The Remarkable Robustness of LLMs: Stages of Inference?

• URL: http://arxiv.org/abs/2406.19384v3
• Date: Mon, 16 Jun 2025 10:21:00 GMT
• Title: The Remarkable Robustness of LLMs: Stages of Inference?
• Authors: Vedang Lad, Jin Hwa Lee, Wes Gurnee, Max Tegmark,
• Abstract summary: We investigate the robustness of Large Language Models (LLMs) to structural interventions by deleting and swapping adjacent layers during inference.<n>Surprisingly, models retain 72-95% of their original top-1 prediction accuracy without any fine-tuning.
• Score: 5.346230590800585
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We investigate the robustness of Large Language Models (LLMs) to structural interventions by deleting and swapping adjacent layers during inference. Surprisingly, models retain 72-95% of their original top-1 prediction accuracy without any fine-tuning. We find that performance degradation is not uniform across layers: interventions to the early and final layers cause the most degradation, while the model is remarkably robust to dropping middle layers. This pattern of localized sensitivity motivates our hypothesis of four stages of inference, observed across diverse model families and sizes: (1) detokenization, where local context is integrated to lift raw token embeddings into higher-level representations; (2) feature engineering, where task- and entity-specific features are iteratively refined; (3) prediction ensembling, where hidden states are aggregated into plausible next-token predictions; and (4) residual sharpening, where irrelevant features are suppressed to finalize the output distribution. Synthesizing behavioral and mechanistic evidence, we provide a framework for interpreting depth-dependent computations in LLMs.

