add new training files for possible new loading approach

commit my first llama-7b-hf model

initial commit