parameter.md

Parameter

Parameter contains tensors with data and gradients. Used to store trainable parameters of Model.

Methods

• data - returns tensor with initial data.
• grad - if is_trainable == true then attribute will contain the gradients computed and future calls to backward() will accumulate (add) gradients into it, otherwise grad is an empty tensor.
• train - set is_trainable to true.
• is_trainable - is true if gradients need to be computed for this Tensor, false otherwise.