Compare runs with Guild

Guild supports a variety of techniques for comparing runs. You may want to compare runs for a variety of reasons:

  • Find the training run with the lowest loss or highest accuracy
  • Compare hyperparameters and other configuration to understand why an operation failed or succeeded
  • Compare output generated by runs

For more information, see help for the following commands:

Compare run loss and accuracy.
Diff two runs.
Open Guild View, which supports run exploration and comparison.
Open TensorBoard, which supports run scalar comparison.