Fine-Tuning LR Groups

Build two optimiser parameter groups for fine-tuning: the backbone gets the learning rate base_lr * backbone_factor and the head gets base_lr.

Implement solve(backbone, head, base_lr, backbone_factor).

Returns. Return [{'params': list(backbone), 'lr': base_lr * backbone_factor}, {'params': list(head), 'lr': base_lr}].

Examples

Example 1

Input:

solve(['backbone'], ['head'], 0.1, 0.1)

Output:

[{'params': ['backbone'], 'lr': 0.01}, {'params': ['head'], 'lr': 0.1}]

Example 2

Input:

solve([], ['head'], 0.01, 0.2)

Output:

[{'params': [], 'lr': 0.002}, {'params': ['head'], 'lr': 0.01}]