtils.hpp>
13 
14 namespace armnn
15 {
16 
17 LstmLayer::LstmLayer(const LstmDescriptor& param, const char* name)
18  : LayerWithParameters(3, 4, LayerType::Lstm, param, name)
19 {
20 }
21 
22 std::unique_ptr<IWorkload> LstmLayer::CreateWorkload(const IWorkloadFactory& factory) const
23 {
24  LstmQueueDescriptor descriptor;
25 
26  // Basic parameters
34  descriptor.m_CellBias = m_BasicParameters.m_CellBias.get();
36 
37  // Cifg parameters
39  {
43  }
44 
45  // Projection parameters
47  {
50  }
51 
52  // Peephole parameters
54  {
56  {
58  }
61  }
62 
63  // Layer normalisation parameters
65  {
67  {
69  }
73  }
74 
75  SetAdditionalInfo(descriptor);
76 
77  return factory.CreateWorkload(LayerType::Lstm, descriptor, PrepInfoAndDesc(descriptor));
78 }
79 
81 {
82  auto layer = CloneBase<LstmLayer>(graph, m_Param, GetName());
83 
84  layer->m_BasicParameters.m_InputToForgetWeights = m_BasicParameters.m_InputToForgetWeights ?
86  : nullptr;
87  layer->m_BasicParameters.m_InputToCellWeights = m_BasicParameters.m_InputToCellWeights ?
89  layer->m_BasicParameters.m_InputToOutputWeights = m_BasicParameters.m_InputToOutputWeights ?
91  layer->m_BasicParameters.m_RecurrentToForgetWeights = m_BasicParameters.m_RecurrentToForgetWeights ?
93  layer->m_BasicParameters.m_RecurrentToCellWeights = m_BasicParameters.m_RecurrentToCellWeights ?
95  layer->m_BasicParameters.m_RecurrentToOutputWeights = m_BasicParameters.m_RecurrentToOutputWeights ?
97  layer->m_BasicParameters.m_ForgetGateBias = m_BasicParameters.m_ForgetGateBias ?
99  layer->m_BasicParameters.m_CellBias = m_BasicParameters.m_CellBias ?
100  m_BasicParameters.m_CellBias : nullptr;
101  layer->m_BasicParameters.m_OutputGateBias = m_BasicParameters.m_OutputGateBias ?
103 
104  if (!m_Param.m_CifgEnabled)
105  {
106  layer->m_CifgParameters.m_InputToInputWeights = m_CifgParameters.m_InputToInputWeights ?
108  layer->m_CifgParameters.m_RecurrentToInputWeights = m_CifgParameters.m_RecurrentToInputWeights ?
110  layer->m_CifgParameters.m_InputGateBias = m_CifgParameters.m_InputGateBias ?
112  }
113 
115  {
116  layer->m_ProjectionParameters.m_ProjectionWeights = m_ProjectionParameters.m_ProjectionWeights ?
118  layer->m_ProjectionParameters.m_ProjectionBias = m_ProjectionParameters.m_ProjectionBias ?
120  }
121 
123  {
124  if (!m_Param.m_CifgEnabled)
125  {
126  layer->m_PeepholeParameters.m_CellToInputWeights = m_PeepholeParameters.m_CellToInputWeights ?
128  }
129  layer->m_PeepholeParameters.m_CellToForgetWeights = m_PeepholeParameters.m_CellToForgetWeights ?
131  layer->m_PeepholeParameters.m_CellToOutputWeights = m_PeepholeParameters.m_CellToOutputWeights ?
133  }
134 
136  {
137  layer->m_LayerNormParameters.m_InputLayerNormWeights = m_LayerNormParameters.m_InputLayerNormWeights ?
139  layer->m_LayerNormParameters.m_ForgetLayerNormWeights = m_LayerNormParameters.m_ForgetLayerNormWeights ?
141  layer->m_LayerNormParameters.m_CellLayerNormWeights = m_LayerNormParameters.m_CellLayerNormWeights ?
143  layer->m_LayerNormParameters.m_OutputLayerNormWeights = m_LayerNormParameters.m_OutputLayerNormWeights ?
