Sparse or Dense? A Mechanistic Estimation of Computation Density in Transformer-based LLMs

• URL: http://arxiv.org/abs/2601.22795v1
• Date: Fri, 30 Jan 2026 10:18:29 GMT
• Title: Sparse or Dense? A Mechanistic Estimation of Computation Density in Transformer-based LLMs
• Abstract summary: Transformer-based large language models (LLMs) are comprised of billions of parameters arranged in deep and wide computational graphs.<n>We introduce here a technique to systematically quantify computation density in LLMs.
• Score: 7.375393542449447
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Transformer-based large language models (LLMs) are comprised of billions of parameters arranged in deep and wide computational graphs. Several studies on LLM efficiency optimization argue that it is possible to prune a significant portion of the parameters, while only marginally impacting performance. This suggests that the computation is not uniformly distributed across the parameters. We introduce here a technique to systematically quantify computation density in LLMs. In particular, we design a density estimator drawing on mechanistic interpretability. We experimentally test our estimator and find that: (1) contrary to what has been often assumed, LLM processing generally involves dense computation; (2) computation density is dynamic, in the sense that models shift between sparse and dense processing regimes depending on the input; (3) per-input density is significantly correlated across LLMs, suggesting that the same inputs trigger either low or high density. Investigating the factors influencing density, we observe that predicting rarer tokens requires higher density, and increasing context length often decreases the density. We believe that our computation density estimator will contribute to a better understanding of the processing at work in LLMs, challenging their symbolic interpretation.

Related papers

• Tracing Computation Density in LLMs [8.097942585959956]
  Transformer-based large language models (LLMs) are comprised of billions of parameters arranged in deep and wide computational graphs.<n>We introduce the s-Trace method to efficiently estimate the subgraph of size s that best approximates a full model output.
  arXiv  Detail & Related papers  (2026-05-26T13:55:08Z)
• SparseLoRA: Accelerating LLM Fine-Tuning with Contextual Sparsity [52.88892280536302]
  We introduce SparseLoRA, a method that accelerates fine-tuning through contextual sparsity.<n>We show that SparseLoRA reduces computational cost by up to 2.2 times and a measured speedup of up to 1.6 times.
  arXiv  Detail & Related papers  (2025-06-19T17:53:34Z)
• Densing Law of LLMs [81.06644243978101]
  Large Language Models (LLMs) have emerged as a milestone in artificial intelligence, and their performance can improve as the model size increases.<n>This paper introduces the concept of textitcapacity density'' as a new metric to evaluate the quality of the LLMs across different scales.
  arXiv  Detail & Related papers  (2024-12-05T16:31:13Z)
• Density estimation with LLMs: a geometric investigation of in-context learning trajectories [3.281128493853064]
  Large language models (LLMs) demonstrate remarkable emergent abilities to perform in-context learning across various tasks.<n>This work investigates LLMs' ability to estimate probability density functions from data observed in-context.<n>We leverage the Intensive Principal Component Analysis (InPCA) to visualize and analyze the in-context learning dynamics of LLaMA-2 models.
  arXiv  Detail & Related papers  (2024-10-07T17:22:56Z)
• STBLLM: Breaking the 1-Bit Barrier with Structured Binary LLMs [28.70239743254508]
  We present the first structural binarization method for LLM compression to less than 1-bit precision. We observe that some weights in binarized LLMs can be randomly flipped without significant performance degradation. Our approach performs better than other compressed binarization methods while significantly reducing memory requirements.
  arXiv  Detail & Related papers  (2024-08-03T15:07:44Z)
• Enabling High-Sparsity Foundational Llama Models with Efficient Pretraining and Deployment [56.44025052765861]
  Large language models (LLMs) have revolutionized Natural Language Processing (NLP), but their size creates computational bottlenecks. We introduce a novel approach to create accurate, sparse foundational versions of performant LLMs. We show a total speedup on CPUs for sparse-quantized LLaMA models of up to 8.6x.
  arXiv  Detail & Related papers  (2024-05-06T16:03:32Z)
• Scaling Sparse Fine-Tuning to Large Language Models [67.59697720719672]
  Large Language Models (LLMs) are difficult to fully fine-tune due to their sheer number of parameters. We propose SpIEL, a novel sparse finetuning method which maintains an array of parameter indices and the deltas of these parameters relative to their pretrained values. We show that SpIEL is superior to popular parameter-efficient fine-tuning methods like LoRA in terms of performance and comparable in terms of run time.
  arXiv  Detail & Related papers  (2024-01-29T18:43:49Z)
• Compress, Then Prompt: Improving Accuracy-Efficiency Trade-off of LLM Inference with Transferable Prompt [96.24800696597707]
  We introduce a new perspective to optimize this trade-off by prompting compressed models. We propose a soft prompt learning method where we expose the compressed model to the prompt learning process. Our experimental analysis suggests our soft prompt strategy greatly improves the performance of the 8x compressed LLaMA-7B model.
  arXiv  Detail & Related papers  (2023-05-17T20:45:13Z)
• Quantum Adaptive Fourier Features for Neural Density Estimation [0.0]
  This paper presents a method for neural density estimation that can be seen as a type of kernel density estimation. The method is based on density matrices, a formalism used in quantum mechanics, and adaptive Fourier features. The method was evaluated in different synthetic and real datasets, and its performance compared against state-of-the-art neural density estimation methods.
  arXiv  Detail & Related papers  (2022-08-01T01:39:11Z)
• TAKDE: Temporal Adaptive Kernel Density Estimator for Real-Time Dynamic Density Estimation [16.45003200150227]
  We name the temporal adaptive kernel density estimator (TAKDE) TAKDE is theoretically optimal in terms of the worst-case AMISE. We provide numerical experiments using synthetic and real-world datasets, showing that TAKDE outperforms other state-of-the-art dynamic density estimators.
  arXiv  Detail & Related papers  (2022-03-15T23:38:32Z)
• Density-Based Clustering with Kernel Diffusion [59.4179549482505]
  A naive density corresponding to the indicator function of a unit $d$-dimensional Euclidean ball is commonly used in density-based clustering algorithms. We propose a new kernel diffusion density function, which is adaptive to data of varying local distributional characteristics and smoothness.
  arXiv  Detail & Related papers  (2021-10-11T09:00:33Z)
• Learning Optical Flow from a Few Matches [67.83633948984954]
  We show that the dense correlation volume representation is redundant and accurate flow estimation can be achieved with only a fraction of elements in it. Experiments show that our method can reduce computational cost and memory use significantly, while maintaining high accuracy.
  arXiv  Detail & Related papers  (2021-04-05T21:44:00Z)

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.