Tag
This paper introduces LinTree, which improves LLM reasoning by adding explicit parent pointers to linearized search histories, showing that making the tree structure explicit boosts both task performance and search efficiency compared to implicit reasoning and heuristic-guided search.
ETCHR is a novel image editing approach that decouples visual reasoning from image generation, using a two-stage training process (Reasoning Imitation and Reasoning Enhancement) to improve multimodal language model performance across five visual reasoning tasks. It achieves consistent gains of 4-5% Pass@1 on models like Qwen3-VL-8B, Gemini-3.1-Flash-Lite, and Kimi K2.5.
OptiLLM is an open-source inference proxy that boosts LLM reasoning accuracy by up to 10x using advanced techniques without requiring retraining, compatible with various AI APIs.