No need to toss conversion because of branching inputs. (#393)

Co-authored-by: dkroell <david.kroell@exptechinc.com>
Co-authored-by: Benjamin Fineran <bfineran@users.noreply.github.com>
Co-authored-by: Mark Kurtz <mark@neuralmagic.com>

Author: 4 people

src/sparseml/onnx/optim/quantization/quantize.py

@@ -9,9 +9,6 @@
 # neuralmagic: no copyright
 # flake8: noqa
 
-import os
-import struct
-
 import numpy as np
 import onnx
 import onnx.numpy_helper
@@ -1086,6 +1083,10 @@ def _quantize_bias(self, node, new_node_list):
         bias_data = self.find_weight_data(bias_initializer)
         quantized_bias_name = bias_name + "_quantized"
 
+        if bias_name in self.quantized_value_map:
+            print(f"WARNING: {bias_name} already has a quantization value.")
+            return quantized_bias_name
+
         # input scale is not provided and this input is dynamically quantized so it is not pre-computed at this point
         # so resort to dynamic quantization for bias
         if (
@@ -1152,7 +1153,6 @@ def _quantize_bias(self, node, new_node_list):
             )
             self._quantized_weights.append(quantized_bias_entry)
 
-            assert bias_name not in self.quantized_value_map
             quantized_value = QuantizedValue(
                 bias_name,
                 quantized_bias_name,