search

mlcommons / algorithmic-efficiency

MLCommons Algorithmic Efficiency is a benchmark and competition measuring neural network training speedups due to algorithmic improvements in both training algorithms and models.
https://mlcommons.org/en/groups/research-algorithms/
Apache License 2.0
319 stars 60 forks source link

Models will always be initialized without dropout layers in self-tuning ruleset #753

Open georgedahl opened 4 months ago

georgedahl commented 4 months ago

In submission_runner.py, if we are in the self-tuning rules, the hyperparameters argument to train_once will always be None.

Then in this code snippet

    dropout_rate = None
    aux_dropout_rate = None
    if hasattr(hyperparameters, 'dropout_rate'):
      dropout_rate = hyperparameters.dropout_rate
    if hasattr(hyperparameters, 'aux_dropout_rate'):
      aux_dropout_rate = hyperparameters.aux_dropout_rate
    model_params, model_state = workload.init_model_fn(
        model_init_rng, dropout_rate, aux_dropout_rate)

workload.init_model_fn will always get None for dropout_rate and aux_dropout_rate, so Dropout layers won't ever be added to the model.

Although submissions could call workload.init_model_fn again themselves to make use of its side effect of setting workload._model, this is awkward and also challenging for workloads near the memory limit since it involves superfluously reconstructing model_params again on device.