145  }
146 
147  return std::move(layer);
148 }
149 
150 std::vector<TensorShape> LstmLayer::InferOutputShapes(const std::vector<TensorShape>& inputShapes) const
151 {
152  if (inputShapes.size() != 3)
153  {
154  throw armnn::Exception("inputShapes' size is \"" + std::to_string(inputShapes.size()) +
155  "\" - should be \"3\".");
156  }
157 
158  // Get input values for validation
159  unsigned int batchSize = inputShapes[0][0];
160  unsigned int outputSize = inputShapes[1][1];
161  unsigned int numUnits = inputShapes[2][1];
162 
163  std::vector<TensorShape> outShapes;
164  outShapes.push_back(TensorShape({batchSize, numUnits * (m_Param.m_CifgEnabled ? 3 : 4)}));
165  outShapes.push_back(TensorShape({batchSize, outputSize}));
166  outShapes.push_back(TensorShape({batchSize, numUnits}));
167  outShapes.push_back(TensorShape({batchSize, outputSize}));
168 
169  return outShapes;
170 }
171 
173 {
175 
176  const TensorShape& outputShape = GetOutputSlot(0).GetTensorInfo().GetShape();
177 
179 
180  auto inferredShapes = InferOutputShapes( {
184  });
185 
186  if (inferredShapes.size() != 4)
187  {
188  throw armnn::Exception("inferredShapes has "
189  + std::to_string(inferredShapes.size()) +
190  " element(s) - should only have 4.");
191  }
192 
193  // Check if the weights are nullptr
195  {
196  throw armnn::NullPointerException("LstmLayer: "
197  "m_BasicParameters.m_InputToForgetWeights should not be null.");
198  }
199 
201  {
202  throw armnn::NullPointerException("LstmLayer: "
203  "m_BasicParameters.m_InputToCellWeights should not be null.");
204  }
205 
207  {
208  throw armnn::NullPointerException("LstmLayer: "
209  "m_BasicParameters.m_InputToOutputWeights should not be null.");
210  }
211 
213  {
214  throw armnn::NullPointerException("LstmLayer: "
215  "m_BasicParameters.m_RecurrentToForgetWeights should not be null.");
216  }
217 
219  {
220  throw armnn::NullPointerException("LstmLayer: "
221  "m_BasicParameters.m_RecurrentToCellWeights should not be null.");
222  }
223 
225  {
226  throw armnn::NullPointerException("LstmLayer: "
227  "m_BasicParameters.m_RecurrentToOutputWeights should not be null.");
228  }
229 
231  {
232  throw armnn::NullPointerException("LstmLayer: "
233  "m_BasicParameters.m_ForgetGateBias should not be null.");
234  }
235 
237  {
238  throw armnn::NullPointerException("LstmLayer: "
239  "m_BasicParameters.m_CellBias should not be null.");
240  }
241 
243  {
244  throw armnn::NullPointerException("LstmLayer: "
245  "m_BasicParameters.m_OutputGateBias should not be null.");
246  }
247 
248  if (!m_Param.m_CifgEnabled)
249  {
251  {
252  throw armnn::NullPointerException("LstmLayer: "
253  "m_CifgParameters.m_InputToInputWeights should not be null.");
254  }
255 
257  {
258  throw armnn::NullPointerException("LstmLayer: "
259  "m_CifgParameters.m_RecurrentToInputWeights should not be null.");
260  }
261 
263  {
264  throw armnn::NullPointerException("LstmLayer: "
265  "m_CifgParameters.m_InputGateBias should not be null.");
266  }
267 
268  ValidateAndCopyShape(outputShape, inferredShapes[0], m_ShapeInferenceMethod, "LstmLayer");
269  }
270  else
271  {
273  {
274  throw armnn::Exception("LstmLayer: "
275  "m_CifgParameters.m_InputToInputWeights should not have a value "
276  "when CIFG is enabled.");
277  }
278 
280  {
281  throw armnn::Exception("LstmLayer: "
282  "m_CifgParameters.m_RecurrentToInputWeights should not have a value "
283  "when CIFG is enabled.");
284  }
285 
287  {
288  throw armnn::Exception("LstmLayer: "
