Fix dataset info: FineWebEdu -> TinyStories

Initial upload: checkpoint 20000 with accurate model card

Update to checkpoint 20000 (best quality, loss 0.758)

Update model card: professional format, remove MLX version reference

Update usage example with trust_remote_code parameter

Update README for checkpoint 35000 with distillation details

Update README with checkpoint 20000 details and benchmarks

Upload model - 35000 iterations, loss: 3.4640

initial commit