mlsanityData to 14 Oct 2026

Know When To Fold 'Em: Token-Efficient LLM Synthetic Data Generation via Multi-Stage In-Flight Rejection

arXiv 2605.14062 ↗First posted 13 May 2026cs.AI cs.CLcited by 0

On the map: Synthetic data and self-improvement for LLMs → Code and instruction data synthesis

Abstract (arXiv)

While synthetic data generation with large language models (LLMs) is widely used in post-training pipelines, existing approaches typically generate full outputs before applying quality filters, leading to substantial token waste on samples that are ultimately discarded. To address this, we propose Multi-Stage In-Flight Rejection (MSIFR), a lightweight, training-free framework that detects and terminates low-quality generation trajectories at intermediate checkpoints before they reach full completion. MSIFR decomposes the generation process into sequential stages and applies fast rule-based validators to identify arithmetic inconsistencies, hallucination patterns, and formatting violations, enabling early rejection of faulty samples. We formalize in-flight rejection as a sequential decision process and show that any non-trivial discard policy reduces expected token consumption, with stage-wise savings increasing when rejection occurs earlier in the generation pipeline. We further demonstrate that conditional utility estimates form a martingale, ensuring that early, in-flight rejection does not bias the expected utility of retained samples. Across five instruction-tuned models and seven reasoning benchmarks, MSIFR reduces token consumption by 11%-77% as a standalone method, and up to 78.2% when combined with early-exit methods, while preserving or improving evaluation accuracy. These results confirm that MSIFR provides a practical mechanism for improving the efficiency of LLM-based synthetic data generation without additional training or architectural changes.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Code and instruction data synthesis, one of 13 directions in Synthetic data and self-improvement for LLMs. Builds on 33 of its 46 references that are in the map. First posted on arXiv 13 May 2026.

Builds on

33 of 46 references.

References · 33

  1. 2020Language Models are Few-Shot Learnerscited by 32,419
  2. 2023GPT-4 Technical Reportcited by 18,396
  3. 2024The Llama 3 Herd of Modelscited by 14,715
  4. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
  5. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  6. 2020Scaling Laws for Neural Language Modelscited by 6,586
  7. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,944
  8. 2024Qwen2.5 Technical Reportcited by 4,942
  9. 2023Qwen Technical Reportcited by 3,344
  10. 2023Mistral 7Bcited by 3,164
  11. 2019Fine-Tuning Language Models from Human Preferencescited by 2,150
  12. 2024Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phonecited by 1,917
21 more
  1. 2021A Gentle Introduction to Conformal Prediction and Distribution-Free Uncertainty Quantificationcited by 792
  2. 2019MathQA: Towards Interpretable Math Word Problem Solving with Operation-Based Formalismscited by 662
  3. 2024DeepSeek LLM: Scaling Open-Source Language Models with Longtermismcited by 642
  4. 2019Analysing Mathematical Reasoning Abilities of Neural Modelscited by 335
  5. 2025Dynamic Early Exit in Reasoning Modelscited by 219
  6. 2025Deep Think with Confidencecited by 205
  7. 2016Sequence Tutor: Conservative Fine-Tuning of Sequence Generation Models with KL-controlcited by 101
  8. 2025S-GRPO: Early Exit via Reinforcement Learning in Reasoning Modelscited by 87
  9. 2025Synthetic Data Generation Using Large Language Models: Advances in Text and Codecited by 61
  10. 2025Think or Not? Exploring Thinking Efficiency in Large Reasoning Models via an Information-Theoretic Lenscited by 34
  11. 2025FlashThink: An Early Exit Method For Efficient Reasoningcited by 28
  12. 2024Synthesizing Post-Training Data for LLMs through Multi-Agent Simulationcited by 24
  13. 2024Prune 'n Predict: Optimizing LLM Decision-making with Conformal Predictioncited by 18
  14. 2025Lost at the Beginning of Reasoningcited by 15
  15. 2026Abstain-R1: Calibrated Abstention and Post-Refusal Clarification via Verifiable RLcited by 14
  16. 2024SynthesizRR: Generating Diverse Datasets with Retrieval Augmentationcited by 13
  17. 2026Knowing When to Quit: A Principled Framework for Dynamic Abstention in LLM Reasoningcited by 8
  18. 2025LYNX: Learning Dynamic Exits for Confidence-Controlled Reasoningcited by 6
  19. 2025Revisiting Privacy, Utility, and Efficiency Trade-offs when Fine-Tuning Large Language Modelscited by 5
  20. 2025Accelerating LLM Reasoning via Early Rejection with Partial Reward Modelingcited by 4
  21. 2026The Detection-Extraction Gap: Models Know the Answer Before They Can Say Itcited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025BARE: Leveraging Base Language Models for Few-Shot Synthetic Data Generation
  2. 2025Scaling Instruction-Tuned LLMs to Million-Token Contexts via Hierarchical Synthetic Data Generation
  3. 2024SPADE: Synthesizing Data Quality Assertions for Large Language Model Pipelines
  4. 2025FastMTP: Accelerating LLM Inference with Enhanced Multi-Token Prediction
  5. 2025HAMburger: Accelerating LLM Inference via Token Smashing
  6. 2026Demystifying Numerical Instability in LLM Inference: Achieving Reproducible Inference for Mission-Critical Tasks with HEAL
  7. 2026Locally Confident, Globally Stuck: The Quality-Exploration Dilemma in Diffusion Language Models
  8. 2025On Evaluating Performance of LLM Inference Serving Systems