289  "m_CifgParameters.m_InputGateBias should not have a value "
290  "when CIFG is enabled.");
291  }
292 
293  ValidateAndCopyShape(outputShape, inferredShapes[0], m_ShapeInferenceMethod, "LstmLayer");
294  }
295 
297  {
299  {
300  throw armnn::NullPointerException("LstmLayer: "
301  "m_ProjectionParameters.m_ProjectionWeights should not be null.");
302  }
303  }
304 
306  {
307  if (!m_Param.m_CifgEnabled)
308  {
310  {
311  throw armnn::NullPointerException("LstmLayer: "
312  "m_PeepholeParameters.m_CellToInputWeights should not be null "
313  "when Peephole is enabled and CIFG is disabled.");
314  }
315  }
316 
318  {
319  throw armnn::NullPointerException("LstmLayer: "
320  "m_PeepholeParameters.m_CellToForgetWeights should not be null.");
321  }
322 
324  {
325  throw armnn::NullPointerException("LstmLayer: "
326  "m_PeepholeParameters.m_CellToOutputWeights should not be null.");
327  }
328  }
329 
331  GetOutputSlot(1).GetTensorInfo().GetShape(), inferredShapes[1], m_ShapeInferenceMethod, "LstmLayer", 1);
333  GetOutputSlot(2).GetTensorInfo().GetShape(), inferredShapes[2], m_ShapeInferenceMethod, "LstmLayer", 2);
335  GetOutputSlot(3).GetTensorInfo().GetShape(), inferredShapes[3], m_ShapeInferenceMethod, "LstmLayer", 3);
336 
338  {
340  {
342  {
343  throw armnn::NullPointerException("LstmLayer: "
344  "m_LayerNormParameters.m_inputLayerNormWeights should not be null.");
345  }
346  }
347 
349  {
350  throw armnn::NullPointerException("LstmLayer: "
351  "m_LayerNormParameters.m_forgetLayerNormWeights should not be null.");
352  }
353 
355  {
356  throw armnn::NullPointerException("LstmLayer: "
357  "m_LayerNormParameters.m_cellLayerNormWeights should not be null.");
358  }
359 
361  {
362  throw armnn::NullPointerException("LstmLayer: "
363  "m_LayerNormParameters.m_outputLayerNormWeights should not be null.");
364  }
365  }
366 }
367 
369 {
370  // For API stability DO NOT ALTER order and add new members to the end of vector
380 
381  // Cifg parameters
385 
386  // Projection parameters
389 
390  // Peephole parameters
394 
395  // Layer normalisation parameters
400 }
401 
403 {
404  std::vector<ConstTensor> constTensors;
405 
406  LstmDescriptor descriptor = GetParameters();
407 
417 
418  // Cifg parameters
422 
423  // Projection parameters
426 
427  // Peephole parameters
431 
432  // Layer normalisation parameters
437 
438  // First add mandatory/basic parameters
440  {
441  constTensors.emplace_back(ConstTensor(managedInputToForgetWeights.GetTensorInfo(),
442  managedInputToForgetWeights.Map()));
443  }
445  {
446  constTensors.emplace_back(ConstTensor(managedInputToCellWeights.GetTensorInfo(),
447  managedInputToCellWeights.Map()));
448  }
450  {
451  constTensors.emplace_back(ConstTensor(managedInputToOutputWeights.GetTensorInfo(),
452  managedInputToOutputWeights.Map()));
453  }
455  {
456  constTensors.emplace_back(ConstTensor(
457  managedRecurrentToForgetWeights.GetTensorInfo(),
458  managedRecurrentToForgetWeights.Map()));
459  }
461  {
462  constTensors.emplace_back(ConstTensor(
463  managedRecurrentToCellWeights.GetTensorInfo(),
464  managedRecurrentToCellWeights.Map()));
465  }
467  {
468  constTensors.emplace_back(ConstTensor(
469  managedRecurrentToOutputWeights.GetTensorInfo(),
470  managedRecurrentToOutputWeights.Map()));
471  }
472  if (m_BasicParameters.m_ForgetGateBias != nullptr)
473  {
474  constTensors.emplace_back(ConstTensor(managedForgetGateBias.GetTensorInfo(),
475  managedForgetGateBias.Map()));
476  }
477  if (m_BasicParameters.m_CellBias != nullptr)
478  {
