saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-30 20:18:08 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Make quantization parameters as constant weigth instead of overrideable (#3160)

Browse source
This commit is contained in:
Yufeng Li 2020-03-10 08:35:02 -07:00 committed by GitHub
parent 3d928de778
commit 876d0c5430
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 0 additions and 14 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

14
onnxruntime/python/tools/quantization/quantize.py
View file

@ -197,9 +197,7 @@ def _add_initializer_if_not_present(graph, name, value, shape, type):
'''
if _find_by_name(name, graph.initializer) is None:
initializer = onnx.helper.make_tensor(name, type, shape, value)
value_info = onnx.helper.make_tensor_value_info(name, type, shape)
graph.initializer.extend([initializer])
graph.input.extend([value_info])
def _get_qrange_for_qType(qType):
'''
@ -355,15 +353,6 @@ class ONNXQuantizer:
self.model.graph.initializer.extend([packed_weight_initializer, scale_initializer, zero_initializer])
# Create input for initialized scale and zeros
packed_weight_value_info = onnx.helper.make_tensor_value_info(packed_weight_name, weight.qType,
weight.initializer.dims)
scale_value_info = onnx.helper.make_tensor_value_info(scale_name, onnx_proto.TensorProto.FLOAT, zero_scale_shape)
zero_point_value_info = onnx.helper.make_tensor_value_info(zero_point_name,
zero_point_type, zero_scale_shape) # zero_point is int for dequantize operator
self.model.graph.input.extend([packed_weight_value_info, scale_value_info, zero_point_value_info])
self._quantized_weights.append(weight)
def _get_quantized_weight(self, initializer, qType):
@ -780,9 +769,6 @@ class ONNXQuantizer:
packed_bias_initializer = onnx.numpy_helper.from_array(bias_np_data, quantized_bias_name)
self.model.graph.initializer.extend([packed_bias_initializer])
bias_value_info = onnx.helper.make_tensor_value_info(quantized_bias_name, onnx_proto.TensorProto.INT32, bias_initializer.dims)
self.model.graph.input.extend([bias_value_info])
# log entries for this quantized bias value
quantized_bias_entry = QuantizedInitializer(bias_name, bias_initializer, [0], [0], [0], [bias_scale],
bias_data, quantized_data, qType=onnx_proto.TensorProto.INT32)