Deconstructing Pre-training: Knowledge Attribution Analysis in MoE and Dense Models

• URL: http://arxiv.org/abs/2601.08383v1
• Date: Tue, 13 Jan 2026 09:44:00 GMT
• Title: Deconstructing Pre-training: Knowledge Attribution Analysis in MoE and Dense Models
• Abstract summary: Mixture-of-Experts (MoE) architectures decouple model capacity from per-token computation.<n>We present a time-resolved comparison of knowledge acquisition dynamics in MoE and dense architectures.<n>Masking the ten most important MoE attention heads reduces relational HIT@10 by 10%, compared with > 50% for the dense model.
• Score: 37.90956602792573
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Mixture-of-Experts (MoE) architectures decouple model capacity from per-token computation, enabling scaling beyond the computational limits imposed by dense scaling laws. Yet how MoE architectures shape knowledge acquisition during pre-training, and how this process differs from dense architectures, remains unknown. To address this issue, we introduce Gated-LPI (Log-Probability Increase), a neuron-level attribution metric that decomposes log-probability increase across neurons. We present a time-resolved comparison of knowledge acquisition dynamics in MoE and dense architectures, tracking checkpoints over 1.2M training steps (~ 5.0T tokens) and 600K training steps (~ 2.5T tokens), respectively. Our experiments uncover three patterns: (1) Low-entropy backbone. The top approximately 1% of MoE neurons capture over 45% of positive updates, forming a high-utility core, which is absent in the dense baseline. (2) Early consolidation. The MoE model locks into a stable importance profile within < 100K steps, whereas the dense model remains volatile throughout training. (3) Functional robustness. Masking the ten most important MoE attention heads reduces relational HIT@10 by < 10%, compared with > 50% for the dense model, showing that sparsity fosters distributed -- rather than brittle -- knowledge storage. These patterns collectively demonstrate that sparsity fosters an intrinsically stable and distributed computational backbone from early in training, helping bridge the gap between sparse architectures and training-time interpretability.

Related papers

• LLaDA MoE v2: Scaling Mixture-of-Experts Diffusion Language Models [129.64558694024126]
  Diffusion language models (dLLMs) offer an alternative to autoregressive (AR) language modeling.<n>We characterize how optimization hyper parameters, compute allocation, and architecture scale for MoE dLLMs.<n>We train LLaDA MoE v2, a 30B-A3B dLLM, from scratch on 23.5T tokens.
  arXiv  Detail & Related papers  (2026-08-04T10:53:02Z)
• Every Step Evolves: Scaling Reinforcement Learning for Trillion-Scale Thinking Model [100.86587937568832]
  Ring-1T is the first open-source, state-of-the-art thinking model with a trillion-scale parameter.<n>It features 1 trillion total parameters and activates approximately 50 billion per token.
  arXiv  Detail & Related papers  (2025-10-21T17:46:14Z)
• CAST: Continuous and Differentiable Semi-Structured Sparsity-Aware Training for Large Language Models [27.682531424487564]
  Sparsity-aware training is an effective approach for transforming large language models into hardware-friendly sparse patterns.<n>We propose Continuous Adaptive Sparse Trainer (CAST), a continuous and differentiable sparsity-aware training framework for sparse models.<n>Our results demonstrate significant improvements over previous state-of-the-art methods in both perplexity and zero-shot accuracy with minimal training resources.
  arXiv  Detail & Related papers  (2025-09-30T09:28:47Z)
• MobileLLM-R1: Exploring the Limits of Sub-Billion Language Model Reasoners with Open Training Recipes [60.57770396565211]
  We show that strong reasoning abilities can emerge with far less data.<n>MobileLLM-R50M achieves an AIME score of 15.5, compared to just 0.6 for OLMo-2-1.48B and 0.3 for SmolLM-2-1.7B.
  arXiv  Detail & Related papers  (2025-09-29T15:43:59Z)
• Curriculum Learning for Mesh-based simulations [2.867517731896504]
  Graph neural networks (GNNs) have emerged as powerful surrogates for mesh-based computational fluid dynamics (CFD)<n>We study a emphcoarse-to-fine curriculum that accelerates convergence by first training on very coarse meshes.<n>We achieve comparable accuracy while reducing total wall-clock time by up to 50%.
  arXiv  Detail & Related papers  (2025-09-16T14:54:11Z)
• Revolutionizing Reinforcement Learning Framework for Diffusion Large Language Models [49.911784762244814]
  TraceRL is a trajectory-aware reinforcement learning framework for diffusion language models (DLMs)<n>We derive a series of state-of-the-art diffusion language models, namely TraDo.<n>TraDo-8B-Instruct achieves relative accuracy improvements of 6.1% over Qwen2.5-7B-Instruct and 51.3% over Llama3.1-8B-Instruct on mathematical reasoning benchmarks.
  arXiv  Detail & Related papers  (2025-09-08T17:58:06Z)
• DualSparse-MoE: Coordinating Tensor/Neuron-Level Sparsity with Expert Partition and Reconstruction [15.261077484922616]
  Mixture of Experts (MoE) has become a mainstream architecture for building Large Language Models (LLMs)<n>We identify dual sparsity at the tensor and neuron levels in pre-trained MoE modules as a key factor for both accuracy and efficiency.<n>We propose DualSparse-MoE, an inference system that integrates dynamic tensor-level dropping with static neuron-level reconstruction.
  arXiv  Detail & Related papers  (2025-08-25T18:08:32Z)
• Towards Greater Leverage: Scaling Laws for Efficient Mixture-of-Experts Language Models [25.608085561102566]
  We introduce Leverage Efficiency (EL), a metric quantifying the computational advantage of an MoE model over a dense equivalent.<n>EL is driven by the expert activation ratio and the total compute budget, both following predictable power laws.<n>We integrate these discoveries into a unified scaling law that accurately predicts the EL of an MoE architecture based on its configuration.
  arXiv  Detail & Related papers  (2025-07-23T17:10:23Z)
• Can Mixture-of-Experts Surpass Dense LLMs Under Strictly Equal Resources? [58.56306556151929]
  Mixture-of-Experts (MoE) language models dramatically expand model capacity and achieve remarkable performance without increasing per-token compute.<n>Can MoEs surpass dense architectures under strictly equal resource constraints?<n>We show that an MoE model with activation rate in an optimal region is able to outperform its dense counterpart under the same total parameter, training compute and data resource.
  arXiv  Detail & Related papers  (2025-06-13T17:59:05Z)
• T1: Advancing Language Model Reasoning through Reinforcement Learning and Inference Scaling [52.34735382627312]
  Large language models (LLMs) have demonstrated remarkable capabilities in complex reasoning tasks.<n>Existing approaches mainly rely on imitation learning and struggle to achieve effective test-time scaling.<n>We present T1 to scale reinforcement learning by encouraging exploration and understand inference scaling.
  arXiv  Detail & Related papers  (2025-01-20T18:33:33Z)
• The Languini Kitchen: Enabling Language Modelling Research at Different Scales of Compute [66.84421705029624]
  We introduce an experimental protocol that enables model comparisons based on equivalent compute, measured in accelerator hours. We pre-process an existing large, diverse, and high-quality dataset of books that surpasses existing academic benchmarks in quality, diversity, and document length. This work also provides two baseline models: a feed-forward model derived from the GPT-2 architecture and a recurrent model in the form of a novel LSTM with ten-fold throughput.
  arXiv  Detail & Related papers  (2023-09-20T10:31:17Z)

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.