Comment by loveparade

If you could easily benchmark the quality of code then models would be trained on these benchmarks/metrics.