rome-target-layer-gpt2xl

IN premise — summaries/2026/08/24/meng-2022-rome-s3-interventions-on-weights-for-understanding-factual-associati.md

Created 2026-08-25T02:58:15+00:00

In the ROME experiments on GPT-2 XL (1.5B parameters), the editing layer was layer 18, and the model tested for main causal tracing results was GPT-2 XL.

Summary

This records the exact experimental setup behind the main causal tracing results: a single edit was applied at layer 18 of the 1.5B GPT-2 XL model. It matters because any attempt to reproduce, extend, or interpret those results must target that specific layer and model size rather than assuming a different configuration.