use top_k instead of return_all_scores

add note about evaluate.py

add config.json

add model weights

first commit

initial commit