2 ms·I Pruned Layers from a 30B Code Model and It Got 123% Better at Benchmarks3 points by egonschiele 7mo ago