479  constTensors.emplace_back(ConstTensor(managedCellBias.GetTensorInfo(),
480  managedCellBias.Map()));
481  }
482  if (m_BasicParameters.m_OutputGateBias != nullptr)
483  {
484  constTensors.emplace_back(ConstTensor(managedOutputGateBias.GetTensorInfo(),
485  managedOutputGateBias.Map()));
486  }
487 
488  // Add cifg parameters
489  if (!descriptor.m_CifgEnabled)
490  {
492  {
493  constTensors.emplace_back(ConstTensor(managedInputToInputWeights.GetTensorInfo(),
494  managedInputToInputWeights.Map()));
495  }
497  {
498  constTensors.emplace_back(ConstTensor(
499  managedRecurrentToInputWeights.GetTensorInfo(),
500  managedRecurrentToInputWeights.Map()));
501  }
502  if (m_CifgParameters.m_InputGateBias != nullptr)
503  {
504  constTensors.emplace_back(ConstTensor(managedInputGateBias.GetTensorInfo(),
505  managedInputGateBias.Map()));
506  }
507  }
508 
509  // Add peephole parameters
510  if (descriptor.m_PeepholeEnabled)
511  {
512  if (!descriptor.m_CifgEnabled)
513  {
515  {
516  constTensors.emplace_back(ConstTensor(managedCellToInputWeights.GetTensorInfo(),
517  managedCellToInputWeights.Map()));
518  }
519  }
521  {
522  constTensors.emplace_back(ConstTensor(managedCellToForgetWeights.GetTensorInfo(),
523  managedCellToForgetWeights.Map()));
524  }
526  {
527  constTensors.emplace_back(ConstTensor(managedCellToOutputWeights.GetTensorInfo(),
528  managedCellToOutputWeights.Map()));
529  }
530  }
531 
532  // Add projection parameters
533  if (descriptor.m_ProjectionEnabled)
534  {
536  {
537  constTensors.emplace_back(ConstTensor(managedProjectionWeights.GetTensorInfo(),
538  managedProjectionWeights.Map()));
539  }
541  {
542  constTensors.emplace_back(ConstTensor(managedProjectionBias.GetTensorInfo(),
543  managedProjectionBias.Map()));
544  }
545  }
546 
547  // Add norm parameters
548  if (descriptor.m_LayerNormEnabled)
549  {
550  if (!descriptor.m_CifgEnabled)
551  {
553  {
554  constTensors.emplace_back(ConstTensor(managedInputLayerNormWeights.GetTensorInfo(),
555  managedInputLayerNormWeights.Map()));
556  }
557  }
559  {
560  constTensors.emplace_back(ConstTensor(managedForgetLayerNormWeights.GetTensorInfo(),
561  managedForgetLayerNormWeights.Map()));
562  }
564  {
565  constTensors.emplace_back(ConstTensor(managedCellLayerNormWeights.GetTensorInfo(),
566  managedCellLayerNormWeights.Map()));
567  }
569  {
570  constTensors.emplace_back(ConstTensor(managedOutputLayerNormWeights.GetTensorInfo(),
571  managedOutputLayerNormWeights.Map()));
572  }
573  }
574 
575  strategy.ExecuteStrategy(this, GetParameters(), constTensors, GetName());
576 }
577 
578 } // namespace armnn
#define CHECK_LOCATION()
Definition: Exceptions.hpp:203
A tensor defined by a TensorInfo (shape and data type) and an immutable backing store.
Definition: Tensor.hpp:330
Base class for all ArmNN exceptions so that users can filter to just those.
Definition: Exceptions.hpp:47
std::vector< std::reference_wrapper< const std::shared_ptr< ConstTensorHandle > >> ImmutableConstantTensors
Definition: INetwork.hpp:141
virtual void ExecuteStrategy(const IConnectableLayer *layer, const armnn::BaseDescriptor &descriptor, const std::vector< armnn::ConstTensor > &constants, const char *name, const armnn::LayerBindingId id=0)=0
virtual std::unique_ptr< IWorkload > CreateWorkload(LayerType type, const QueueDescriptor &descriptor, const WorkloadInfo &info) const =0
Backends should implement their own CreateWorkload function with a switch statement.
const TensorInfo & GetTensorInfo() const override
Gets the TensorInfo for this InputSlot.
