Module's parameters get changed during training, that is, they are what is learnt during training of a neural network, but what is a buffer?
and is it learnt during neural network training?
model. parameters(): PyTorch modules have a a method called parameters() which returns an iterator over all the parameters. param.numel(): We use the Iterator object returned by the model.parameters() and calculate the number of elements in it using the .numel() function.
nn modules. Any deep learning model is developed using the subclass of the torch. nn module it uses method like forward(input) which returns the output. A simple neural network takes input to add weights and bias to it feed the input through multiple hidden layers and finally returns the output.
A state_dict is an integral entity if you are interested in saving or loading models from PyTorch. Because state_dict objects are Python dictionaries, they can be easily saved, updated, altered, and restored, adding a great deal of modularity to PyTorch models and optimizers.
The self. modules() method returns an iterable to the many layers or “modules” defined in the model class. This particular piece of code is using that self. modules() iterable to initialize the weights of the different layers present in the model.
Registering these "arguments" as the model's buffer allows pytorch to track them and save them like regular parameters, but prevents pytorch from updating them using SGD mechanism.
The main difference between register and buffer is that the register is a temporary storage area in the processor that allows transferring data faster while the buffer is a temporary storage area in the main memory that holds data before using them. A register is a fast memory location built into the processor.
When you register a new parameter it will appear inside the module.parameters () iterator, but when you register a buffer it will not. Buffers are named tensors that do not update gradients at every step, like parameters.
Both parameters and buffers you create for a module ( nn.Module ). Say you have a linear layer nn.Linear. You already have weight and bias parameters. But if you need a new parameter you use register_parameter () to register a new named parameter that is a tensor.
Pytorch doc for register_buffer()
method reads
This is typically used to register a buffer that should not to be considered a model parameter. For example, BatchNorm’s
running_mean
is not a parameter, but is part of the persistent state.
As you already observed, model parameters are learned and updated using SGD during the training process.
However, sometimes there are other quantities that are part of a model's "state" and should be
- saved as part of state_dict
.
- moved to cuda()
or cpu()
with the rest of the model's parameters.
- cast to float
/half
/double
with the rest of the model's parameters.
Registering these "arguments" as the model's buffer
allows pytorch to track them and save them like regular parameters, but prevents pytorch from updating them using SGD mechanism.
An example for a buffer can be found in _BatchNorm
module where the running_mean
, running_var
and num_batches_tracked
are registered as buffers and updated by accumulating statistics of data forwarded through the layer. This is in contrast to weight
and bias
parameters that learns an affine transformation of the data using regular SGD optimization.
Both parameters and buffers you create for a module (nn.Module
).
Say you have a linear layer nn.Linear
. You already have weight
and bias
parameters. But if you need a new parameter you use register_parameter()
to register a new named parameter that is a tensor.
When you register a new parameter it will appear inside the module.parameters()
iterator, but when you register a buffer it will not.
The difference:
Buffers are named tensors that do not update gradients at every step, like parameters. For buffers, you create your custom logic (fully up to you).
The good thing is when you save the model, all params and buffers are saved, and when you move the model to or off the CUDA params and buffers will go as well.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With