Emergent effects of scaling on the functional hierarchies within large language models

• URL: http://arxiv.org/abs/2501.07359v1
• Date: Mon, 13 Jan 2025 14:27:39 GMT
• Title: Emergent effects of scaling on the functional hierarchies within large language models
• Authors: Paul C. Bogdan,
• Abstract summary: Analyses using a small model (Llama-3.2-3b; 28 layers) Item-level semantics are most strongly represented early (layers 2-7), then two-item relations (layers 8-12), and then four-item analogies (layers 10-15) Deep layers also compress information from early portions of the context window without meaningful abstraction.
• Score: 0.0
• License:
• Abstract: Large language model (LLM) architectures are often described as functionally hierarchical: Early layers process syntax, middle layers begin to parse semantics, and late layers integrate information. The present work revisits these ideas. This research submits simple texts to an LLM (e.g., "A church and organ") and extracts the resulting activations. Then, for each layer, support vector machines and ridge regressions are fit to predict a text's label and thus examine whether a given layer encodes some information. Analyses using a small model (Llama-3.2-3b; 28 layers) partly bolster the common hierarchical perspective: Item-level semantics are most strongly represented early (layers 2-7), then two-item relations (layers 8-12), and then four-item analogies (layers 10-15). Afterward, the representation of items and simple relations gradually decreases in deeper layers that focus on more global information. However, several findings run counter to a steady hierarchy view: First, although deep layers can represent document-wide abstractions, deep layers also compress information from early portions of the context window without meaningful abstraction. Second, when examining a larger model (Llama-3.3-70b-Instruct), stark fluctuations in abstraction level appear: As depth increases, two-item relations and four-item analogies initially increase in their representation, then markedly decrease, and afterward increase again momentarily. This peculiar pattern consistently emerges across several experiments. Third, another emergent effect of scaling is coordination between the attention mechanisms of adjacent layers. Across multiple experiments using the larger model, adjacent layers fluctuate between what information they each specialize in representing. In sum, an abstraction hierarchy often manifests across layers, but large models also deviate from this structure in curious ways.

Related papers

• The Representation and Recall of Interwoven Structured Knowledge in LLMs: A Geometric and Layered Analysis [0.0]
  Large language models (LLMs) represent and recall multi-associated attributes across transformer layers. intermediate layers encode factual knowledge by superimposing related attributes in overlapping spaces. later layers refine linguistic patterns and progressively separate attribute representations.
  arXiv  Detail & Related papers  (2025-02-15T18:08:51Z)
• Layer by Layer: Uncovering Hidden Representations in Language Models [28.304269706993942]
  We show that intermediate layers can encode even richer representations, often improving performance on a wide range of downstream tasks. Our framework highlights how each model layer balances information compression and signal preservation. These findings challenge the standard focus on final-layer embeddings and open new directions for model analysis and optimization.
  arXiv  Detail & Related papers  (2025-02-04T05:03:42Z)
• Learning Visual Hierarchies with Hyperbolic Embeddings [28.35250955426006]
  We introduce a learning paradigm that can encode user-defined multi-level visual hierarchies in hyperbolic space without requiring explicit hierarchical labels. We show significant improvements in hierarchical retrieval tasks, demonstrating the capability of our model in capturing visual hierarchies.
  arXiv  Detail & Related papers  (2024-11-26T14:58:06Z)
• Looking into Black Box Code Language Models [2.5324062203985935]
  We use two state-of-the-art codeLMs, Codegen-Mono and Ploycoder, and three widely used programming languages, Java, Go, and Python. We show concepts of interest can be edited within feed-forward layers without compromising codeLM performance.
  arXiv  Detail & Related papers  (2024-07-05T21:13:41Z)
• Talking Heads: Understanding Inter-layer Communication in Transformer Language Models [32.2976613483151]
  We analyze a mechanism used in two LMs to selectively inhibit items in a context in one task. We find that models write into low-rank subspaces of the residual stream to represent features which are then read out by later layers.
  arXiv  Detail & Related papers  (2024-06-13T18:12:01Z)
• Exploring Concept Depth: How Large Language Models Acquire Knowledge and Concept at Different Layers? [57.04803703952721]
  Large language models (LLMs) have shown remarkable performances across a wide range of tasks. However, the mechanisms by which these models encode tasks of varying complexities remain poorly understood. We introduce the idea of "Concept Depth" to suggest that more complex concepts are typically acquired in deeper layers.
  arXiv  Detail & Related papers  (2024-04-10T14:56:40Z)
• Fantastic Semantics and Where to Find Them: Investigating Which Layers of Generative LLMs Reflect Lexical Semantics [50.982315553104975]
  We investigate the bottom-up evolution of lexical semantics for a popular large language model, namely Llama2. Our experiments show that the representations in lower layers encode lexical semantics, while the higher layers, with weaker semantic induction, are responsible for prediction. This is in contrast to models with discriminative objectives, such as mask language modeling, where the higher layers obtain better lexical semantics.
  arXiv  Detail & Related papers  (2024-03-03T13:14:47Z)
• TopicNet: Semantic Graph-Guided Topic Discovery [51.71374479354178]
  Existing deep hierarchical topic models are able to extract semantically meaningful topics from a text corpus in an unsupervised manner. We introduce TopicNet as a deep hierarchical topic model that can inject prior structural knowledge as an inductive bias to influence learning.
  arXiv  Detail & Related papers  (2021-10-27T09:07:14Z)
• Exploring the Hierarchy in Relation Labels for Scene Graph Generation [75.88758055269948]
  The proposed method can improve several state-of-the-art baselines by a large margin (up to $33%$ relative gain) in terms of Recall@50. Experiments show that the proposed simple yet effective method can improve several state-of-the-art baselines by a large margin.
  arXiv  Detail & Related papers  (2020-09-12T17:36:53Z)
• Fine-grained Video-Text Retrieval with Hierarchical Graph Reasoning [72.52804406378023]
  Cross-modal retrieval between videos and texts has attracted growing attentions due to the rapid emergence of videos on the web. To improve fine-grained video-text retrieval, we propose a Hierarchical Graph Reasoning model, which decomposes video-text matching into global-to-local levels.
  arXiv  Detail & Related papers  (2020-03-01T03:44:19Z)
• Cross-layer Feature Pyramid Network for Salient Object Detection [102.20031050972429]
  We propose a novel Cross-layer Feature Pyramid Network to improve the progressive fusion in salient object detection. The distributed features per layer own both semantics and salient details from all other layers simultaneously, and suffer reduced loss of important information.
  arXiv  Detail & Related papers  (2020-02-25T14:06:27Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.