So, re the first part, when I run:
delattr(model._modules[key], param_key)
setattr(model._modules[key], param_key, updated)
print(model._modules[key]._parameters[param_key])
I get the error:
print(model._modules[key]._parameters[param_key])KeyError: ‘weight’
where param_key is weight. The point being, when I loop over parameters using
for param_key in model._modules[key]._parameters:
# module._modules[key]._parameters[param_key] = memo[p] # old
delattr(module, param_key)
setattr(module, param_key, memo[p])
upon the second iteration I get the error:
for param_key in model._modules[key]._parameters:RuntimeError: OrderedDict mutated during iteration
Re the second part, I tried using _stateless with the snippet you provided here, and I’m getting the following error:
for name, tensor in parameters_and_buffers.items():AttributeError: ‘generator’ object has no attribute ‘items’
when I
print(params)
I get
<generator object Module.named_parameters at 0x7f98f8dc1450>
My model parameters include:
self.fc = nn.Linear(10, 10)
self.relu = nn.ReLU()
self.params_list = nn.ParameterList([self.fc1.weight, self.fc1.bias])