2 ms·

Is it possible to offload some layers to CPU and still train in a reasonable amount of time?

• generalizations 3y ago
  There’s also that pruning tool that was on hn in the last couple weeks. It seemed to work really well on the larger models, and could reduce size by 30-50%