Reduce code duplication

Author: Xeratec

Deeploy/Targets/Generic/Parsers.py

@@ -2696,9 +2696,9 @@ def parseNodeCtxt(self, ctxt: NetworkContext, node: gs.Node, channels_first: boo
         self.operatorRepresentation['data_out'] = node.outputs[0].name
         if len(node.inputs) == 3:
             self.operatorRepresentation['bias'] = node.inputs[2].name
-            self.operatorRepresentation['has_bias'] = True
+            self.operatorRepresentation['has_bias'] = "true"
         else:
-            self.operatorRepresentation['has_bias'] = False
+            self.operatorRepresentation['has_bias'] = "false"
         # Get output shape from context
         data_out = ctxt.lookup(node.outputs[0].name)
         out_shape = data_out.shape
@@ -2779,6 +2779,7 @@ def parseNodeCtxt(self,
             data_out = newCtxt.lookup(node.outputs[0].name)
             in_shape = data_in.shape
             out_shape = data_out.shape
+            self.operatorRepresentation['batch'] = in_shape[0]
             self.operatorRepresentation['ch_im_in'] = in_shape[1]
             self.operatorRepresentation['dim_im_in_y'] = in_shape[2]
             self.operatorRepresentation['ch_im_out'] = out_shape[1]

Deeploy/Targets/Generic/Templates/ConvTransposeTemplate.py

@@ -2,63 +2,9 @@
 #
 # SPDX-License-Identifier: Apache-2.0
 
-from typing import Dict, List, Tuple
+from Deeploy.DeeployTypes import NodeTemplate
 
-from Deeploy.DeeployTypes import NetworkContext, NodeTemplate, OperatorRepresentation
-
-
-class _ConvTranspose1D_Template(NodeTemplate):
-
-    def __init__(self, templateStr):
-        super().__init__(templateStr)
-
-    def alignToContext(self, ctxt: NetworkContext,
-                       operatorRepresentation: OperatorRepresentation) -> Tuple[NetworkContext, Dict, List[str]]:
-
-        # input/output tensors
-        # input/output tensors
-        data_in = ctxt.lookup(operatorRepresentation["data_in"])
-        data_out = ctxt.lookup(operatorRepresentation["data_out"])
-
-        #quantized tensor offset computation
-        operatorRepresentation["input_offset"] = 0
-        if hasattr(data_in, "_signed") and hasattr(data_in, "nLevels"):
-            operatorRepresentation["input_offset"] = (data_in._signed == 0) * int(data_in.nLevels // 2)
-
-        operatorRepresentation["output_offset"] = 0
-        if hasattr(data_out, "_signed") and hasattr(data_out, "nLevels"):
-            operatorRepresentation["output_offset"] = -(data_out._signed == 0) * int(data_out.nLevels // 2)
-
-        # Batch size
-        operatorRepresentation["batch"] = data_in.shape[0]
-
-        # input/output shape (format NCW)
-        operatorRepresentation["ch_im_in"] = data_in.shape[1]
-        operatorRepresentation["dim_im_in_y"] = data_in.shape[2]
-
-        operatorRepresentation["ch_im_out"] = data_out.shape[1]
-        operatorRepresentation["dim_im_out_y"] = data_out.shape[2]
-
-        # weights and kernel
-        weight = ctxt.lookup(operatorRepresentation["weight"])
-        operatorRepresentation["dim_kernel_y"] = weight.shape[2]  # Shape: [C_out, C_in, K]
-
-        # Stride
-        operatorRepresentation["stride_y"] = operatorRepresentation.get("stride_y", 1)
-
-        # Bias (optional)
-        operatorRepresentation["has_bias"] = "true" if "bias" in operatorRepresentation else "false"
-        operatorRepresentation["bias"] = operatorRepresentation.get("bias", "NULL")
-
-        operatorRepresentation[
-            "batchOffsetIn"] = operatorRepresentation["ch_im_in"] * operatorRepresentation["dim_im_in_y"]
-        operatorRepresentation[
-            "batchOffsetOut"] = operatorRepresentation["ch_im_out"] * operatorRepresentation["dim_im_out_y"]
-
-        return ctxt, operatorRepresentation, []
-
-
-referenceTemplate = _ConvTranspose1D_Template("""
+referenceTemplate = NodeTemplate("""
 <%
 batchOffsetIn = ch_im_in * dim_im_in_y
 batchOffsetOut = ch_im_out * dim_im_out_y