Towards Active Synthetic Data Generation for Finetuning Language Models

• URL: http://arxiv.org/abs/2512.00884v1
• Date: Sun, 30 Nov 2025 13:13:00 GMT
• Title: Towards Active Synthetic Data Generation for Finetuning Language Models
• Authors: Samuel Kessler, Menglin Xia, Daniel Madrigal Diaz, Dongge Han, Helia Heshemi, Saravan Rajmohan, Victor Ruehle, Jordan T. Ash,
• Abstract summary: This paper studies and advocates for the latter case, where data are generated in an iterative, closed-loop fashion.<n>For a fixed budget of generated samples, or a budget in terms of compute spent querying a teacher, we show that this curation of finetuning data affords improved student performance.
• Score: 22.595504847732823
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: A common and effective means for improving language model capabilities involves finetuning a ``student'' language model's parameters on generations from a more proficient ``teacher'' model. Termed ``synthetic data'', these generations are often produced before any student finetuning, but some work has considered generating new synthetic samples as training progresses. This paper studies and advocates for the latter case, where data are generated in an iterative, closed-loop fashion that is guided by the current state of the student model. For a fixed budget of generated samples, or a budget in terms of compute spent querying a teacher, we show that this curation of finetuning data affords improved student performance over static generation. Further, while there have been several LLM-specific methods proposed that operate in this regime, we find that simple, inexpensive selection criteria from the active learning literature tend to be most performant. We validate these claims across four mathematical and logical reasoning datasets using four different small language models.

Related papers

• Self-Improving LLM Agents at Test-Time [49.9396634315896]
  One paradigm of language model (LM) fine-tuning relies on creating large training datasets.<n>In practice, gathering large sets of data is inefficient, and training on them is prohibitively expensive.<n>We study two variants of this approach: Test-Time Self-Improvement (TT-SI) and Test-Time Distillation (TT-D)
  arXiv  Detail & Related papers  (2025-10-09T06:37:35Z)
• Assessing Generative Models for Structured Data [0.0]
  This paper introduces rigorous methods for assessing synthetic data against real data by looking at inter-column dependencies within the data.<n>We find that large language models (GPT-2), both when queried via few-shot prompting, and when fine-tuned, and GAN (CTGAN) models do not produce data with dependencies that mirror the original real data.
  arXiv  Detail & Related papers  (2025-03-26T18:19:05Z)
• How Hard is this Test Set? NLI Characterization by Exploiting Training Dynamics [49.9329723199239]
  We propose a method for the automated creation of a challenging test set without relying on the manual construction of artificial and unrealistic examples. We categorize the test set of popular NLI datasets into three difficulty levels by leveraging methods that exploit training dynamics. When our characterization method is applied to the training set, models trained with only a fraction of the data achieve comparable performance to those trained on the full dataset.
  arXiv  Detail & Related papers  (2024-10-04T13:39:21Z)
• Generative Pre-training for Speech with Flow Matching [81.59952572752248]
  We pre-trained a generative model, named SpeechFlow, on 60k hours of untranscribed speech with Flow Matching and masked conditions. Experiment results show the pre-trained generative model can be fine-tuned with task-specific data to match or surpass existing expert models on speech enhancement, separation, and synthesis.
  arXiv  Detail & Related papers  (2023-10-25T03:40:50Z)
• TrueTeacher: Learning Factual Consistency Evaluation with Large Language Models [20.09470051458651]
  We introduce TrueTeacher, a method for generating synthetic data by annotating diverse model-generated summaries. Unlike prior work, TrueTeacher does not rely on human-written summaries, and is multilingual by nature.
  arXiv  Detail & Related papers  (2023-05-18T17:58:35Z)
• A Multi-dimensional Evaluation of Tokenizer-free Multilingual Pretrained Models [87.7086269902562]
  We show that subword-based models might still be the most practical choice in many settings. We encourage future work in tokenizer-free methods to consider these factors when designing and evaluating new models.
  arXiv  Detail & Related papers  (2022-10-13T15:47:09Z)
• An Application of Pseudo-Log-Likelihoods to Natural Language Scoring [5.382454613390483]
  A language model with relatively few parameters and training steps can outperform it on a recent large data set. We produce some absolute state-of-the-art results for common sense reasoning in binary choice tasks. We argue that robustness of the smaller model ought to be understood in terms of compositionality.
  arXiv  Detail & Related papers  (2022-01-23T22:00:54Z)
• Turning Tables: Generating Examples from Semi-structured Tables for Endowing Language Models with Reasoning Skills [32.55545292360155]
  We propose to leverage semi-structured tables, and automatically generate at scale question-paragraph pairs. We add a pre-training step over this synthetic data, which includes examples that require 16 different reasoning skills. We show that our model, PReasM, substantially outperforms T5, a popular pre-trained encoder-decoder model.
  arXiv  Detail & Related papers  (2021-07-15T11:37:14Z)
• Fine-tuning BERT for Low-Resource Natural Language Understanding via Active Learning [30.5853328612593]
  In this work, we explore fine-tuning methods of BERT -- a pre-trained Transformer based language model. Our experimental results show an advantage in model performance by maximizing the approximate knowledge gain of the model. We analyze the benefits of freezing layers of the language model during fine-tuning to reduce the number of trainable parameters.
  arXiv  Detail & Related papers  (2020-12-04T08:34:39Z)
• Selecting Informative Contexts Improves Language Model Finetuning [66.26521454263343]
  We present a general fine-tuning method that we call information gain filtration. During fine-tuning, a secondary learner selects informative examples and skips uninformative ones. We show that our method has consistent improvement across datasets, fine-tuning tasks, and language model architectures.
  arXiv  Detail & Related papers  (2020-05-01T02:01:18Z)
• Exploring Versatile Generative Language Model Via Parameter-Efficient Transfer Learning [70.81910984985683]
  We propose an effective way to fine-tune multiple down-stream generation tasks simultaneously using a single, large pre-trained model. The experiments on five diverse language generation tasks show that by just using an additional 2-3% parameters for each task, our model can maintain or even improve the performance of fine-tuning the whole model.
  arXiv  Detail & Related papers  (2020-04-08T06:18:44Z)

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.