implementation
0nutation opened this issue · 1 comments
0nutation commented
I write a pytorch version of Lookup-Free Quantization in https://github.com/0nutation/Lookup-Free-Quantization.
lucidrains commented
@0nutation looks good! so torch sign produces 0 for 0, which is an edge case you'll have to handle (one way is (x > 0) * 2 - 1
)