Frei aber einsam
Notes on hardware, cryptography, and the mathematics underneath.
Research
-
The four parts of a transformer — embedding, attention, MLPs, and unembedding — worked through with GPT-3's numbers, accounting for all 175 billion parameters.
-
Tensor product is a good and powerful way to describe and analyze NTTs. This note gives the basics on using tensor product to present NTT and derive the famous constant-geometry using this presentation.
Reading
Coming soon.
Field Notes
Coming soon.