I made an explicit reverse pass (no autodiff), it was 8x faster in Python

hu3 · 2026-03-02T04:56:42 1772427402

I made an explicit double-reverse pass (no code!), it was 80x faster in my head!

bear3r · 2026-03-02T05:13:02 1772428382

tradeoff worth naming: you avoid the autodiff graph overhead (hence the speedup), but any architecture change means rewriting every gradient by hand. fine for a pedagogical project, but that's exactly why autodiff exists.

love2read · 2026-03-01T23:05:08 1772406308

Can you share a link?