Transfer Learning Freeze

Prepare a model for transfer learning: set requires_grad = False on every backbone parameter and requires_grad = True on every task-head parameter. The parameter objects are modified in place.

Implement solve(backbone, head).

Returns. Return a tuple (backbone_list, head_list) containing the same parameter objects as lists.

Examples

Example 1

Input:

solve([SimpleNamespace(requires_grad=True)], [SimpleNamespace(requires_grad=False)])

Output:

([namespace(requires_grad=False)], [namespace(requires_grad=True)])

Example 2

Input:

solve([], [SimpleNamespace(requires_grad=False), SimpleNamespace(requires_grad=False)])

Output:

([], [namespace(requires_grad=True), namespace(requires_grad=True)])