Pull request pytorch#88: Turn off optimization REPLACE_AVERAGE_POOL_BEFORE_FULLY_CONNECTED_WITH_SUM

Merge in AITEC/executorch from feature/nxg11066/EIEX-194-turn-off-optimization-ReplaceAveragePoolBeforeFullyConnectedWithSum to main-nxp

* commit '5ef87a69f050c257230779421e87eddd666cd8ec':
  Remove optimization REPLACE_AVERAGE_POOL_BEFORE_FULLY_CONNECTED_WITH_SUM

Author: roman-janik-nxp

backends/nxp/backend/ir/tflite_optimizer/optimizer.py

@@ -32,8 +32,6 @@
     RemoveIdentityTransposeOperators
 from executorch.backends.nxp.backend.ir.tflite_optimizer.optimizations.remove_unused_tensors_and_buffers import \
     RemoveUnusedTensorsAndBuffers
-from executorch.backends.nxp.backend.ir.tflite_optimizer.optimizations.replace_average_pool_before_fully_connected_with_sum import \
-    ReplaceAveragePoolBeforeFullyConnectedWithSum
 
 
 class Optimization(Enum):
@@ -60,7 +58,6 @@ class Optimization(Enum):
 
     MOVE_ACTIVATION_BEFORE_CONCAT = 15
     COMBINE_HARD_SIGMOID_AND_MUL_INTO_HARD_SWISH = 16
-    REPLACE_AVERAGE_POOL_BEFORE_FULLY_CONNECTED_WITH_SUM = 17
 
 
 class Optimizer:
@@ -111,9 +108,7 @@ def __init__(self, builder: 'model_builder.ModelBuilder', conversion_config: Con
 
             Optimization.MOVE_ACTIVATION_BEFORE_CONCAT: MoveActivationBeforeConcatenation(builder, conversion_config),
             Optimization.COMBINE_HARD_SIGMOID_AND_MUL_INTO_HARD_SWISH:
-                CombineHardSigmoidAndMulIntoHardSwish(builder, conversion_config),
-            Optimization.REPLACE_AVERAGE_POOL_BEFORE_FULLY_CONNECTED_WITH_SUM:
-                ReplaceAveragePoolBeforeFullyConnectedWithSum(builder, conversion_config)
+                CombineHardSigmoidAndMulIntoHardSwish(builder, conversion_config)
         }
 
     def optimize(self, optimization_whitelist: list[Optimization] | None = None,