Freeze Deep, Train Shallow: Interpretable Layer Allocation for Continued Pre-Training

• URL: http://arxiv.org/abs/2605.11416v1
• Date: Tue, 12 May 2026 02:11:31 GMT
• Title: Freeze Deep, Train Shallow: Interpretable Layer Allocation for Continued Pre-Training
• Abstract summary: We propose LayerTracer to locate task execution positions and quantify layer sensitivity.<n>We show that deep layers act as critical regions for task execution and maintain high stability against disruptive updates.<n>We also present a hybrid model case study, which validates that placing high-quality pre-trained modules in deep layers effectively preserves inherent knowledge of the model.
• Score: 5.338568138489752
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Selective layer-wise updates are essential for low-cost continued pre-training of Large Language Models (LLMs), yet determining which layers to freeze or train remains an empirical black-box problem due to the lack of interpretable guidance. To address this issue, we propose LayerTracer, an architecture-agnostic diagnostic framework that reveals the evolution patterns of layer-wise representations and stability by locating task execution positions and quantifying layer sensitivity. Analysis results reveal that deep layers act as critical regions for task execution and maintain high stability against disruptive updates. Guided by this finding, we conduct three controlled continued pre-training trials to compare diverse freeze-train strategies, demonstrating that training shallow layers while freezing deep layers consistently outperforms full-parameter fine-tuning and the opposite allocation on both C-Eval and CMMLU benchmarks. We further present a hybrid model case study, which validates that placing high-quality pre-trained modules in deep layers effectively preserves inherent knowledge of the model. This work delivers a low-cost and interpretable solution for resource-constrained teams, offering actionable guidance for layer-wise parameter allocation in continued pre-training and hybrid model construction.

Related papers

• S3: Stable Subgoal Selection by Constraining Uncertainty of Coarse Dynamics in Hierarchical Reinforcement Learning [1.160208922584163]
  We study whether subgoal selection by the high-level agent can be performed more strategically, by providing it with dynamics-aware intrinsic motivation.<n>This approach stabilizes the high-level policy by learning to minimize the predictive uncertainty associated with the coarse dynamics.<n> Empirically, we observe that a dense, dynamics-aware intrinsic reward leads to risk-averse subgoal selection, enabling it to outperform state-of-the-art HRL methods in non-stationary long-horizon environments.
  arXiv  Detail & Related papers  (2026-07-21T16:03:34Z)
• Trust, but Verify: Peeling Low-Bit Transformer Networks for Training Monitoring [20.854288216118423]
  We propose a layer-wise peeling framework for training dynamics, in which each layer is locally optimized against representations of the trained model.<n>Experiments on decoder-only transformer models show that these layer-wise reference bounds can match or even surpass the trained model at various stages.
  arXiv  Detail & Related papers  (2026-05-04T17:30:50Z)
• SENTINEL: Stagewise Integrity Verification for Pipeline Parallel Decentralized Training [54.8494905524997]
  Decentralized training introduces critical security risks when executed across untrusted, geographically distributed nodes.<n>We propose SENTINEL, a verification mechanism for pipeline parallelism (PP) training without duplication.<n>Experiments demonstrate successful training of up to 4B- parameter LLMs across untrusted distributed environments with up to 176 workers while maintaining model convergence and performance.
  arXiv  Detail & Related papers  (2026-03-03T23:51:10Z)
• LESA: Learnable LLM Layer Scaling-Up [57.0510934286449]
  Training Large Language Models (LLMs) from scratch requires immense computational resources, making it prohibitively expensive.<n>Model scaling-up offers a promising solution by leveraging the parameters of smaller models to create larger ones.<n>We propose textbfLESA, a novel learnable method for depth scaling-up.
  arXiv  Detail & Related papers  (2025-02-19T14:58:48Z)
• Continuous Knowledge-Preserving Decomposition with Adaptive Layer Selection for Few-Shot Class-Incremental Learning [73.59672160329296]
  CKPD-FSCIL is a unified framework that unlocks the underutilized capacity of pretrained weights.<n>Our method consistently outperforms state-of-the-art approaches in both adaptability and knowledge retention.
  arXiv  Detail & Related papers  (2025-01-09T07:18:48Z)
• AdaRankGrad: Adaptive Gradient-Rank and Moments for Memory-Efficient LLMs Training and Fine-Tuning [9.51289606759621]
  Training and fine-tuning large language models (LLMs) come with challenges related to memory and computational requirements.<n>Various techniques have been developed to tackle these challenges, such as low-rank adaptation (LoRA)<n>We introduce a new method inspired by a phenomenon we formally prove: as training progresses, the rank of the estimated gradient gradually decreases.
  arXiv  Detail & Related papers  (2024-10-23T13:53:26Z)
• Avoiding mode collapse in diffusion models fine-tuned with reinforcement learning [0.0]
  Fine-tuning foundation models via reinforcement learning (RL) has proven promising for aligning to downstream objectives. We exploit the hierarchical nature of diffusion models (DMs) and train them dynamically at each epoch with a tailored RL method. We show that models trained with HRF achieve better preservation of diversity in downstream tasks, thus enhancing the fine-tuning robustness and at uncompromising mean rewards.
  arXiv  Detail & Related papers  (2024-10-10T19:06:23Z)
• 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)
• Robust Stochastically-Descending Unrolled Networks [85.6993263983062]
  Deep unrolling is an emerging learning-to-optimize method that unrolls a truncated iterative algorithm in the layers of a trainable neural network.<n>We show that convergence guarantees and generalizability of the unrolled networks are still open theoretical problems.<n>We numerically assess unrolled architectures trained under the proposed constraints in two different applications.
  arXiv  Detail & Related papers  (2023-12-25T18:51:23Z)
• Exploring Heterogeneous Characteristics of Layers in ASR Models for More Efficient Training [1.3999481573773072]
  We study the stability of these layers across runs and model sizes. We propose that group normalization may be used without disrupting their formation. We apply these findings to Federated Learning in order to improve the training procedure.
  arXiv  Detail & Related papers  (2021-10-08T17:25:19Z)
• Layer-wise Conditioning Analysis in Exploring the Learning Dynamics of DNNs [115.35745188028169]
  We extend conditioning analysis to deep neural networks (DNNs) in order to investigate their learning dynamics. We show that batch normalization (BN) can stabilize the training, but sometimes result in the false impression of a local minimum. We experimentally observe that BN can improve the layer-wise conditioning of the optimization problem.
  arXiv  Detail & Related papers  (2020-02-25T11:40:27Z)

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.