File size: 2,472 Bytes
610d9c2 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 | import onnx
from onnx import onnx_pb as onnx_proto # noqa: F401
from ..quant_utils import TENSOR_NAME_QUANT_SUFFIX, QuantizedValue, QuantizedValueType, attribute_to_kwarg, ms_domain
from .base_operator import QuantOperatorBase
class QLinearBinaryOp(QuantOperatorBase):
def __init__(self, onnx_quantizer, onnx_node):
super().__init__(onnx_quantizer, onnx_node)
def quantize(self):
node = self.node
(
data_found,
output_scale_name,
output_zp_name,
_,
_,
) = self.quantizer._get_quantization_params(node.output[0])
(
quantized_input_names,
zero_point_names,
scale_names,
nodes,
) = self.quantizer.quantize_activation(node, [0, 1])
if not data_found or quantized_input_names is None:
return super().quantize()
qlinear_binary_math_output = node.output[0] + TENSOR_NAME_QUANT_SUFFIX
qlinear_binary_math_name = node.name + "_quant" if node.name else ""
kwargs = {}
for attribute in node.attribute:
kwargs.update(attribute_to_kwarg(attribute))
kwargs["domain"] = ms_domain
qlinear_binary_math_inputs = []
# Input 0
qlinear_binary_math_inputs.append(quantized_input_names[0])
qlinear_binary_math_inputs.append(scale_names[0])
qlinear_binary_math_inputs.append(zero_point_names[0])
# Input 1
qlinear_binary_math_inputs.append(quantized_input_names[1])
qlinear_binary_math_inputs.append(scale_names[1])
qlinear_binary_math_inputs.append(zero_point_names[1])
# Output
qlinear_binary_math_inputs.append(output_scale_name)
qlinear_binary_math_inputs.append(output_zp_name)
qlinear_binary_math_node = onnx.helper.make_node(
"QLinear" + node.op_type,
qlinear_binary_math_inputs,
[qlinear_binary_math_output],
qlinear_binary_math_name,
**kwargs,
)
nodes.append(qlinear_binary_math_node)
# Create an entry for this quantized value
q_output = QuantizedValue(
node.output[0],
qlinear_binary_math_output,
output_scale_name,
output_zp_name,
QuantizedValueType.Input,
)
self.quantizer.quantized_value_map[node.output[0]] = q_output
self.quantizer.new_nodes += nodes
|