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
335 stars 69 forks source link

Add function that submissions can call that can change the dropout value #787

Open priyakasimbeg opened 2 months ago

priyakasimbeg commented 2 months ago

Our current API has 2 dropout related limitations:

  1. Currently, in the external tuning ruleset we read the dropout value from the hparam config and pass it to the model initialization functions. In the self-tuning ruleset there exist no convenient way to specify the dropout value in the model initialization.
  2. Furthermore, there is no way to change the dropout value during training.

Having a workload function to change the dropout value that submitters can call will remove both of these limitations.