A training job fails immediately because the configured algorithm expects a supported data format different from the supplied files. What is the best correction?

1. Validate the algorithm input contract and transform the dataset into a supported format and channel configuration.
2. Ignore the error and register the absent model artifact as approved.
3. Increase the number of epochs without changing the input.
4. Use a larger inference endpoint instance.