It would be good to provide functionality to fix the random seeds used for generating training datasets, training networks, and sampling. This will help to ensure reproducibility of results.
One way might be to specify the seed in the various config files (or on the command-line for inference).
It would be good to provide functionality to fix the random seeds used for generating training datasets, training networks, and sampling. This will help to ensure reproducibility of results.
One way might be to specify the seed in the various config files (or on the command-line for inference).