Tag
The Infinity-Parser2 technical report presents a large multimodal model for end-to-end document parsing, featuring a scalable data synthesis pipeline and multi-task reinforcement learning. It achieves state-of-the-art results on multiple benchmarks while releasing open-source model variants and a 5-million-sample bilingual corpus.
This paper introduces Goku, a million-scale dataset and benchmark for instruction-based video editing, supporting multi-task and structural manipulations. The accompanying model, Goku-Edit, achieves up to +8% improvement on instruction following over open-source models.
This paper introduces RODS, a reward-driven online data synthesis method that addresses the depletion of informative samples in static datasets for multi-turn tool-use agent training. It achieves comparable performance to larger offline pipelines with significantly fewer trajectories.
Edu-Theater is a data-efficient agent framework that uses LLM-powered generative agents to simulate learner behavior in educational settings. It employs a cohort-aware roll-call paradigm to infer learner states with fewer data and computational resources, achieving higher simulation accuracy.
Introduces Geometry-Aware Tabular Diffusion (GATD), which augments tabular diffusion denoisers with explicit pairwise geometric features. Achieves state-of-the-art performance on ten benchmarks while using significantly fewer parameters.
DOMINO is a novel framework that learns minimal sufficient domain representations from reference examples to synthesize domain-specific data for LLMs, improving code benchmark performance without requiring explicit domain descriptions.
This paper proposes a knowledge-aware Text-to-SQL framework that uses knowledge distillation to improve performance in low-resource settings by constructing task-specific knowledge bases and generating synthetic training data. Experiments on seven benchmarks show substantial improvements, especially for open-source models.
Terminal-World introduces a fully automated pipeline that uses agent skills to synthesize high-quality training data for terminal agents, enabling models to outperform baselines with only 1.2% of the training data. The method co-derives task instructions, environments, and teacher trajectories from skill primitives.
This paper proposes a pipeline for fine-tuning LLMs specifically for explainable misinformation detection and introduces LonsRex, a data synthesis method to generate necessary and sufficient rationales, addressing limitations of naive filtering based solely on label correctness.
Uni-Edit proposes using intelligent image editing as a single general task to simultaneously improve unified multimodal models' understanding, generation, and editing capabilities, with an automated data synthesis pipeline creating complex editing instructions.
FrontierSmith automatically generates diverse open-ended coding problems from closed-ended tasks, improving LLM coding performance on benchmarks through enhanced agent interactions and training data synthesis.
This paper introduces CUActSpot, a multimodal benchmark for evaluating computer-use agents, and a renderer-based data synthesis pipeline. The proposed Phi-Ground-Any-4B model outperforms open-source models under 32B parameters.
CoEvolve proposes an agent-data mutual evolution framework for training LLM agents through closed-loop, interaction-driven learning that adapts both the agent and its training data distribution. The method extracts feedback signals from rollout trajectories to guide LLM-based task synthesis, demonstrating significant improvements (15-19% absolute gains) across multiple Qwen models on AppWorld and BFCL benchmarks.
A dual-view data synthesis method using polarity reversal boosts instruction-following retrieval performance by 45% on the FollowIR benchmark.
WebShaper is a formalization-driven framework for synthesizing information-seeking datasets using set theory and Knowledge Projections, achieving state-of-the-art performance on GAIA and WebWalkerQA benchmarks among open-source agents.