Synthetic Mixed Training: Scaling Parametric Knowledge Acquisition Beyond RAG

• URL: http://arxiv.org/abs/2603.23562v2
• Date: Mon, 30 Mar 2026 08:42:02 GMT
• Title: Synthetic Mixed Training: Scaling Parametric Knowledge Acquisition Beyond RAG
• Abstract summary: We introduce Synthetic Mixed Training, which combines synthetic QAs and synthetic documents.<n>This enables log-linear improvements as both synthetic data volume and generator strength increase.<n>Across models and benchmarks, our training enables models to beat RAG in five of six settings, outperforms by 2.6%, and achieves a 9.1% gain when combined with RAG.
• Score: 56.95387658211215
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Synthetic data augmentation helps language models learn new knowledge in data-constrained domains. However, naively scaling existing synthetic data methods by training on more synthetic tokens or using stronger generators yields diminishing returns below the performance of RAG. To break the RAG ceiling, we introduce Synthetic Mixed Training, which combines synthetic QAs and synthetic documents. This leverages their complementary training signals, and enables log-linear improvements as both synthetic data volume and generator strength increase. This allows the model to outperform RAG by a 2.6% relative gain on QuaLITY, a long-document reading comprehension benchmark. In addition, we introduce Focal Rewriting, a simple technique for synthetic document generation that explicitly conditions document generation on specific questions, improving the diversity of synthetic documents and yielding a steeper log-linear scaling curve. On QuaLITY, our final recipe trains a Llama 8B model that outperforms RAG by 4.4% relatively. Across models and benchmarks (QuaLITY, LongHealth, FinanceBench), our training enables models to beat RAG in five of six settings, outperforms by 2.6%, and achieves a 9.1% gain when combined with RAG.

Related papers

• From Memorization to Creativity: LLM as a Designer of Novel Neural-Architectures [48.83701310501069]
  Large language models (LLMs) excel in program synthesis, yet their ability to autonomously navigate neural architecture design-balancing reliability, performance, and structural novelty--remains underexplored.<n>We address this by placing a code-oriented LLM within a closed-loop synthesis framework, analyzing its evolution over 22 supervised fine-tuning cycles.
  arXiv  Detail & Related papers  (2026-01-06T13:20:28Z)
• Characterizing Model Behavior Under Synthetic Data Training: An Empirical Study Across Scales and Mixing Ratios [1.631115063641726]
  This paper presents a controlled empirical study examining model performance, calibration, and output characteristics when trained on varying synthetic-to-external data ratios.<n>Our key findings include: models maintain stable performance with up to 20% synthetic data, but degradation accelerates beyond 30%.<n>Current best practices, such as those employed in STaR and Self-Instruct systems that maintain greater than 80% external data, operate well within safe regimes identified by our experiments.
  arXiv  Detail & Related papers  (2025-10-01T03:28:01Z)
• Synthline: A Product Line Approach for Synthetic Requirements Engineering Data Generation using Large Language Models [0.5156484100374059]
  This paper introduces Synthline, a Product Line (PL) approach that leverages Large Language Models to generate synthetic Requirements Engineering (RE) data.<n>Our analysis reveals that while synthetic datasets exhibit less diversity than real data, they are good enough to serve as viable training resources.<n>Our evaluation shows that combining synthetic and real data leads to substantial performance improvements.
  arXiv  Detail & Related papers  (2025-05-06T07:57:16Z)
• Scaling Laws of Synthetic Data for Language Models [125.41600201811417]
  We introduce SynthLLM, a scalable framework that transforms pre-training corpora into diverse, high-quality synthetic datasets.<n>Our approach achieves this by automatically extracting and recombining high-level concepts across multiple documents using a graph algorithm.
  arXiv  Detail & Related papers  (2025-03-25T11:07:12Z)
• AugGen: Synthetic Augmentation using Diffusion Models Can Improve Recognition [14.525986333650417]
  Synthetic data generation offers a promising alternative to external datasets or pre-trained models.<n>AugGen strategically samples from a class-conditional generative model trained exclusively on the target FR dataset.<n>Our findings demonstrate that carefully integrated synthetic data can both mitigate privacy constraints and substantially enhance recognition performance.
  arXiv  Detail & Related papers  (2025-03-14T16:10:21Z)
• Little Giants: Synthesizing High-Quality Embedding Data at Scale [71.352883755806]
  We introduce SPEED, a framework that aligns open-source small models to efficiently generate large-scale embedding data. SPEED uses only less than 1/10 of the GPT API calls, outperforming the state-of-the-art embedding model E5_mistral when both are trained solely on their synthetic data.
  arXiv  Detail & Related papers  (2024-10-24T10:47:30Z)
• TarGEN: Targeted Data Generation with Large Language Models [51.87504111286201]
  TarGEN is a multi-step prompting strategy for generating high-quality synthetic datasets. We augment TarGEN with a method known as self-correction empowering LLMs to rectify inaccurately labeled instances. A comprehensive analysis of the synthetic dataset compared to the original dataset reveals similar or higher levels of dataset complexity and diversity.
  arXiv  Detail & Related papers  (2023-10-27T03:32:17Z)
• Towards a Competitive End-to-End Speech Recognition for CHiME-6 Dinner Party Transcription [73.66530509749305]
  In this paper, we argue that, even in difficult cases, some end-to-end approaches show performance close to the hybrid baseline. We experimentally compare and analyze CTC-Attention versus RNN-Transducer approaches along with RNN versus Transformer architectures. Our best end-to-end model based on RNN-Transducer, together with improved beam search, reaches quality by only 3.8% WER abs. worse than the LF-MMI TDNN-F CHiME-6 Challenge baseline.
  arXiv  Detail & Related papers  (2020-04-22T19:08:33Z)

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.