Definition: Layer.cpp:614
void VerifyLayerConnections(unsigned int expectedConnections, const CheckLocation &location) const
Definition: Layer.cpp:410
const OutputSlot & GetOutputSlot(unsigned int index=0) const override
Get the const output slot handle by slot index.
Definition: Layer.hpp:339
void VerifyShapeInferenceType(const TensorShape &outputShape, ShapeInferenceMethod shapeInferenceMethod)
Definition: Layer.cpp:526
const char * GetName() const override
Returns the name of the layer.
Definition: Layer.hpp:332
const InputSlot & GetInputSlot(unsigned int index) const override
Get a const input slot handle by slot index.
Definition: Layer.hpp:337
void ValidateAndCopyShape(const TensorShape &outputShape, const TensorShape &inferredShape, const ShapeInferenceMethod shapeInferenceMethod, const std::string &layerName, const unsigned int outputSlotIndex=0)
Definition: Layer.cpp:457
void SetAdditionalInfo(QueueDescriptor &descriptor) const
Definition: Layer.cpp:303
ShapeInferenceMethod m_ShapeInferenceMethod
Definition: Layer.hpp:441
WorkloadInfo PrepInfoAndDesc(QueueDescriptor &descriptor) const
Helper function to reduce duplication in *Layer::CreateWorkload.
LstmDescriptor m_Param
The parameters for the layer (not including tensor-valued weights etc.).
const LstmDescriptor & GetParameters() const override
This layer represents a LSTM operation.
Definition: LstmLayer.hpp:17
LstmOptCifgParameters m_CifgParameters
Definition: LstmLayer.hpp:21
Layer::ImmutableConstantTensors GetConstantTensorsByRef() const override
Retrieve the handles to the constant values stored by the layer.
Definition: LstmLayer.cpp:368
LstmOptProjectionParameters m_ProjectionParameters
Definition: LstmLayer.hpp:22
LstmOptLayerNormParameters m_LayerNormParameters
Definition: LstmLayer.hpp:24
void ExecuteStrategy(IStrategy &strategy) const override
Apply a visitor to this layer.
Definition: LstmLayer.cpp:402
LstmOptPeepholeParameters m_PeepholeParameters
Definition: LstmLayer.hpp:23
std::vector< TensorShape > InferOutputShapes(const std::vector< TensorShape > &inputShapes) const override
By default returns inputShapes if the number of inputs are equal to number of outputs,...
Definition: LstmLayer.cpp:150
LstmBasicParameters m_BasicParameters
Definition: LstmLayer.hpp:20
void ValidateTensorShapesFromInputs() override
Check if the input tensor shape(s) will lead to a valid configuration of LstmLayer.
Definition: LstmLayer.cpp:172
LstmLayer * Clone(Graph &graph) const override
Creates a dynamically-allocated copy of this layer.
Definition: LstmLayer.cpp:80
LstmLayer(const LstmDescriptor &param, const char *name)
Constructor to create a LstmLayer.
Definition: LstmLayer.cpp:17
virtual std::unique_ptr< IWorkload > CreateWorkload(const IWorkloadFactory &factory) const override
Makes a workload for the LSTM type.
Definition: LstmLayer.cpp:22
const TensorInfo & GetTensorInfo() const
const void * Map(bool blocking=true)
RAII Managed resource Unmaps MemoryArea once out of scope.
const TensorInfo & GetTensorInfo() const override
Definition: Layer.cpp:100
const TensorShape & GetShape() const
Definition: Tensor.hpp:193
Copyright (c) 2021 ARM Limited and Contributors.
LayerType
When adding a new layer, adapt also the LastLayer enum value in the enum class LayerType below.
Definition: Types.hpp:494
const TensorInfo & GetTensorInfo(const ITensorHandle *tensorHandle)
float32 helpers
std::shared_ptr< ConstTensorHandle > m_RecurrentToForgetWeights
A unique pointer to represent 2D weights tensor with dimensions [output_size, num_units].
std::shared_ptr< ConstTensorHandle > m_CellBias
A unique pointer to represent 1D weights tensor with dimensions [num_units].
std::shared_ptr< ConstTensorHandle > m_InputToOutputWeights
A unique pointer to represent 2D weights tensor with dimensions [input_size, num_units].
