mlsanityData to 14 Oct 2026

Approximating Language Model Training Data from Weights

arXiv 2506.15553 ↗First posted 18 Jun 2025cs.CLcited by 4

Not on the map. Why?

Abstract (arXiv)

Modern language models often have open weights but closed training data. We formalize the problem of data approximation from model weights and propose several baselines and metrics. We develop a gradient-based approach that selects the highest-matching data from a large public text corpus and show its effectiveness at recovering useful data given only weights of the original and finetuned models. Even when none of the true training data is known, our method is able to locate a small subset of public Web documents can be used to train a model to close to the original model performance given models trained for both classification and supervised-finetuning. On the AG News classification task, our method improves performance from 65% (using randomly selected data) to 80%, approaching the expert benchmark of 88%. When applied to a model trained with SFT on MSMARCO web documents, our method reduces perplexity from 3.3 to 2.3, compared to an expert LLAMA model's perplexity of 2.0.

Key facts. Cited by 4 papers in the map (data through 14 Oct 2026). Builds on 31 of its 48 references that are in the map. First posted on arXiv 18 Jun 2025.

Builds on

31 of 48 references.

References · 31

  1. 2015Character-level Convolutional Networks for Text Classificationcited by 2,330
  2. 2013Sinkhorn Distances: Lightspeed Computation of Optimal Transportation Distancescited by 2,316
  3. 2020Extracting Training Data from Large Language Modelscited by 2,189
  4. 2016MS MARCO: A Human Generated MAchine Reading COmprehension Datasetcited by 1,605
  5. 2022Quantifying Memorization Across Neural Language Modelscited by 763
  6. 2020Estimating Training Data Influence by Tracing Gradient Descentcited by 580
  7. 2024LESS: Selecting Influential Data for Targeted Instruction Tuningcited by 532
  8. 2020Dataset Condensation with Gradient Matchingcited by 516
  9. 2023Scalable Extraction of Training Data from (Production) Language Modelscited by 434
  10. 2021Large Language Models Can Be Strong Differentially Private Learnerscited by 412
  11. 2023Data Selection for Language Models via Importance Resamplingcited by 329
  12. 2021GRAD-MATCH: Gradient Matching based Data Subset Selection for Efficient Deep Model Trainingcited by 274
19 more
  1. 2020GLISTER: Generalization based Data Subset Selection for Efficient and Robust Learningcited by 226
  2. 2020Coresets via Bilevel Optimization for Continual Learning and Streamingcited by 192
  3. 2022Scaling Up Dataset Distillation to ImageNet-1K with Constant Memorycited by 181
  4. 2022Dataset Distillation using Neural Feature Regressioncited by 177
  5. 2024Stealing Part of a Production Language Modelcited by 159
  6. 2022Reconstructing Training Data with Informed Adversariescited by 148
  7. 2023Squeeze, Recover and Relabel: Dataset Condensation at ImageNet Scale From A New Perspectivecited by 138
  8. 2022Reconstructing Training Data from Trained Neural Networkscited by 133
  9. 2023Towards Lossless Dataset Distillation via Difficulty-Aligned Trajectory Matchingcited by 118
  10. 2024DsDm: Model-Aware Dataset Selection with Datamodelscited by 105
  11. 2024Logits of API-Protected LLMs Leak Proprietary Informationcited by 40
  12. 2023Making Large Language Models Better Data Creatorscited by 28
  13. 2023Deconstructing Data Reconstruction: Multiclass, Weight Decay and General Lossescited by 26
  14. 2024DiLM: Distilling Dataset into Language Model for Text-level Dataset Distillationcited by 20
  15. 2022Automatic Document Selection for Efficient Encoder Pretrainingcited by 11
  16. 2025Synthetic Text Generation for Training Large Language Models via Gradient Matchingcited by 10
  17. 2025Efficient Model Development through Fine-tuning Transfercited by 6
  18. 2024Training Data Reconstruction: Privacy due to Uncertainty?cited by 4
  19. 2025Can We Infer Confidential Properties of Training Data from LLMs?cited by 4

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 2; 2026: 2.

4 papers.

  1. 2025Extracting alignment data in open modelscited by 8
  2. 2025Beyond Data Privacy: New Privacy Risks for Large Language Modelscited by 6
  3. 2026WARP: Weight-Space Analysis for Recovering Training Data Portfolioscited by 1
  4. 2026Learning to Detect Language Model Training Data via Active Reconstructioncited by 0