Fix dataset info: FineWebEdu -> TinyStories

Fix: Add HuggingFace standard attributes to config for generate() compatibility

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

Add auto_map to config.json for automatic model loading

Update usage example with trust_remote_code parameter

Update README for checkpoint 35000 with distillation details

Update to checkpoint 35000 (loss 3.46, includes distillation)

Update README with checkpoint 20000 details and benchmarks

Update to checkpoint 20000 (loss 0.758, best quality)

Remove old/unnecessary file: __pycache__/upload_to_hf.cpython-310.pyc

Remove old/unnecessary file: __pycache__/convert_to_hf.cpython-310.pyc

Remove old/unnecessary file: README_SCRIPTS.md

Remove old/unnecessary file: requirements.txt

Remove old/unnecessary file: upload_to_hf.py

Remove old/unnecessary file: test_model.py

Remove old/unnecessary file: publish_model.py

Remove old/unnecessary file: convert_to_hf.py

Remove old/unnecessary file: add_tokenizer.py

Remove old/unnecessary file: training_metadata.json

Remove old/unnecessary file: model.safetensors

Update to checkpoint 35000 with fixed Pre-LN architecture (iter 35k, loss 3.46)

Upload model - 20000 iterations, loss: 0.7583

Upload model - 35000 iterations, loss: 3.4640

Converted from MLX checkpoint_35000 (transpose + lm_head included)

Upload model - 35000 iterations, loss: 3.4640

Upload tokenizer_config.json with huggingface_hub

Upload merges.txt with huggingface_hub

Upload vocab.json with huggingface_hub

Upload tokenizer.json with huggingface_hub

Upload model - 35000 iterations, loss: 3.4640

initial commit