Evolution of Concepts in Language Model Pre-Training

• URL: http://arxiv.org/abs/2509.17196v1
• Date: Sun, 21 Sep 2025 18:53:12 GMT
• Title: Evolution of Concepts in Language Model Pre-Training
• Authors: Xuyang Ge, Wentao Shu, Jiaxing Wu, Yunhua Zhou, Zhengfu He, Xipeng Qiu,
• Abstract summary: We track linear interpretable feature evolution across pre-training snapshots using a sparse dictionary learning method called crosscoders.<n>We find that most features begin to form around a specific point, while more complex patterns emerge in later training stages.
• Score: 53.994470178155105
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Language models obtain extensive capabilities through pre-training. However, the pre-training process remains a black box. In this work, we track linear interpretable feature evolution across pre-training snapshots using a sparse dictionary learning method called crosscoders. We find that most features begin to form around a specific point, while more complex patterns emerge in later training stages. Feature attribution analyses reveal causal connections between feature evolution and downstream performance. Our feature-level observations are highly consistent with previous findings on Transformer's two-stage learning process, which we term a statistical learning phase and a feature learning phase. Our work opens up the possibility to track fine-grained representation progress during language model learning dynamics.

Related papers

• Towards Understanding Multimodal Fine-Tuning: Spatial Features [25.349396112139214]
  Vision-Language Models (VLMs) achieve strong performance on a wide range of tasks by pairing a vision encoder with a pre-trained language model.<n>We present the first mechanistic analysis of VLM adaptation using stage-wise model diffing.
  arXiv  Detail & Related papers  (2026-02-06T18:48:18Z)
• Crosscoding Through Time: Tracking Emergence & Consolidation Of Linguistic Representations Throughout LLM Pretraining [33.22703101533052]
  Large language models learn non-trivial abstractions during pretraining.<n>We use sparse crosscoders to discover and align features across model checkpoints.<n>We show that crosscoders can detect feature emergence, maintenance, and discontinuation during pretraining.
  arXiv  Detail & Related papers  (2025-09-05T17:56:24Z)
• Dwell in the Beginning: How Language Models Embed Long Documents for Dense Retrieval [31.9252824152673]
  We build on previous research that demonstrated loss of information in the middle of input sequences for causal language models. We examine positional biases at various stages of training for an encoder-decoder model, including language model pre-training, contrastive pre-training, and contrastive fine-tuning.
  arXiv  Detail & Related papers  (2024-04-05T15:16:16Z)
• Premonition: Using Generative Models to Preempt Future Data Changes in Continual Learning [63.850451635362425]
  Continual learning requires a model to adapt to ongoing changes in the data distribution. We show that the combination of a large language model and an image generation model can similarly provide useful premonitions. We find that the backbone of our pre-trained networks can learn representations useful for the downstream continual learning problem.
  arXiv  Detail & Related papers  (2024-03-12T06:29:54Z)
• Unveiling Multilinguality in Transformer Models: Exploring Language Specificity in Feed-Forward Networks [12.7259425362286]
  We investigate how multilingual models might leverage key-value memories. For autoregressive models trained on two or more languages, do all neurons (across layers) respond equally to all languages? Our findings reveal that the layers closest to the network's input or output tend to exhibit more language-specific behaviour compared to the layers in the middle.
  arXiv  Detail & Related papers  (2023-10-24T06:45:00Z)
• Expedited Training of Visual Conditioned Language Generation via Redundancy Reduction [61.16125290912494]
  $textEVL_textGen$ is a framework designed for the pre-training of visually conditioned language generation models. We show that our approach accelerates the training of vision-language models by a factor of 5 without a noticeable impact on overall performance.
  arXiv  Detail & Related papers  (2023-10-05T03:40:06Z)
• Few-shot Subgoal Planning with Language Models [58.11102061150875]
  We show that language priors encoded in pre-trained language models allow us to infer fine-grained subgoal sequences. In contrast to recent methods which make strong assumptions about subgoal supervision, our experiments show that language models can infer detailed subgoal sequences without any fine-tuning.
  arXiv  Detail & Related papers  (2022-05-28T01:03:30Z)
• Exploring Fine-tuning Techniques for Pre-trained Cross-lingual Models via Continual Learning [74.25168207651376]
  Fine-tuning pre-trained language models to downstream cross-lingual tasks has shown promising results. We leverage continual learning to preserve the cross-lingual ability of the pre-trained model when we fine-tune it to downstream tasks. Our methods achieve better performance than other fine-tuning baselines on the zero-shot cross-lingual part-of-speech tagging and named entity recognition tasks.
  arXiv  Detail & Related papers  (2020-04-29T14:07:18Z)
• Pre-training Text Representations as Meta Learning [113.3361289756749]
  We introduce a learning algorithm which directly optimize model's ability to learn text representations for effective learning of downstream tasks. We show that there is an intrinsic connection between multi-task pre-training and model-agnostic meta-learning with a sequence of meta-train steps.
  arXiv  Detail & Related papers  (2020-04-12T09:05:47Z)

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.