1.0.0.1 | Prototype Trainer

Every 100 steps (configurable), the trainer saves a full state: weights, optimizer momentum, epoch number, and even random seed states. With version 1.0.0.1, checkpoints are now compressed using LZ4, reducing disk usage by up to 40% compared to 1.0.0.0.

This isn't just a rename; this update focuses heavily on backend stability after the initial beta testing phase. If you were experiencing crashes during the loading screen or hotkey conflicts, this build should resolve those issues. prototype trainer 1.0.0.1

trainer.analyze() # Outputs detailed report on gradient flow, class-wise accuracy, and suggestions. Every 100 steps (configurable), the trainer saves a