3 ms·

So marginally better (and occasionally worse) performance for an order of magnitude larger training costs…?

• watsonmusic 1y ago
  14b model performs comparably with 32b size. the improvement is huge
  • 85392_school 1y ago
    are we only comparing them in terms of text completion accuracy? does it also improve perf on benchmarks?