1 / Log(loss)

Open ouirin opened this issue 8 months ago • 2 comments

if loss < 1.0, the result will become a negative number, then the compress ratio can not be calculated correctly.

Apr 21 '25 05:04 ouirin

I believe they used XX^T without dividing the number of samples. The truncation loss is quite big in that case and should avoid getting the negative problem. However, I didn't manage to reproduce their results.

Apr 21 '25 20:04 JeffreyWong20