hendel-pythia-2.8b-architecture

IN premise — summaries/2026/08/24/hendel-2023-icl-task-vectors-sR-references.md

Created 2026-08-25T02:58:05+00:00

Pythia 2.8B has 2560 hidden dimensions, 32 layers, and 32 attention heads.

Summary

This pins down the fixed "shape" of the Pythia 2.8B model: how wide its internal text representations are at each step, how many sequential processing stages it applies, and how many parallel channels it uses to track relationships between words. These numbers set the baseline for any comparison with other models, memory-budget calculations, or claims about what the model can and cannot do.