requires_grad

metatensor.requires_grad(tensor: TensorMap, requires_grad: bool = True) → TensorMap

Set requires_grad on all arrays (blocks and gradients of blocks) in this tensor to the provided value.

This is mainly intended for torch arrays, and will warn if trying to set requires_grad=True with numpy arrays.

Parameters:

• tensor (TensorMap) – TensorMap to modify

• requires_grad (bool) – new value for requires_grad

Return type:

TensorMap

metatensor.requires_grad_block(block: TensorBlock, requires_grad: bool = True) → TensorBlock

Set requires_grad on the values and all gradients in this block to the provided value.

This is mainly intended for torch arrays, and will warn if trying to set requires_grad=True with numpy arrays.

Parameters:

• block (TensorBlock) – TensorBlock to modify

• requires_grad (bool) – new value for requires_grad

Return type:

TensorBlock