Hacker News Reader

14b model performs comparably with 32b size. the improvement is huge

are we only comparing them in terms of text completion accuracy? does it also improve perf on benchmarks?