Research

A novel policy optimization framework for multimodal large language models that addresses semantic alignment challenges in GUI grounding.

A novel policy optimization framework for multimodal large language models that addresses semantic alignment challenges in GUI grounding.

An instructed version of the InfiR small language model, part of the reasoning-enhanced low-resource training pipeline.

Part of the InfiR reasoning-enhanced low-resource training pipeline, crafted to be an effective small language model with improved reasoning.

A multimodal generalist GUI agent with native hierarchical and expectation-reflection reasoning through a unique two-stage supervised pipeline.

A GUI agent developed via the Actor2Reasoner framework, evolving a reactive model into a deliberative reasoner through spatial reasoning distillation and reinforcement learning.

A structure-aware extension that builds co-activation graphs from logits and aligns them via an efficient Gromov-Wasserstein loss approximation.

A logit-level fusion pipeline based on Universal Logit Distillation, enhanced with Top-K filtering and logits standardization. Supports both pairwise and unified fusion strategies to balance performance and efficiency.

A lightweight fusion method during the preference alignment phase that injects fused model behavior into preference learning.

A multimodal model developed via the InfiMMR three-phase curriculum framework, enhancing multimodal reasoning capabilities in small language models.