Add model card

Add inference checkpoint (optimizer/cache stripped)

initial commit