aiqbench analysis
This study introduces MechaTerp-TRACE, a novel ablation framework that quantifies the causal contribution of specific transformer components to named entity generation across diverse model families. The findings reveal that entity knowledge is largely distributed within generic generation machinery rather than localized in specific neurons, challenging assumptions underlying targeted knowledge editing methods. This work advances mechanistic interpretability by providing a unified scale to compare architectural components from blocks down to individual logits.
MechaTerp-TRACE: A Novel Approach for Component Ablation Analysis in Language Models
arXiv:2609.22163v1 Announce Type: new Abstract: 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.
Abstract: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.
Computation and Language (cs.CL)
arXiv:2609.22163 [cs.CL]
(or arXiv:2609.22163v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.22163 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 321 words
