Prototype Language Models

• URL: http://arxiv.org/abs/2607.00510v1
• Date: Wed, 01 Jul 2026 06:45:02 GMT
• Title: Prototype Language Models
• Abstract summary: Prototype language models form each prediction via a sparse, non-negative mixture of learned prototypes.<n>We show that sparse prototype structure localizes curvature in the loss landscape, yielding a more tractable Hessian.
• Score: 32.76699413390825
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Knowing which training examples drive outputs is fundamental to auditing, correcting, and understanding language models, yet for modern LLMs this remains expensive, approximate, and largely post-hoc. Standard language models generate tokens through a dense network pathway, causing training data's influence to be distributed across parameters rather than organized along explicit, traceable components. We introduce a prototype language model architecture, Prototypes for Interpretable Sequence Modeling (PRISM), that forms each prediction via a sparse, non-negative mixture of learned prototypes, trained with clustering objectives that anchor each prototype to coherent neighborhoods of training examples. Across architectures from 130M to 1.6B parameters trained on up to 50B tokens, prototype language models either surpass or remain within 2.5 percentage points on average downstream accuracy of matched dense baselines. We show that sparse prototype structure localizes curvature in the loss landscape, yielding a more tractable Hessian and enabling training data attribution that is ~500x faster than post hoc baselines when consuming equivalent memory. Calibrating linear prototype controllers can improve downstream accuracy by roughly 3 points while tracing those corrections back to training neighborhoods, and targeted prototype suppression can remove model behaviors without finetuning or measurable loss in generation quality.

Related papers

• ProtoTTA: Prototype-Guided Test-Time Adaptation [8.178991486617972]
  We introduce ProtoTTA, a framework for prototypical models that leverage intermediate prototype signals rather than relying solely on model outputs.<n>ProtoTTA minimizes the entropy of the prototype-similarity distribution to encourage more confident and prototype-specific activations on shifted data.<n>We also introduce novel interpretability metrics and a vision-language minimization model (VLM) evaluation framework to explain TTA dynamics.
  arXiv  Detail & Related papers  (2026-04-16T19:52:57Z)
• MinerU2.5-Pro: Pushing the Limits of Data-Centric Document Parsing at Scale [92.09717763663873]
  We present MinerU2.5-Pro, which advances the state of the art purely through data engineering and training strategy design.<n>At its core is a Data Engine co-designed around coverage, informativeness, and annotation accuracy.<n>We show that MinerU2.5-Pro achieves 95.69 on OmniDocBench v1.6, improving over the same-architecture baseline by 2.71 points and surpassing all existing methods.
  arXiv  Detail & Related papers  (2026-04-06T15:44:18Z)
• 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)
• Probabilistic Prototype Calibration of Vision-Language Models for Generalized Few-shot Semantic Segmentation [75.18058114915327]
  Generalized Few-Shot Semanticnative (GFSS) aims to extend a segmentation model to novel classes with only a few annotated examples.<n>We propose FewCLIP, a probabilistic prototype calibration framework over multi-modal prototypes from the pretrained CLIP.<n>We show FewCLIP significantly outperforms state-of-the-art approaches across both GFSS and class-incremental setting.
  arXiv  Detail & Related papers  (2025-06-28T18:36:22Z)
• What Do Learning Dynamics Reveal About Generalization in LLM Reasoning? [83.83230167222852]
  We find that a model's generalization behavior can be effectively characterized by a training metric we call pre-memorization train accuracy. By connecting a model's learning behavior to its generalization, pre-memorization train accuracy can guide targeted improvements to training strategies.
  arXiv  Detail & Related papers  (2024-11-12T09:52:40Z)
• This Looks Better than That: Better Interpretable Models with ProtoPNeXt [14.28283868577614]
  Prototypical-part models are a popular interpretable alternative to black-box deep learning models for computer vision. We create a new framework for integrating components of prototypical-part models -- ProtoPNeXt.
  arXiv  Detail & Related papers  (2024-06-20T18:54:27Z)
• Rethinking Few-shot 3D Point Cloud Semantic Segmentation [62.80639841429669]
  This paper revisits few-shot 3D point cloud semantic segmentation (FS-PCS) We focus on two significant issues in the state-of-the-art: foreground leakage and sparse point distribution. To address these issues, we introduce a standardized FS-PCS setting, upon which a new benchmark is built.
  arXiv  Detail & Related papers  (2024-03-01T15:14:47Z)
• 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)
• Same Pre-training Loss, Better Downstream: Implicit Bias Matters for Language Models [46.24479693469042]
  This paper shows that 1) pre-training loss cannot fully explain downstream performance and 2) flatness of the model is well-correlated with downstream performance where pre-training loss is not.
  arXiv  Detail & Related papers  (2022-10-25T17:45:36Z)
• Meta-learning Pathologies from Radiology Reports using Variance Aware Prototypical Networks [3.464871689508835]
  We propose a simple extension of the Prototypical Networks for few-shot text classification. Our main idea is to replace the class prototypes by Gaussians and introduce a regularization term that encourages the examples to be clustered near the appropriate class centroids.
  arXiv  Detail & Related papers  (2022-10-22T05:22:29Z)
• Scale Efficiently: Insights from Pre-training and Fine-tuning Transformers [57.931830650323]
  This paper presents scaling insights from pretraining and finetuning Transformers. We show that aside from only the model size, model shape matters for downstream fine-tuning. We present improved scaling protocols whereby our redesigned models achieve similar downstream fine-tuning quality.
  arXiv  Detail & Related papers  (2021-09-22T12:29:15Z)
• Unsupervised Paraphrasing with Pretrained Language Models [85.03373221588707]
  We propose a training pipeline that enables pre-trained language models to generate high-quality paraphrases in an unsupervised setting. Our recipe consists of task-adaptation, self-supervision, and a novel decoding algorithm named Dynamic Blocking. We show with automatic and human evaluations that our approach achieves state-of-the-art performance on both the Quora Question Pair and the ParaNMT datasets.
  arXiv  Detail & Related papers  (2020-10-24T11:55:28Z)

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.