Skip to main content

How-To Guides

How to Predict Training Time Using Metadata​

  • Problem: Estimating runtime for deep learning models on HPC resources.
  • Steps:
    1. Load dataset-new.csv.
    2. Select relevant features (batch, epochs, gpu_core_count, etc.).
    3. Train a regression model.
  • Advanced Tips: Integrate HLO graph features for improved accuracy using GNNs.
  • Troubleshooting: Ensure consistent preprocessing of categorical fields.
  • Example: Refer to the baseline notebook.