Document Llama metadata-only dependency

Fix default checkpoint layout for direct loading

Keep one default checkpoint for direct loader compatibility

Document mirrored Llama metadata and CogACT training base

Add Llama-2 tokenizer and config metadata for offline loading

Add Llama-2 tokenizer and config metadata for offline loading

Add Llama-2 tokenizer and config metadata for offline loading

Add Llama-2 tokenizer and config metadata for offline loading

Add Llama-2 tokenizer and config metadata for offline loading

Add CogACT-Large base checkpoint for future real-task training

Document base model dependencies and direct loading

Add root checkpoint alias for direct loading

Add runtime aliases and CogACT base metadata

Add runtime aliases and CogACT base metadata

Add runtime aliases and CogACT base metadata

Add runtime aliases and CogACT base metadata

Add runtime aliases and CogACT base metadata

Add runtime aliases and CogACT base metadata

Add runtime aliases and CogACT base metadata

Link reproducible GitHub code commit

Restore Model Card with exact runtime config paths

Document exact runtime config paths

Add exact runtime config for piper color sorting (config.json)

Add exact runtime config for piper color sorting (config.yaml)

Add latest Frozen EMA step-20000 checkpoint

Add configs/piper-color-sorting/dataset_statistics.json

Add configs/piper-color-sorting/training_config.json

Add configs/piper-color-sorting/training_config.yaml

Add README.md

initial commit