- Rating
- 2(+1)
- Confidence
- 3
- Informativeness
- 1
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper introduces two scalar quantities, \rho and \alpha, which are derived from the solely from the query-key weights of RoPE-based attention. The authors interpret ρ as positional capacity and α as the definiteness of content matching. The results show separation of three canonical tasks, induction-heads, duplicate-token, and previous-token, across five language models, together with cross-model transfer and a selective RoPE-scaling application.
Critical strengths
- The scalar quantities are independent of the activations, and depend only on the weights of the attention-heads. Much of mechanistic interpretability relies on carefully designed inputs, forward passes, activation inspection, or interventions, and this complementary parameter-space approach is interesting.
- The authors evaluate five models spanning four architecture families, rather than demonstrating the method on a single transformer, and the empirical consistencies well validated.
- I liked the unsupervised clustering experiment and an attempt to connect the proposed descriptors to an actionable inference-time modification.
Critical weaknesses
- The theoretical presentation needs to be improved. Since most results are focused on RoPE based architectures, complete mathematical definitions need to be provided. Other notation like SO(2)^{d/2}, and important quantities like G, P etc. need to be properly defined before using directly in equations.
- The paper contains errors in its theoretical derivations. Equation (5) assumes that the squared Frobenius norm of a sum of matrices equals the sum of their squared norms, ignoring cross-terms, which is incorrect. Furthermore, Proposition 1 only establishes that \rho=0 eliminates positional dependence at the sampled offsets D, not at all relative positions. The stronger claim that positional computation necessarily requires \rho>0 is therefore not established for the paper's definition of \rho.
- Similarly, the claimed forbidden region of near-zero \rho needs more explanation or a quantitative theorem.
- The majority of attention heads in Figure 1 belong to the "other" category, which is excluded from the primary classification evaluation. The method is therefore evaluated on a relatively small, selected subset of the model's heads. Its usefulness for general-purpose head characterization is less clear.
- The paper does not adequately compare against existing weight-based head-characterization methods, conventional QK similarity measures, spectral statistics, RoPE-frequency features, or simple architectural baselines such as layer index. Without these comparisons, it is difficult to determine whether the proposed invariants provide genuinely new information beyond QK geometry.
- Although the authors demonstrate correlations between (\rho,\alpha) and three behavioral head types, these observations do not establish that the invariants causally determine head functionality. For example, a positive-semidefinite symmetric QK component does not necessarily imply duplicate-token detection, and induction behavior can depend on interactions between multiple attention heads and the value-output pathway.
Questions for authors
Please see weaknesses above.
Reply to Reviewer cJEM
We thank the reviewer for the thoughtful review. We answer the six weaknesses in order.
We define WQ,WK∈Rdh×d, R(Δ), GQK(Δ)=WQ⊤R(Δ)WK, P(Δ)=GQK(Δ)−GQK(0) and the symmetry group before first use. The group is (C∗)dh/2 (per-plane rotation with reciprocal rescaling of WQ and WK), which contains the SO(2)dh/2 of the paper. Since ρ and α are functions of GQK, they are invariant under all of it. No number changes.
Eq. 5 is wrong and we remove Eq. 5. You are also right that Proposition 1 holds on the sampled offsets D, for fixed inputs. We state it that way and add a bound: for every Δ∈D, ∣xi⊤P(Δ)xj∣≤∣D∣ρ∥GQK(0)∥F∥xi∥∥xj∥. So ρ>0 is necessary for a head to compute relative position through its own RoPE pathway on D. We do not claim other offsets, or position that reaches a head through its inputs.
The bound is the quantitative statement, and we drop "forbidden region" and "without exception". Table 2 is an empirical separation among labeled heads, not a theorem: previous-token floors are 0.27 to 0.74, the smallest ρ over all heads is 0.004 to 0.024, and the floor is that class's own minimum.
Yes, most heads are unlabeled and excluded from the supervised evaluation. We claim only that the three labeled types occupy the same relative regions across five models, not general-purpose head characterization. We state this in the limitations.
A comparison:
High-frequency energy is strong on the positional pairs, as expected for a proxy of ρ. On the content pair (induction vs. duplicate) no baseline exceeds 0.92 on the 3 models, against 0.98 to 1.00 for (ρ,α). We did not run published methods and we did not find a directly comparable method (they were either VO, or plane specific).