No Dataset Size Limitation: Training jobs can now process datasets of any size. Previously, FCS recommended that datasets be limited to 150GB for training jobs.
Faster Training Job start times: Dataset management optimizations have reduced the overall starting time of Training Jobs, by speeding up the time to load a dataset.Previously, training job start times could be much longer depending on the size of the dataset.
Read-Only /input/ Directory: The /input/ directory ─ where Datasets are mounted ─ is now set as read-only, improving clarity on its intended usage. You can continue to use the /output/ directory to save artifacts and results as well as any data processing operations you need to perform during training.
Faster Training Fetch Availability: The training fetch <name> command is now available 35% sooner after a training job completes due to an optimized save process.
UX improvement for training fetch to display a smoother data download progress.