Hello !
I can’t understand where I have the error, in the configuration I write that I want fake per_tensor_symmetric quantization, but when I display the picture of the graph, he writes that I have a FakeQuantizePerTensorAffineBackward node.
What am I doing wrong?
Thank you in advance!
import torch
from torch import nn
from torchviz import make_dot
model = nn.Sequential()
model.add_module('W0', nn.Linear(8, 16))
model.add_module('tanh', nn.Tanh())
model.add_module('W1', nn.Linear(16, 1))
model.qconfig = torch.quantization.QConfig(activation=torch.quantization.FakeQuantize.with_args(observer=torch.quantization.MinMaxObserver,
quant_min=0,
quant_max=255,
qscheme=torch.per_tensor_symmetric,
dtype=torch.quint8,
reduce_range=False),
weight=torch.quantization.FakeQuantize.with_args(observer=torch.quantization.MinMaxObserver, quant_min=-128, quant_max=127,
dtype=torch.qint8, qscheme=torch.per_tensor_symmetric, reduce_range=False))
torch.quantization.prepare_qat(model, inplace=True)
x = torch.randn(1,8)
make_dot(model(x), params=dict(model.named_parameters()))