Is this if statement necessary?

[darthjaja6]

pytorch-lora/lora.ipynb

Line 459 in a2bdead

" if 'lora' not in name:\n",

When I do

for name, param in net.named_parameters():
  print(name)

I got output

linear1.bias
linear1.parametrizations.weight.original
linear2.bias
linear2.parametrizations.weight.original
linear3.bias
linear3.parametrizations.weight.original

So it seems that anyways the parametrization doesn't add any named_params that has "lora" in path. And I wrote a simple experimental code:

import torch
import torch.nn as nn
import torch.nn.utils.parametrize as parametrize

class WeightParametrization(nn.Module):
    def __init__(self):
        super().__init__()

    def forward(self, weight):
        return weight * 2  # 简单的示例，实际上可以是更复杂的变换

class MyLinear(nn.Module):
    def __init__(self):
        super(MyLinear, self).__init__()
        self.linear = nn.Linear(10, 5)

    def forward(self, x):
        return self.linear(x)

model = MyLinear()
print("Before parametrization:")
for name, param in model.named_parameters():
    print(name, param.shape)

# 注册参数化
parametrization = WeightParametrization()
parametrize.register_parametrization(model.linear, 'weight', parametrization)

print("\nAfter parametrization:")
for name, param in model.named_parameters():
    print(name, param.shape)  # 可以看到权重参数的变化

So it seems that after registering parametrization, the named_params don't really get new param.
What do you think? Did I miss something?