aiqbench analysis
This research introduces a scalable "communication map" for transformer models that quantifies interaction channels between attention heads and neurons directly from weights. By identifying distinct functional communities and critical subspaces responsible for capabilities like in-context copying, the study advances mechanistic interpretability beyond manual circuit tracing. The method efficiently processes large models on consumer hardware, offering new tools for understanding internal model dynamics without relying on activation-based approximations.
The Communication Map of a Transformer
arXiv:2608.22007v1 Announce Type: cross Abstract: The components of a transformer communicate by writing to and reading from a shared residual stream, and mechanistic interpretability has mapped these connections by hand, one circuit at a time. We present the communication map, which charts every potential communication channel in a language model from weights alone, generalizing the composition score of Elhage et al. (2021) into a single coupling coefficient covering all 18 connection classes, from entire attention head circuits to single neurons. The census of all candidate channels, from $6.3\times10^{8}$ in GPT-2 to $1.3\times10^{11}$ in Pythia-6.9B, finds that 70-89% of head pairs are oriented far from chance, some coupled strongly and others actively avoiding each other. The full map costs 15 seconds for GPT-2 and 11 minutes for Pythia-6.9B on one consumer GPU. Two applications demonstrate the utility of the map. In Application 1, the strongest head-to-head couplings recover the known induction circuits blind and group them into communities, and ablating one such community destroys the model's in-context copying. In Application 2, pooling every head's coupling coefficients identifies a distinct two-dimensional stream subspace, whose deletion abolishes the induction capability in six models up to Pythia-6.9B. This subspace is different from those identified by either activation PCA or outlier dimensions. We release the map, the statistical machinery, and the intervention suite.
Abstract:The components of a transformer communicate by writing to and reading from a shared residual stream, and mechanistic interpretability has mapped these connections by hand, one circuit at a time. We present the communication map, which charts every potential communication channel in a language model from weights alone, generalizing the composition score of Elhage et al. (2021) into a single coupling coefficient covering all 18 connection classes, from entire attention head circuits to single neurons. The census of all candidate channels, from $6.3\times10^{8}$ in GPT-2 to $1.3\times10^{11}$ in Pythia-6.9B, finds that 70-89% of head pairs are oriented far from chance, some coupled strongly and others actively avoiding each other. The full map costs 15 seconds for GPT-2 and 11 minutes for Pythia-6.9B on one consumer GPU. Two applications demonstrate the utility of the map. In Application 1, the strongest head-to-head couplings recover the known induction circuits blind and group them into communities, and ablating one such community destroys the model's in-context copying. In Application 2, pooling every head's coupling coefficients identifies a distinct two-dimensional stream subspace, whose deletion abolishes the induction capability in six models up to Pythia-6.9B. This subspace is different from those identified by either activation PCA or outlier dimensions. We release the map, the statistical machinery, and the intervention suite.
28 pages. Code and results: this https URL
Machine Learning (cs.LG); Computation and Language (cs.CL)
arXiv:2608.22007 [cs.LG]
(or arXiv:2608.22007v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2608.22007 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
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 · 319 words
Related coverage
Accelerating Dropless MoE Training in JAX with NVIDIA Transformer Engine
NVIDIA · Sep 14, 2026
Latent reasoning architectures would likely undermine CoT, our strongest oversight tool
AI Alignment Forum · Sep 23, 2026
Article: Artificial Intelligence-Driven Phishing: How Phishing Technique Is Evolving and Implemented
InfoQ · Jun 8, 2026
Accelerating Transformers Fine-Tuning with NVIDIA NeMo AutoModel
Hugging Face · Jun 24, 2026