saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-05-18 21:21:17 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Fix bug in the back to back quantization of matmul and conv (#5264)

Browse source 
* fix bug in the back to back quantization of matmul and conv

* fix bug in back to back gather
This commit is contained in:
Yufeng Li 2020-09-23 08:47:20 -07:00 committed by GitHub
parent b5a6a8e847
commit 61ba5b501a
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
4 changed files with 4 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
onnxruntime/python/tools/quantization/onnx_model.py
View file

@ -127,7 +127,7 @@ class ONNXModel:
nodes.append(node)
return nodes
def save_model_to_file(self, output_path, use_external_data_format=False):
def save_model_to_file(self, output_path, use_external_data_format=False):
'''
Save model to external data, which is needed for model size > 2GB
'''

2
onnxruntime/python/tools/quantization/operators/conv.py
View file

@ -22,7 +22,7 @@ class ConvInteger(QuantOperatorBase):
quantized_bias_name = self.quantizer.quantize_bias(node, nodes)
bias_present = True
conv_integer_output = node.output[0] + "_quantized"
conv_integer_output = node.output[0] + "_output_quantized"
conv_integer_name = node.name + "_quant" if node.name != "" else ""
kwargs = {}

2
onnxruntime/python/tools/quantization/operators/gather.py
View file

@ -15,7 +15,7 @@ class GatherQuant(QuantOperatorBase):
node = self.node
assert (node.op_type == "Gather")
if (not self.quantizer._is_valid_quantize_weight(node.input[0])):
self.quantizer.new_nodes += [node]
super().quantize()
return
(quantized_input_names, zero_point_names, scale_names, nodes) = \

2
onnxruntime/python/tools/quantization/operators/matmul.py
View file

@ -18,7 +18,7 @@ class MatMulInteger(QuantOperatorBase):
(quantized_input_names, zero_point_names, scale_names, nodes) = \
self.quantizer.quantize_inputs(node, [0, 1])
matmul_integer_output = node.output[0] + "_quantized"
matmul_integer_output = node.output[0] + "_output_quantized"
matmul_integer_name = node.name + "_quant" if node.name != "" else ""
matmul_integer_node = onnx.helper.make_node("MatMulInteger", quantized_input_names + zero_point_names,
[matmul_integer_output], matmul_integer_name)