What does the get_model_size() function mean?

[leemgs]

I thought that the execution result of get_model_size() was the amount of memory required for the model to be loaded into memory. The current execution result of the get_model_siz() function was just like model.num_parameters().

So I wonder what the meaning and intent of the get_model_size() function are. Do you want to print the number of parameters? Or, do you want to print the size of the model loaded into memory?

• https://github.com/salesforce/CodeT5/blob/main/models.py#L17

  def get_model_size(model):
  model_parameters = filter(lambda p: p.requires_grad, model.parameters())
  model_size = sum([np.prod(p.size()) for p in model_parameters])
  return "{}M".format(round(model_size / 1e+6))

• codeT5-base model: pytorch_model.bin (892MB)

  • https://huggingface.co/Salesforce/codet5-base/tree/main

[yuewang-cuhk]

Hi, it is the same as model.num_parameters().