Secret mixtures of experts inside your LLM

• URL: http://arxiv.org/abs/2512.18452v1
• Date: Sat, 20 Dec 2025 17:53:24 GMT
• Title: Secret mixtures of experts inside your LLM
• Authors: Enric Boix-Adsera,
• Abstract summary: Multilayer Perceptron (MLP) is arguably one of the least understood parts of the transformer architecture.<n>This paper seeks to understand the layers in dense LLM models by hypothesizing that these layers secretly approximately perform a sparse computation.
• Score: 0.11603243575080534
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Despite being one of the earliest neural network layers, the Multilayer Perceptron (MLP) is arguably one of the least understood parts of the transformer architecture due to its dense computation and lack of easy visualization. This paper seeks to understand the MLP layers in dense LLM models by hypothesizing that these layers secretly approximately perform a sparse computation -- namely, that they can be well approximated by sparsely-activating Mixture of Experts (MoE) layers. Our hypothesis is based on a novel theoretical connection between MoE models and Sparse Autoencoder (SAE) structure in activation space. We empirically validate the hypothesis on pretrained LLMs, and demonstrate that the activation distribution matters -- these results do not hold for Gaussian data, but rather rely crucially on structure in the distribution of neural network activations. Our results shine light on a general principle at play in MLP layers inside LLMs, and give an explanation for the effectiveness of modern MoE-based transformers. Additionally, our experimental explorations suggest new directions for more efficient MoE architecture design based on low-rank routers.

Related papers

• Detailed balance in large language model-driven agents [1.2687030176231846]
  Large language model (LLM)-driven agents are emerging as a powerful new paradigm for solving complex problems.<n>This Letter proposes a method to estimate the underlying generative directionality of LLMs embedded within agents.
  arXiv  Detail & Related papers  (2025-12-10T20:04:23Z)
• RL makes MLLMs see better than SFT [96.508432109136]
  We conduct a critical yet under-explored analysis of the vision encoder of Multimodal Language Model (MLLM)<n>Our results demonstrate that MLLM's post-training strategy (i.e., SFT or RL) not only leads to distinct outcomes on MLLM downstream tasks, but also fundamentally reshapes MLLM's underlying visual representations.<n>We then reframe our findings into a simple recipe for building strong vision encoders for MLLMs, Preference-Instructed Vision OpTimization (PIVOT)
  arXiv  Detail & Related papers  (2025-10-18T03:37:17Z)
• LLMs for Bayesian Optimization in Scientific Domains: Are We There Yet? [25.116042999105193]
  Large language models (LLMs) have been proposed as general-purpose agents for experimental design.<n>We evaluate this hypothesis using both open- and closed-source instruction-tuned LLMs applied to genetic perturbation and molecular property discovery tasks.<n>We find that LLM-based agents show no sensitivity to experimental feedback, replacing true outcomes with randomly permuted labels has no impact on performance.
  arXiv  Detail & Related papers  (2025-09-24T15:50:17Z)
• Cluster Topology-Driven Placement of Experts Reduces Network Traffic in MoE Inference [49.141930185079325]
  We propose an integer linear program (ILP) that determines the optimal placement of experts, minimizing the expected number of transmissions.<n>We demonstrate that ILP-based placement strategy yields lower network traffic than competitors for small-scale (DeepSeekMoE16B) and large-scale (DeepSeek-R1671B) models.
  arXiv  Detail & Related papers  (2025-08-12T07:08:48Z)
• Bifrost-1: Bridging Multimodal LLMs and Diffusion Models with Patch-level CLIP Latents [55.82787697101274]
  Bifrost-1 is a unified framework that bridges pretrained multimodal LLMs (MLLMs) and diffusion models.<n>By seamlessly integrating pretrained MLLMs and diffusion models with patch-level CLIP latents, our framework enables high-fidelity controllable image generation.<n>Our experiments demonstrate that Bifrost-1 achieves comparable or better performance than previous methods in terms of visual fidelity and multimodal understanding.
  arXiv  Detail & Related papers  (2025-08-08T02:38:47Z)
• Training-Free Reasoning and Reflection in MLLMs [45.134271969594614]
  This paper introduces FRANK Model, a training-FRee ANd r1-liKe MLLM that imbues off-the-shelf MLLMs with reasoning and reflection abilities.<n>Our key insight is to decouple perception and reasoning across MLLM decoder layers.<n>To this end, we propose a layer-wise, Taylor-derived closed-form fusion mechanism that integrates reasoning capacity into deep decoder layers.
  arXiv  Detail & Related papers  (2025-05-22T02:51:12Z)
• MoLe-VLA: Dynamic Layer-skipping Vision Language Action Model via Mixture-of-Layers for Efficient Robot Manipulation [24.200547898713126]
  Multimodal Large Language Models (MLLMs) excel in understanding complex language and visual data.<n>Their real-world deployment is hindered by substantial computational and storage demands.<n>We propose a Mixture-of-Layers Vision-Language-Action model (MoLe) architecture for dynamic LLM layer activation.
  arXiv  Detail & Related papers  (2025-03-26T10:05:38Z)
• Mitigating Forgetting in LLM Fine-Tuning via Low-Perplexity Token Learning [65.23593936798662]
  We show that fine-tuning with LLM-generated data improves target task performance and reduces non-target task degradation.<n>This is the first work to provide an empirical explanation based on token perplexity reduction to mitigate catastrophic forgetting in LLMs after fine-tuning.
  arXiv  Detail & Related papers  (2025-01-24T08:18:56Z)
• AlphaPruning: Using Heavy-Tailed Self Regularization Theory for Improved Layer-wise Pruning of Large Language Models [94.82766517752418]
  We propose AlphaPruning, which uses shape metrics to allocate layerwise sparsity ratios in a more theoretically principled manner. Our results show that AlphaPruning prunes LLaMA-7B to 80% sparsity while maintaining reasonable perplexity, marking a first in the literature on LLMs.
  arXiv  Detail & Related papers  (2024-10-14T03:35:11Z)
• 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.<n>However, the mechanisms by which these models encode tasks of varying complexities remain poorly understood.<n>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)
• Streamlining Redundant Layers to Compress Large Language Models [21.27944103424621]
  This paper introduces LLM-Streamline, a pioneer work on layer pruning for large language models (LLMs)<n>It is based on the observation that different layers have varying impacts on hidden states, enabling the identification of less important layers to be pruned.<n>Experiments show that LLM-Streamline outperforms both previous and concurrent state-of-the-art pruning methods in terms of both performance and training efficiency.
  arXiv  Detail & Related papers  (2024-03-28T04:12:13Z)
• ShortGPT: Layers in Large Language Models are More Redundant Than You Expect [38.148626520751385]
  We show that many layers of Large Language Models (LLMs) exhibit high similarity, and some layers play a negligible role in network functionality. We propose a straightforward pruning approach: layer removal, in which we directly delete the redundant layers. Experiments demonstrate that our method, which we call ShortGPT, significantly outperforms previous state-of-the-art (SOTA) methods in model pruning.
  arXiv  Detail & Related papers  (2024-03-06T17:04:18Z)
• Characterizing Truthfulness in Large Language Model Generations with Local Intrinsic Dimension [63.330262740414646]
  We study how to characterize and predict the truthfulness of texts generated from large language models (LLMs) We suggest investigating internal activations and quantifying LLM's truthfulness using the local intrinsic dimension (LID) of model activations.
  arXiv  Detail & Related papers  (2024-02-28T04:56:21Z)

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.