REVIEW 9 cited by
Interpreting Key Mechanisms of Factual Recall in Transformer-Based Language Models
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Interpreting Key Mechanisms of Factual Recall in Transformer-Based Language Models
read the original abstract
In this paper, we delve into several mechanisms employed by Transformer-based language models (LLMs) for factual recall tasks. We outline a pipeline consisting of three major steps: (1) Given a prompt ``The capital of France is,'' task-specific attention heads extract the topic token, such as ``France,'' from the context and pass it to subsequent MLPs. (2) As attention heads' outputs are aggregated with equal weight and added to the residual stream, the subsequent MLP acts as an ``activation,'' which either erases or amplifies the information originating from individual heads. As a result, the topic token ``France'' stands out in the residual stream. (3) A deep MLP takes ``France'' and generates a component that redirects the residual stream towards the direction of the correct answer, i.e., ``Paris.'' This procedure is akin to applying an implicit function such as ``get\_capital($X$),'' and the argument $X$ is the topic token information passed by attention heads. To achieve the above quantitative and qualitative analysis for MLPs, we proposed a novel analytic method aimed at decomposing the outputs of the MLP into components understandable by humans. Additionally, we observed a universal anti-overconfidence mechanism in the final layer of models, which suppresses correct predictions. We mitigate this suppression by leveraging our interpretation to improve factual recall confidence. The above interpretations are evaluated across diverse tasks spanning various domains of factual knowledge, using various language models from the GPT-2 families, 1.3B OPT, up to 7B Llama-2, and in both zero- and few-shot setups.
Forward citations
Cited by 9 Pith papers
-
Your UnEmbedding Matrix is Secretly a Feature Lens for Text Embeddings
EmbedFilter applies a linear filter derived from the LLM unembedding matrix to suppress high-frequency token influences in text embeddings, yielding improved zero-shot performance and inherent dimensionality reduction.
-
Revisiting Parameter-Based Knowledge Editing in Large Language Models: Theoretical Limits and Empirical Evidence
Parameter-based knowledge editing in LLMs induces reasoning collapse via dimensional collapse and is consistently outperformed by a retrieval baseline across varied edit counts, knowledge complexity, and evaluation metrics.
-
Relational Linear Properties in Language Models: An Empirical Investigation
A KL-divergence probing method shows relational linearity in language models varies across models and layers while being sensitive to relation phrasing, extending prior linear embedding work.
-
Context-Gated Associative Retrieval: From Theory to Transformers
Context gating in associative memories boosts inter-memory separation and sparsity for exponential retrieval gains, admits a unique fixed point driven by direct bias and feedback, and matches in-context learning dynam...
-
Do Transformers Use their Depth Adaptively? Evidence from a Relational Reasoning Task
Transformers show limited adaptive depth use on relational reasoning, with clearer evidence after finetuning on the task.
-
Deep sequence models tend to memorize geometrically; it is unclear why
Deep sequence models develop geometric memory in embeddings that encodes novel global relationships, transforming l-fold composition tasks into 1-step navigation via a natural spectral bias connected to Node2Vec.
-
Relational Linear Properties in Language Models: An Empirical Investigation
Introduces KL-divergence probing to test relational linearity and reports its variation across models, layers, and paraphrased queries on four datasets.
-
Tracing Relational Knowledge Recall in Large Language Models
Per-head attention contributions to the residual stream serve as strong linear features for classifying relational knowledge in LLMs, with probe accuracy correlating to relation specificity and signal distribution.
-
Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models
The survey organizes mechanistic interpretability techniques into a Locate-Steer-Improve framework to enable actionable improvements in LLM alignment, capability, and efficiency.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.