How QLinearConv layer absorb the Relu function about onnxruntime HOT 4 CLOSED

The Relu in the ONNX model is removed by quantization tool:

do you mind to share the onnx model(both quantized and non-quantized)?

from onnxruntime.

The Relu in the ONNX model is removed by quantization tool:

onnxruntime/onnxruntime/python/tools/quantization/quantize.py

Line 189 in 981893c

QDQKeepRemovableActivations = True/False:

do you mind to share the onnx model(both quantized and non-quantized)?

yes, but where these -128 are from? The answer is still unfindable. the model I provided are quantized, here I attatched the unquantized model. This model is pretty simple and only for test use.

from onnxruntime.

Could you explain how the Relu is handled in QLinearConv? I cannot find anything related to "QuantReLU".

from onnxruntime.

Sorry, the problem is solved, the place where I put ReLU is the key, the Relu is part of requantization.

from onnxruntime.