Tag
GraphForge introduces an evidence-graph framework that grounds both agent task generation and verification in real files, enabling synthesis of verifiable training trajectories for working agents. Fine-tuning Qwen3.6-27B on 2,169 GraphForge trajectories improves GDPVal, Workspace-Bench-Lite, and SpreadsheetBench II, with rejection fine-tuning yielding further gains.
Perplexity combines rejection sampling fine-tuning with hint-guided self-distillation to train its Computer model on real-world tool use, learning from successful sessions and user-corrected failures.
This paper proposes RIMS, a three-stage preference optimization framework for small-scale language models in retrieval-augmented generation, using synthetic chain-of-thought data and a differentiable soft aggregation mechanism to improve robustness against noisy evidence. Experiments show consistent gains over baselines on multi-hop QA benchmarks.
This paper isolates Fixed-Source Synthesis (FSS) from Source Expansion in synthetic data scaling, proposing a rectified scaling law that predicts performance at high budgets from low-budget fits. Empirical results show FSS is bounded and that adding seed questions outperforms increasing response budgets at large scales.
This paper systematically evaluates data referencing errors (DREs) in LLMs processing tables and proposes critic-based filtering and rejection sampling to improve answer accuracy. A lightweight 4B-parameter critic model achieves high detection accuracy for DREs.
The article proposes Implicit Variational Rejection Sampling (IVRS), which integrates implicit distributions with rejection sampling to improve posterior approximation in variational inference, and introduces the Implicit Resampling Evidence Lower Bound (IR-ELBO) as a tighter variational lower bound.
This paper proposes Multi-Stage In-Flight Rejection (MSIFR), a training-free framework that reduces token waste in LLM-based synthetic data generation by detecting and terminating low-quality generation trajectories at intermediate checkpoints. Across five models and seven 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 accuracy.
This paper introduces Entrocraft, a rejection-sampling method for RL that controls entropy schedules to prevent performance saturation in LLMs. It demonstrates improved generalization and training longevity, allowing smaller models to outperform larger baselines.