finding

active

finding:a-pair-of-query-and-key-subcomponents-distributed-across-attention-heads-performs-previous-token-behavior

A pair of query and key subcomponents distributed across attention heads performs previous-token behavior

VPD recovers an attention algorithm for attending to the previous token, distributed across multiple heads.

Source paper

extracted_from

cimcWhitepaper

Neighborhood — ranked by edge-count

Claims (1)

claim

Attention algorithms are usually distributed across attention heads
supports
Claim supported by VPD's recovery of cross-head attention subcomponents, noted in footnote.

Communities (4)

community

Mechanistic interpretability & model evaluation
members_of
Spans attention head decomposition, benchmark awareness, and genomic pathogenicity prediction via neural models.
Mechanistic structure of transformer attention computations
members_of
Identifies distributed algorithms implemented across attention heads, with focus on causal masking limitations and emergent capabilities via activation manifold steering.
Distributed attention head decomposition
members_of
Mechanistic interpretability approach decomposing attention heads into query/key subcomponents with distinct algorithmic roles
Distributed computation across attention heads
members_of
Studies how query, key, and value components decompose into specialized subfunctions across heads, enabling routing and token prediction behaviors.

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Previous-token attention behaviorconcept0.842
An attention algorithm recovered by VPD where the model attends to the immediately preceding token.
A pair of query and key subcomponents distributed across attention heads performs syntax-boundary routingfinding0.831
VPD recovers an attention algorithm for routing across syntactic boundaries, distributed across heads.
Some attention heads partially specialize in copying for words that split into two tokens without a space prefix, attending from fragmented token to complete tokenfinding0.822
Interesting special case of copying behavior related to tokenization artifacts; primitive precursor to induction heads
Attention computations distribute across heads via parameter subcomponents with interpretable rolesfinding0.806
Mechanistic discovery about how attention mechanisms decompose into interpretable parameter components.
Induction heads work by using K-composition with a previous token head to shift keys by one token, then matching the current destination token against shifted keys to predict what followsclaim0.778
The mechanistic explanation of how induction heads are implemented in two-layer models
Previous Token Headconcept0.774
An attention head that primarily attends to the immediately preceding token; key building block for induction heads via K-composition
Key, query, and value vectors are intermediary byproducts; W_OV and W_QK are the fundamental low-rank matrices describing attention head behaviorclaim0.770
Reframing observation: the canonical K/Q/V decomposition is computationally convenient but not the most interpretable representation
Each functional token is associated with an internalized visual operation, yet requires no visual supervision and remains a standard token in the tokenizer vocabulary.claim0.769
Describes the properties of the functional token.