NNCF skips insert position for FakeQuanizer about nncf

[vshampor]

@aleksei-kashapov could you retry with latest NNCF version?

[kshpv]

@vshampor
The issue with lacking of FQ on the input tensor was resolved.
As you asked I quantized the same model by NNCF 1.5. with default quantization mode. As a result the problem with the pattern (Power, ReduceSum, Power, Add, Power) still remains

[vshampor]

@aleksei-kashapov which operation in PyTorch does this problematic subgraph correspond to?

[kshpv]

@vshampor

class L2Norm(nn.Module):

    def __init__(self, n_dims, scale=20., eps=1e-10):
        super(L2Norm, self).__init__()
        self.n_dims = n_dims
        self.weight = nn.Parameter(torch.Tensor(self.n_dims))
        self.eps = eps
        self.scale = scale

    def forward(self, x):
        # normalization layer convert to FP32 in FP16 training
        x_float = x.float()
        norm = x_float.pow(2).sum(1, keepdim=True).sqrt() + self.eps
        return (self.weight[None, :, None, None].float().expand_as(x_float) *
                x_float / norm).type_as(x)

[vshampor]

@aleksei-kashapov did @register_module approach work in this case?

[vshampor]

@aleksei-kashapov still waiting for you to verify.

[vshampor]

@aleksei-kashapov did you have a chance to verify it?

[kshpv]

@vshampor I am really sorry. I apologize, I will check it out soon

[kshpv]

should be resolved by #637

[kshpv]

Should be fixed with #984

[kshpv]

The issue is resolved by #984