saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-08 17:17:15 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Fix conv quant reduce range option (#9922)

Browse source 
Co-authored-by: menj <menj@microsoft.com>
This commit is contained in:
MengX1991 2021-12-07 08:43:54 +08:00 committed by GitHub
parent 5871ca1cd1
commit 58728f95a2
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 3 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
onnxruntime/python/tools/quantization/operators/conv.py
View file

@ -51,7 +51,7 @@ class ConvInteger(QuantOperatorBase):
assert (node.op_type == "Conv")
(quantized_input_names, zero_point_names, scale_names, nodes) = \
self.quantizer.quantize_inputs(node, [0, 1])
self.quantizer.quantize_inputs(node, [0, 1], reduce_range=self.quantizer.reduce_range)
conv_integer_output = node.output[0] + "_output_quantized"
conv_integer_name = node.name + "_quant" if node.name != "" else ""
@ -111,7 +111,7 @@ class QLinearConv(QuantOperatorBase):
if self.quantizer.is_input_a_weight(node.input[1]) and self.quantizer.is_per_channel():
(quantized_input_names, zero_point_names, scale_names, nodes) = \
self.quantizer.quantize_inputs(node, [0])
self.quantizer.quantize_inputs(node, [0], reduce_range=self.quantizer.reduce_range)
quant_weight_tuple = self.quantizer.quantize_weight_per_channel(node.input[1], onnx_proto.TensorProto.INT8,
0)
quantized_input_names.append(quant_weight_tuple[0])
@ -119,7 +119,7 @@ class QLinearConv(QuantOperatorBase):
scale_names.append(quant_weight_tuple[2])
else:
(quantized_input_names, zero_point_names, scale_names, nodes) = \
self.quantizer.quantize_inputs(node, [0, 1])
self.quantizer.quantize_inputs(node, [0, 1], reduce_range=self.quantizer.reduce_range)
if not data_found or quantized_input_names is None:
return super().quantize()