Infi-MMR: Curriculum-based Unlocking Multimodal Reasoning via Phased Reinforcement Learning in Multimodal Small Language Models

Recent advancements in large language models (LLMs) have demonstrated substantial progress in reasoning capabilities, such as DeepSeek-R1, which leverages rule-based reinforcement learning to enhance logical reasoning significantly. However, extending these achievements to multimodal large language models (MLLMs) presents critical challenges, which are frequently more pronounced for Multimodal Small Language Models (MSLMs) given their typically weaker foundational reasoning abilities: (1) the scarcity of high-quality multimodal reasoning datasets, (2) the degradation of reasoning capabilities due to the integration of visual processing, and (3) the risk that direct application of reinforcement learning may produce complex yet incorrect reasoning processes. To address these challenges, we design a novel framework Infi-MMR to systematically unlock the reasoning potential of MSLMs through a curriculum of three carefully structured phases and propose our multimodal reasoning model Infi-MMR-3B. The first phase, Foundational Reasoning Activation, leverages high-quality textual reasoning datasets to activate and strengthen the model's logical reasoning capabilities. The second phase, Cross-Modal Reasoning Adaptation, utilizes caption-augmented multimodal data to facilitate the progressive transfer of reasoning skills to multimodal contexts. The third phase, Multimodal Reasoning Enhancement, employs curated, caption-free multimodal data to mitigate linguistic biases and promote robust cross-modal reasoning. Infi-MMR-3B achieves both state-of-the-art multimodal math reasoning ability (43.68% on MathVerse testmini, 27.04% on MathVision test, and 21.33% on OlympiadBench) and general reasoning ability (67.2% on MathVista testmini).

Embracing Imperfection: Simulating Students with Diverse Cognitive Levels Using LLM-based Agents

Large language models (LLMs) are revolutionizing education, with LLM-based agents playing a key role in simulating student behavior. A major challenge in student simulation is modeling the diverse learning patterns of students at various cognitive levels. However, current LLMs, typically trained as ``helpful assistants'', target at generating perfect responses. As a result, they struggle to simulate students with diverse cognitive abilities, as they often produce overly advanced answers, missing the natural imperfections that characterize student learning and resulting in unrealistic simulations. To address this issue, we propose a training-free framework for student simulation. We begin by constructing a cognitive prototype for each student using a knowledge graph, which captures their understanding of concepts from past learning records. This prototype is then mapped to new tasks to predict student performance. Next, we simulate student solutions based on these predictions and iteratively refine them using a beam search method to better replicate realistic mistakes. To validate our approach, we construct the \texttt{Student\_100} dataset, consisting of $100$ students working on Python programming and $5,000$ learning records. Experimental results show that our method consistently outperforms baseline models, achieving $100\%$ improvement in simulation accuracy.

Multimodal LLM-Guided Semantic Correction in Text-to-Image Diffusion

Diffusion models have become the mainstream architecture for text-to-image generation, achieving remarkable progress in visual quality and prompt controllability. However, current inference pipelines generally lack interpretable semantic supervision and correction mechanisms throughout the denoising process. Most existing approaches rely solely on post-hoc scoring of the final image, prompt filtering, or heuristic resampling strategies-making them ineffective in providing actionable guidance for correcting the generative trajectory. As a result, models often suffer from object confusion, spatial errors, inaccurate counts, and missing semantic elements, severely compromising prompt-image alignment and image quality. To tackle these challenges, we propose MLLM Semantic-Corrected Ping-Pong-Ahead Diffusion (PPAD), a novel framework that, for the first time, introduces a Multimodal Large Language Model (MLLM) as a semantic observer during inference. PPAD performs real-time analysis on intermediate generations, identifies latent semantic inconsistencies, and translates feedback into controllable signals that actively guide the remaining denoising steps. The framework supports both inference-only and training-enhanced settings, and performs semantic correction at only extremely few diffusion steps, offering strong generality and scalability. Extensive experiments demonstrate PPAD's significant improvements.

AppealCase: A Dataset and Benchmark for Civil Case Appeal Scenarios

Recent advances in LegalAI have primarily focused on individual case judgment analysis, often overlooking the critical appellate process within the judicial system. Appeals serve as a core mechanism for error correction and ensuring fair trials, making them highly significant both in practice and in research. To address this gap, we present the AppealCase dataset, consisting of 10,000 pairs of real-world, matched first-instance and second-instance documents across 91 categories of civil cases. The dataset also includes detailed annotations along five dimensions central to appellate review: judgment reversals, reversal reasons, cited legal provisions, claim-level decisions, and whether there is new information in the second instance. Based on these annotations, we propose five novel LegalAI tasks and conduct a comprehensive evaluation across 20 mainstream models. Experimental results reveal that all current models achieve less than 50% F1 scores on the judgment reversal prediction task, highlighting the complexity and challenge of the appeal scenario. We hope that the AppealCase dataset will spur further research in LegalAI for appellate case analysis and contribute to improving consistency in judicial decision-making.

