Tommaso Vaccari
Blog

Notes and deep dives

Writeups derived from personal notes, then rewritten after I have built or studied something in enough depth.

Follow new posts via RSS

13 entries

2026

  • Integrating the Vinardo Scoring Function into muDock

    A technical deep dive into integrating Vinardo into muDock, from molecular representation and preprocessing to CPU scoring, validation against smina, and CASF-2016 results.

    C++ · HPC · Molecular Docking · Drug Discovery

  • Deep Dive into a Custom GPT Architecture

    A from-scratch GPT implementation, its computational cost, and experiments on arithmetic and character-level language modeling.

    Neural Networks · Deep Learning · Transformers · Attention · Python

  • Introduction to BLAS

    An introduction to BLAS routines, the roofline model, and SGEMM optimization

    HPC · Linear Algebra · C

  • From Self-Attention to a Small GPT

    A personal from-scratch derivation of causal self-attention, starting from a character language model and ending with the core pieces of a small GPT.

    Neural Networks · Deep Learning · Transformers · Attention · Python

  • Neural Network Training: Statistics, Initialization and Diagnostics

    A structured study of neural network training dynamics, covering statistical preliminaries, loss interpretation, saturation, gradient flow, initialization schemes, diagnostic tools, and stabilization mechanisms.

    Neural Networks · Deep Learning · Optimization · Initialization · Batch Normalization

  • Engineering MovHex: Modeling and Optimizing a Dynamic Hex-Grid Router in C

    Engineering a dynamic routing engine over a hexagonal map in C, from geometric modeling to shortest-path optimization.

    C · Graph Algorithms · Data Structures · Performance

2025

2024