Remix.run Logo
PaulHoule 10 hours ago

I like the extreme boost that PyPy gets here which is most more than you usually get with PyPy, on one hand it shows the potential of PyPy but it is testing just one area where PyPy is strong, other workloads depend on performance that is not well optimized.

My take though is that branchy, old-AI, and heavy code (say keep every fact in a triple store and not as a Python object at the benefit of having an index for every permutation of ?s-?p-?o) that cannot be handled with numpy or similar tools benefits greatly from PyPy, but not 100x.