Metraipolis › archive › dc_e9af0dee45aa
MechaTerp-TRACE: A Novel Approach for Component Ablation Analysis in Language Models — arXiv:2609.22163
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.22163
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-09T06:49:07.349Z
- SHA-256 of the archived text
a5d0cd0c29c7db65da8b148487d278b44dea10f862970b6788d4a2c99e41b682- Stable URL of this page
https://metraipolis.com/documento/dc_e9af0dee45aa
How to cite
MechaTerp-TRACE: A Novel Approach for Component Ablation Analysis in Language Models — arXiv:2609.22163 — arXiv (en), https://arxiv.org/abs/2609.22163 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_e9af0dee45aa (Metraipolis archive dc_e9af0dee45aa, sha256 a5d0cd0c29c7db65da8b148487d278b44dea10f862970b6788d4a2c99e41b682)
BibTeX
@misc{dce9af0dee45aa,
title = {MechaTerp-TRACE: A Novel Approach for Component Ablation Analysis in Language Models — arXiv:2609.22163},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.22163},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_e9af0dee45aa; sha256 a5d0cd0c29c7db65da8b148487d278b44dea10f862970b6788d4a2c99e41b682},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.22163. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Interpretability research on large language models has produced accounts of factual recall in feed-forward layers and of token relationships in self-attention, but little work offers a unified way to compare the causal contribution of different architecture components to a model's output. We introduce MechaTerp (the Mechanistic Interpretability suite) -TRACE (subset for Teacher-forced Registry of Ablated Component Effects), an architecture and study that measures how much each registered component of a language model supports the production of a named entity. TRACE ablates one component at a time and measures the resulting change in the output distribution at a fixed answer token, so component types from whole transformer blocks down to individual neurons and output logits can be compared on a common scale. We apply it to thirteen instruction-tuned dense decoder models spanning five families and one to thirty billion parameters, ablating 49,656 components across 48 medical and 42 general-knowledge prompts. We find that the components carrying the most effect are the same few, positionally fixed components in every model, regardless of which entity a prompt asks about, and that once these are removed, the remaining support is close to evenly spread in eleven of the thirteen models. Apparent localisation of entity knowledge is therefore largely attributable to generic generation machinery, which has direct consequences for methods that assume entity knowledge sits in a findable place, including targeted knowledge editing. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-22 in cs.CL). Categorie arXiv: Computation and Language (cs.CL). Pagina del lavoro: https://arxiv.org/abs/2609.22163 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-09T06:48:44.857Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-22?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).