[watsonmusic]

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

[85392_school]

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