Benchmarking Generalization to New Tasks from Natural Language Instructions - View it on GitHub