std::shared_ptr< ConstTensorHandle > m_RecurrentToCellWeights
A unique pointer to represent 2D weights tensor with dimensions [output_size, num_units].
std::shared_ptr< ConstTensorHandle > m_OutputGateBias
A unique pointer to represent 1D weights tensor with dimensions [num_units].
std::shared_ptr< ConstTensorHandle > m_InputToForgetWeights
A unique pointer to represent 2D weights tensor with dimensions [input_size, num_units].
std::shared_ptr< ConstTensorHandle > m_InputToCellWeights
A unique pointer to represent 2D weights tensor with dimensions [input_size, num_units].
std::shared_ptr< ConstTensorHandle > m_RecurrentToOutputWeights
A unique pointer to represent 2D weights tensor with dimensions [output_size, num_units].
std::shared_ptr< ConstTensorHandle > m_ForgetGateBias
A unique pointer to represent 1D weights tensor with dimensions [num_units].
An LstmDescriptor for the LstmLayer.
bool m_PeepholeEnabled
Enable/disable peephole.
bool m_LayerNormEnabled
Enable/disable layer normalization.
bool m_ProjectionEnabled
Enable/disable the projection layer.
bool m_CifgEnabled
Enable/disable cifg (coupled input & forget gate).
std::shared_ptr< ConstTensorHandle > m_InputToInputWeights
A unique pointer to represent 2D weights tensor with dimensions [input_size, num_units].
std::shared_ptr< ConstTensorHandle > m_InputGateBias
A unique pointer to represent 1D weights tensor with dimensions [num_units].
std::shared_ptr< ConstTensorHandle > m_RecurrentToInputWeights
A unique pointer to represent 2D weights tensor with dimensions [input_size, num_units].
std::shared_ptr< ConstTensorHandle > m_CellLayerNormWeights
A unique pointer to represent 1D weights tensor with dimensions [num_units].
std::shared_ptr< ConstTensorHandle > m_InputLayerNormWeights
A unique pointer to represent 1D weights tensor with dimensions [num_units].
std::shared_ptr< ConstTensorHandle > m_OutputLayerNormWeights
A unique pointer to represent 1D weights tensor with dimensions [num_units].
std::shared_ptr< ConstTensorHandle > m_ForgetLayerNormWeights
A unique pointer to represent 1D weights tensor with dimensions [num_units].
std::shared_ptr< ConstTensorHandle > m_CellToForgetWeights
A unique pointer to represent 1D weights tensor with dimensions [num_units].
std::shared_ptr< ConstTensorHandle > m_CellToInputWeights
A unique pointer to represent 1D weights tensor with dimensions [num_units].
std::shared_ptr< ConstTensorHandle > m_CellToOutputWeights
A unique pointer to represent 1D weights tensor with dimensions [num_units].
std::shared_ptr< ConstTensorHandle > m_ProjectionBias
A unique pointer to represent 1D weights tensor with dimensions [output_size].
std::shared_ptr< ConstTensorHandle > m_ProjectionWeights
A unique pointer to represent 2D weights tensor with dimensions [output_size, num_units].
const ConstTensorHandle * m_OutputLayerNormWeights
const ConstTensorHandle * m_InputToOutputWeights
const ConstTensorHandle * m_InputLayerNormWeights
const ConstTensorHandle * m_CellToForgetWeights
const ConstTensorHandle * m_RecurrentToInputWeights
const ConstTensorHandle * m_ForgetGateBias
const ConstTensorHandle * m_ProjectionWeights
const ConstTensorHandle * m_InputGateBias
const ConstTensorHandle * m_RecurrentToOutputWeights
const ConstTensorHandle * m_OutputGateBias
const ConstTensorHandle * m_CellBias
const ConstTensorHandle * m_InputToCellWeights
const ConstTensorHandle * m_CellToInputWeights
const ConstTensorHandle * m_CellToOutputWeights
const ConstTensorHandle * m_InputToForgetWeights
const ConstTensorHandle * m_InputToInputWeights
const ConstTensorHandle * m_RecurrentToCellWeights
const ConstTensorHandle * m_ProjectionBias
const ConstTensorHandle * m_ForgetLayerNormWeights
const ConstTensorHandle * m_RecurrentToForgetWeights
const ConstTensorHandle * m_CellLayerNormWeights