[TorchFix] requires_grad check fails when condition

[EgShes]

TorchFix failes to detect requres_grad mistyping when using condition instead of plain bool value

Here is the code to reproduce

import torch.nn as nn
model = nn.Module()

for name, param in model.named_parameters():
    param.require_grad = 'specific_layer' in name

TorchFix version: 0.1.1

[kit1980]

The problem is that currently TorchFix doesn't understand that param is a Torch tensor.
So to prevent false positives (maybe it's not PyTorch code at all and just something has a valid attribute name require_grad), it only complains when the value is a boolean literal.

Having said that, it's probably better to change TorchFix to complain in this case.
The probability of false positives seems very low.

[EgShes]

I agree with you. Looking forward to adding this functionality.