fix IT pruning example issue #2772
Conversation
| @@ -145,7 +145,10 @@ def create_model(model_name='naive'): | |||
| def create_pruner(model, pruner_name, optimizer=None): | |||
| pruner_class = prune_config[pruner_name]['pruner_class'] | |||
| config_list = prune_config[pruner_name]['config_list'] | |||
| return pruner_class(model, config_list, optimizer) | |||
| if pruner_name in ['level', 'slim', 'fpgm', 'l1filter']: | |||
There was a problem hiding this comment.
why keep optimizer parameter for ActivationMeanRankFilterPruner and ActivationAPoZRankFilterPruner? they do not need optimizer hook either. For all one-shot pruners, it seems only TaylorFOWeightFilterPruner needs optimizer ( to calculate gradient contribution).
There was a problem hiding this comment.
why keep optimizer parameter for
ActivationMeanRankFilterPrunerandActivationAPoZRankFilterPruner? they do not need optimizer hook either. For all one-shot pruners, it seems only TaylorFOWeightFilterPruner needs optimizer ( to calculate gradient contribution).
In the current implementation, ActivationMeanRankFilterPruner and ActivationAPoZRankFilterPruner need the optimizer hook. If we remove the optimizer param, the sparsity after pruning will always be 0.
There was a problem hiding this comment.
as discussed in the meeting, let's not do this broken change?
(cherry picked from commit d0a9b10)
Parameter 'optimizer ' was removed from Pruners ('level', 'slim', 'fpgm', 'l1', 'l2') in PR #2676. This example should be changed accordingly.