Self-Improving is Often Sudden: Enlightenment-style Finetuning for Large-Scale Models

• URL: http://arxiv.org/abs/2607.13395v1
• Date: Wed, 15 Jul 2026 02:43:34 GMT
• Title: Self-Improving is Often Sudden: Enlightenment-style Finetuning for Large-Scale Models
• Abstract summary: We propose Enlightenment, a training-free post-tuning paradigm for large-scale models.<n>Our approach modifies shortcuts for key modules/layers without weight updates, while existing training-free ones predominantly manipulate attention weights.<n>Experiments show that Enlightenment efficiently unlocks the latent potential of pre-trained networks.
• Score: 23.533065315189223
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: The pursuit of autonomously self-improving models has attracted growing interest in the era of large-scale foundation models. Drawing inspiration from the concept of "enlightenment" or "aha moment" in human brain, we hypothesize that large models exhibit an analogous enlightenment phenomenon-a latent capacity for sudden capability boost. Then, we propose Enlightenment, a novel training-free post-tuning paradigm for large-scale models. Our approach modifies shortcuts for key modules/layers without weight updates, while existing training-free ones predominantly manipulate attention weights. We introduce two architecture-specific instantiations: i) For large language models, we propose attention head-mixing shortcuts that recalibrate attention weights by linking the initial attention head's output to all other target heads, modulated by an adaptive scaling factor initialization strategy. ii) For vision-language models, we apply a lightweight scalar-modulated factor to residual connections in the decoder layers, regulating information flow. Extensive experiments show that Enlightenment efficiently unlocks the latent potential of pre-trained networks, yielding remarkable performance improvements across diverse benchmarks and models.

Related papers

• Large Vision-Language Models Get Lost in Attention [51.851592109135716]
  We propose a unified framework grounded in information theory and geometry to quantify the geometric and entropic nature of residual updates.<n>Applying this unified framework reveals a fundamental functional decoupling: Attention acts as a subspace-language operator focused on reconfiguration, whereas FFNs serve as subspace-language operators driving semantic innovation.
  arXiv  Detail & Related papers  (2026-05-07T04:45:52Z)
• Self-Supervised Weight Templates for Scalable Vision Model Initialization [34.75805112986586]
  SWEET is a self-supervised framework that performs constraint-based pre-training to enable scalable initialization in vision tasks.<n>We introduce width-wise scaling, which regularizes the template along width-related dimensions and encourages robust, width-inwidth representations.<n>Experiments on textscclassification, textscsegmentation and textscgeneration tasks demonstrate the state-of-the-art performance of SWEET.
  arXiv  Detail & Related papers  (2026-01-27T15:15:17Z)
• Continual Learning for Generative AI: From LLMs to MLLMs and Beyond [56.29231194002407]
  We present a comprehensive survey of continual learning methods for mainstream generative AI models.<n>We categorize these approaches into three paradigms: architecture-based, regularization-based, and replay-based.<n>We analyze continual learning setups for different generative models, including training objectives, benchmarks, and core backbones.
  arXiv  Detail & Related papers  (2025-06-16T02:27:25Z)
• SMILE: Zero-Shot Sparse Mixture of Low-Rank Experts Construction From Pre-Trained Foundation Models [85.67096251281191]
  We present an innovative approach to model fusion called zero-shot Sparse MIxture of Low-rank Experts (SMILE) construction. SMILE allows for the upscaling of source models into an MoE model without extra data or further training. We conduct extensive experiments across diverse scenarios, such as image classification and text generation tasks, using full fine-tuning and LoRA fine-tuning.
  arXiv  Detail & Related papers  (2024-08-19T17:32:15Z)
• Expressive and Generalizable Low-rank Adaptation for Large Models via Slow Cascaded Learning [55.5715496559514]
  LoRA Slow Cascade Learning (LoRASC) is an innovative technique designed to enhance LoRA's expressiveness and generalization capabilities. Our approach augments expressiveness through a cascaded learning strategy that enables a mixture-of-low-rank adaptation, thereby increasing the model's ability to capture complex patterns.
  arXiv  Detail & Related papers  (2024-07-01T17:28:59Z)
• DEEM: Diffusion Models Serve as the Eyes of Large Language Models for Image Perception [66.88792390480343]
  We propose DEEM, a simple but effective approach that utilizes the generative feedback of diffusion models to align the semantic distributions of the image encoder.<n>DEEM exhibits enhanced robustness and a superior capacity to alleviate model hallucinations while utilizing fewer trainable parameters, less pre-training data, and a smaller base model size.
  arXiv  Detail & Related papers  (2024-05-24T05:46:04Z)
• Data-efficient Large Vision Models through Sequential Autoregression [58.26179273091461]
  We develop an efficient, autoregression-based vision model on a limited dataset. We demonstrate how this model achieves proficiency in a spectrum of visual tasks spanning both high-level and low-level semantic understanding. Our empirical evaluations underscore the model's agility in adapting to various tasks, heralding a significant reduction in the parameter footprint.
  arXiv  Detail & Related papers  (2024-02-07T13:41:53Z)
• Enhancing Dynamical System Modeling through Interpretable Machine Learning Augmentations: A Case Study in Cathodic Electrophoretic Deposition [0.8796261172196743]
  We introduce a comprehensive data-driven framework aimed at enhancing the modeling of physical systems. As a demonstrative application, we pursue the modeling of cathodic electrophoretic deposition (EPD), commonly known as e-coating.
  arXiv  Detail & Related papers  (2024-01-16T14:58:21Z)
• On the Embedding Collapse when Scaling up Recommendation Models [53.66285358088788]
  We identify the embedding collapse phenomenon as the inhibition of scalability, wherein the embedding matrix tends to occupy a low-dimensional subspace. We propose a simple yet effective multi-embedding design incorporating embedding-set-specific interaction modules to learn embedding sets with large diversity.
  arXiv  Detail & Related papers  (2023-10-06T17:50:38Z)
• Big-model Driven Few-shot Continual Learning [24.392821997721295]
  We propose a Big-model driven Few-shot Continual Learning (B-FSCL) framework to gradually evolve the model. Experimental results of our proposed B-FSCL on three popular datasets completely surpass all state-of-the-art FSCL methods.
  arXiv  Detail & Related papers  (2023-09-02T08:39:46Z)
• Revisiting Implicit Models: Sparsity Trade-offs Capability in Weight-tied Model for Vision Tasks [4.872984658007499]
  Implicit models such as Deep Equilibrium Models (DEQs) have garnered significant attention in the community for their ability to train infinite layer models. We revisit the line of implicit models and trace them back to the original weight-tied models. Surprisingly, we observe that weight-tied models are more effective, stable, as well as efficient on vision tasks, compared to the DEQ variants.
  arXiv  Detail & Related papers  (2023-07-16T11:45:35Z)
• DynamicEmbedding: Extending TensorFlow for Colossal-Scale Applications [0.0]
  One of the limitations of deep learning models with sparse features today stems from the predefined nature of their input. We show that the resulting models are able to perform better and efficiently run at a much larger scale.
  arXiv  Detail & Related papers  (2020-04-17T17:43:51Z)

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.