Releasing a new benchmark and data set for evaluating neural code search models