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

Anjir Ahmed Chowdhury, Syed Zawad, Feng Yan

arXiv:2605.14062 · 2026-05-16 공개 · arXiv · PDF

llm instruction-tuning token-efficiency hallucination-detection reasoning-benchmarks synthetic-data-generation early-exit sequential-decision

Abstract

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.

한국어 요약

📋 한 줄 요약

**[합성 데이터 / LLM 효율]** 저품질 LLM 생성 궤적을 중간 체크포인트에서 조기 거부해 토큰 낭비를 줄이는 학습 불요(training-free) 프레임워크 MSIFR 제안.

🎯 핵심 기여도

💡 핵심 아이디어

"끝까지 생성하고 사후 필터링"이라는 통념을 깨고, 생성 과정 자체를 단계별 검증과 결합한다. 잘못된 궤적을 일찍 끊을수록 토큰 절감 효과는 지수적으로 커지며, martingale 구조 덕분에 통계적 무편향이 보장된다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: 추가 학습·아키텍처 변경 없이 LLM 합성 데이터 생성 비용을 극적으로 줄일 수 있는 실용 메커니즘 제공. **한계**: 규칙 기반 검증기 설계가 도메인 의존적, 매우 창의적이거나 비정형 출력에는 거부율이 비대해질 가능성.

🚀 실용적 활용