Details in smart_quantizer
Opened this issue · 0 comments
TangHengxuan commented
Hello, I would like to ask you some questions.
- The “choose p=1%” in "variance reduction" mentioned in the README: does "p" refer to the sparsity?which means want to reduce the complexity by reducing the "p"?
- In “The ZipML Framework for Training Models with End-to-End Low Precision: The Cans, the Cannots, and a Little Bit of Deep Learning", the paper mentions "2-Approximation in Almost-Linear Time" algorithm (the complexity is O(n log n)). Is there any relevant code? I read the pseudo code, but still not clear.
- I want to confirm if the "smart_quantizer" code implements the algorithm via dynamic programming? It doesn't look like the algorithm used "2-Approximation in Almost-Linear Time".
Looking forward to your reply. Please forgive me for a stupid choice of words.