Giles Thomas

Giles Thomas works through the nuts and bolts of large language models by building them from scratch.

Visit Giles Thomas

Feed URL

Paste it into your feed reader, or add it to Feedly, Inoreader or NewsBlur.

Recent posts

Fun with low-rank vocab matrices (and a bonus test loss reduction?) 9th October 2026
Why do OpenAI's GPT-2 weights beat mine? Part five: data quality 1st October 2026
Extending Raschka's GPT-2: an MoE trained from scratch on an RTX 3090 10th September 2026

Tags

Last posted: 9/10/2026