Related papers

• Beyond Fully Supervised Pixel Annotations: Scribble-Driven Weakly-Supervised Framework for Image Manipulation Localization [11.10178274806454]
  We propose a form of weak supervision that improves the annotation efficiency and detection performance.<n>We re-annotated mainstream IML datasets with scribble labels and propose the first scribble-based IML dataset.<n>We employ self-supervised training with a structural consistency loss to encourage the model to produce consistent predictions.
  arXiv  Detail & Related papers  (2025-07-17T11:45:27Z)
• How Far Are We from Predicting Missing Modalities with Foundation Models? [31.853781353441242]
  Current foundation models often fall short in two critical aspects: (i) fine-grained semantic extraction from the available modalities, and (ii) robust validation of generated modalities.<n>This framework dynamically formulates modality-aware mining strategies based on the input context, facilitating the extraction of richer and more discriminative semantic features.<n> Experimental results show that our method reduces FID for missing image prediction by at least 14% and MER for missing text prediction by at least 10% compared to baselines.
  arXiv  Detail & Related papers  (2025-06-04T03:22:44Z)
• Spectral Insights into Data-Oblivious Critical Layers in Large Language Models [7.486925126518052]
  We introduce a data-oblivious approach to identify intrinsic critical layers in pre-fine-tuned language models.<n>We show that layers with significant shifts in representation space are also those most affected during fine-tuning.
  arXiv  Detail & Related papers  (2025-05-31T04:21:39Z)
• Adversarial Purification by Consistency-aware Latent Space Optimization on Data Manifolds [48.37843602248313]
  Deep neural networks (DNNs) are vulnerable to adversarial samples crafted by adding imperceptible perturbations to clean data, potentially leading to incorrect and dangerous predictions.<n>We propose Consistency Model-based Adversarial Purification (CMAP), which optimize vectors within the latent space of a pre-trained consistency model to generate samples for restoring clean data.<n>CMAP significantly enhances robustness against strong adversarial attacks while preserving high natural accuracy.
  arXiv  Detail & Related papers  (2024-12-11T14:14:02Z)
• Diffusion Forcing: Next-token Prediction Meets Full-Sequence Diffusion [61.03681839276652]
  Diffusion Forcing is a new training paradigm where a diffusion model is trained to denoise a set of tokens with independent per-token noise levels. We apply Diffusion Forcing to sequence generative modeling by training a causal next-token prediction model to generate one or several future tokens.
  arXiv  Detail & Related papers  (2024-07-01T15:43:25Z)
• Entropy Guided Extrapolative Decoding to Improve Factuality in Large Language Models [55.45444773200529]
  Large language models (LLMs) exhibit impressive natural language capabilities but suffer from hallucination. Recent work has focused on decoding techniques to improve factuality during inference.
  arXiv  Detail & Related papers  (2024-04-14T19:45:35Z)
• Identifiable Latent Neural Causal Models [82.14087963690561]
  Causal representation learning seeks to uncover latent, high-level causal representations from low-level observed data. We determine the types of distribution shifts that do contribute to the identifiability of causal representations. We translate our findings into a practical algorithm, allowing for the acquisition of reliable latent causal representations.
  arXiv  Detail & Related papers  (2024-03-23T04:13:55Z)
• When Neural Code Completion Models Size up the Situation: Attaining Cheaper and Faster Completion through Dynamic Model Inference [11.704110756342212]
  We propose a novel dynamic inference method specifically tailored for code completion models. It can averagely skip 1.7 layers out of 16 layers in the models, leading to an 11.2% speedup with only a marginal 1.1% reduction in ROUGE-L.
  arXiv  Detail & Related papers  (2024-01-18T13:26:53Z)
• When Does Confidence-Based Cascade Deferral Suffice? [69.28314307469381]
  Cascades are a classical strategy to enable inference cost to vary adaptively across samples. A deferral rule determines whether to invoke the next classifier in the sequence, or to terminate prediction. Despite being oblivious to the structure of the cascade, confidence-based deferral often works remarkably well in practice.
  arXiv  Detail & Related papers  (2023-07-06T04:13:57Z)
• Opening the Black Box: Analyzing Attention Weights and Hidden States in Pre-trained Language Models for Non-language Tasks [0.8889304968879164]
  We apply a pre-trained language model to constrained arithmetic problems with hierarchical structure, to analyze their attention weight scores and hidden states. The investigation reveals promising results, with the model addressing hierarchical problems in a moderately structured manner, similar to human problem-solving strategies. The attention analysis allows us to hypothesize that the model can generalize to longer sequences in ListOps dataset, a conclusion later confirmed through testing on sequences longer than those in the training set.
  arXiv  Detail & Related papers  (2023-06-21T11:48:07Z)
• Explaining Language Models' Predictions with High-Impact Concepts [11.47612457613113]
  We propose a complete framework for extending concept-based interpretability methods to NLP. We optimize for features whose existence causes the output predictions to change substantially. Our method achieves superior results on predictive impact, usability, and faithfulness compared to the baselines.
  arXiv  Detail & Related papers  (2023-05-03T14:48:27Z)
• Training Trajectories of Language Models Across Scales [99.38721327771208]
  Scaling up language models has led to unprecedented performance gains. How do language models of different sizes learn during pre-training? Why do larger language models demonstrate more desirable behaviors?
  arXiv  Detail & Related papers  (2022-12-19T19:16:29Z)
• Uncertainty-Aware Adaptation for Self-Supervised 3D Human Pose Estimation [70.32536356351706]
  We introduce MRP-Net that constitutes a common deep network backbone with two output heads subscribing to two diverse configurations. We derive suitable measures to quantify prediction uncertainty at both pose and joint level. We present a comprehensive evaluation of the proposed approach and demonstrate state-of-the-art performance on benchmark datasets.
  arXiv  Detail & Related papers  (2022-03-29T07:14:58Z)
• Neuro-Symbolic Entropy Regularization [78.16196949641079]
  In structured prediction, the goal is to jointly predict many output variables that together encode a structured object. One approach -- entropy regularization -- posits that decision boundaries should lie in low-probability regions. We propose a loss, neuro-symbolic entropy regularization, that encourages the model to confidently predict a valid object.
  arXiv  Detail & Related papers  (2022-01-25T06:23:10Z)
• Leveraging recent advances in Pre-Trained Language Models forEye-Tracking Prediction [0.0]
  Natural Language Pro-cessing uses human-derived behavioral data like eye-tracking data to augment the neural nets to solve arange of tasks spanning syntax and semantics. In this paper,we use the ZuCo 1.0 and ZuCo 2.0 dataset to explore differ-ent linguistic models to directly predict thesegaze features for each word with respect to itssentence.
  arXiv  Detail & Related papers  (2021-10-09T06:46:48Z)
• Generative Text Modeling through Short Run Inference [47.73892773331617]
  The present work proposes a short run dynamics for inference. It is variation from the prior distribution of the latent variable and then runs a small number of Langevin dynamics steps guided by its posterior distribution. We show that the models trained with short run dynamics more accurately model the data, compared to strong language model and VAE baselines, and exhibit no sign of posterior collapse.
  arXiv  Detail & Related papers  (2021-05-27T09:14:35Z)
• Explaining and Improving Model Behavior with k Nearest Neighbor Representations [107.24850861390196]
  We propose using k nearest neighbor representations to identify training examples responsible for a model's predictions. We show that kNN representations are effective at uncovering learned spurious associations. Our results indicate that the kNN approach makes the finetuned model more robust to adversarial inputs.
  arXiv  Detail & Related papers  (2020-10-18T16:55:25Z)
• Understanding Neural Abstractive Summarization Models via Uncertainty [54.37665950633147]
  seq2seq abstractive summarization models generate text in a free-form manner. We study the entropy, or uncertainty, of the model's token-level predictions. We show that uncertainty is a useful perspective for analyzing summarization and text generation models more broadly.
  arXiv  Detail & Related papers  (2020-10-15T16:57:27Z)
• Discrete Variational Attention Models for Language Generation [51.88612022940496]
  We propose a discrete variational attention model with categorical distribution over the attention mechanism owing to the discrete nature in languages. Thanks to the property of discreteness, the training of our proposed approach does not suffer from posterior collapse.
  arXiv  Detail & Related papers  (2020-04-21T05:49:04Z)

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.