ThinkRec: Thinking-based recommendation via LLM

Recent advances in large language models (LLMs) have enabled more semantic-aware recommendations through natural language generation. Existing LLM for recommendation (LLM4Rec) methods mostly operate in a System 1-like manner, relying on superficial features to match similar items based on click history, rather than reasoning through deeper behavioral logic. This often leads to superficial and erroneous recommendations. Motivated by this, we propose ThinkRec, a thinking-based framework that shifts LLM4Rec from System 1 to System 2 (rational system). Technically, ThinkRec introduces a thinking activation mechanism that augments item metadata with keyword summarization and injects synthetic reasoning traces, guiding the model to form interpretable reasoning chains that consist of analyzing interaction histories, identifying user preferences, and making decisions based on target items. On top of this, we propose an instance-wise expert fusion mechanism to reduce the reasoning difficulty. By dynamically assigning weights to expert models based on users' latent features, ThinkRec adapts its reasoning path to individual users, thereby enhancing precision and personalization. Extensive experiments on real-world datasets demonstrate that ThinkRec significantly improves the accuracy and interpretability of recommendations. Our implementations are available in anonymous Github: https://anonymous.4open.science/r/ThinkRec_LLM.

InfiFPO: Implicit Model Fusion via Preference Optimization in Large Language Models

Model fusion combines multiple Large Language Models (LLMs) with different strengths into a more powerful, integrated model through lightweight training methods. Existing works on model fusion focus primarily on supervised fine-tuning (SFT), leaving preference alignment (PA) --a critical phase for enhancing LLM performance--largely unexplored. The current few fusion methods on PA phase, like WRPO, simplify the process by utilizing only response outputs from source models while discarding their probability information. To address this limitation, we propose InfiFPO, a preference optimization method for implicit model fusion. InfiFPO replaces the reference model in Direct Preference Optimization (DPO) with a fused source model that synthesizes multi-source probabilities at the sequence level, circumventing complex vocabulary alignment challenges in previous works and meanwhile maintaining the probability information. By introducing probability clipping and max-margin fusion strategies, InfiFPO enables the pivot model to align with human preferences while effectively distilling knowledge from source models. Comprehensive experiments on 11 widely-used benchmarks demonstrate that InfiFPO consistently outperforms existing model fusion and preference optimization methods. When using Phi-4 as the pivot model, InfiFPO improve its average performance from 79.95 to 83.33 on 11 benchmarks, significantly improving its capabilities in mathematics, coding, and reasoning tasks.

InfiGFusion: Graph-on-Logits Distillation via Efficient Gromov-Wasserstein for Model Fusion

Recent advances in large language models (LLMs) have intensified efforts to fuse heterogeneous open-source models into a unified system that inherits their complementary strengths. Existing logit-based fusion methods maintain inference efficiency but treat vocabulary dimensions independently, overlooking semantic dependencies encoded by cross-dimension interactions. These dependencies reflect how token types interact under a model's internal reasoning and are essential for aligning models with diverse generation behaviors. To explicitly model these dependencies, we propose \textbf{InfiGFusion}, the first structure-aware fusion framework with a novel \textit{Graph-on-Logits Distillation} (GLD) loss. Specifically, we retain the top-$k$ logits per output and aggregate their outer products across sequence positions to form a global co-activation graph, where nodes represent vocabulary channels and edges quantify their joint activations. To ensure scalability and efficiency, we design a sorting-based closed-form approximation that reduces the original $O(n^4)$ cost of Gromov-Wasserstein distance to $O(n \log n)$, with provable approximation guarantees. Experiments across multiple fusion settings show that GLD consistently improves fusion quality and stability. InfiGFusion outperforms SOTA models and fusion baselines across 11 benchmarks spanning reasoning, coding, and mathematics. It shows particular strength in complex reasoning tasks, with +35.6 improvement on Multistep Arithmetic and +37.06 on Causal Judgement over SFT, demonstrating superior multi-step and relational inference.

Memory-Efficient Orthogonal Fine-Tuning with Principal Subspace Adaptation

Driven by the relentless growth in model parameters, which renders full fine-tuning prohibitively expensive for large-scale deployment, parameter-efficient fine-tuning (PEFT) has emerged as a crucial approach for rapidly adapting large models to a wide range of downstream tasks. Among the PEFT family, orthogonal fine-tuning and its variants have demonstrated remarkable performance by preserving hyperspherical energy, which encodes pairwise angular similarity between neurons. However, these methods are inherently memory-inefficient due to the need to store intermediate activations from multiple full-dimensional sparse matrices. To address this limitation, we propose Memory-efficient Orthogonal Fine-Tuning (MOFT) with principal subspace adaptation. Specifically, we first establish a theoretical condition under which orthogonal transformations within a low-rank subspace preserve hyperspherical energy. Based on this insight, we constrain orthogonal fine-tuning to the principal subspace defined by the top-r components obtained through singular value decomposition and impose an additional constraint on the projection matrix to satisfy the preservation condition. To enhance MOFT's flexibility across tasks, we relax strict orthogonality by introducing two learnable scaling vectors. Extensive experiments on 37 diverse tasks and four models across NLP and CV demonstrate that MOFT consistently outperforms key baselines while significantly reducing the memory footprint of orthogonal fine-tuning.

EcoAgent: An Efficient Edge-Cloud Collaborative Multi-Agent Framework for Mobile Automation

Cloud-based mobile agents powered by (multimodal) large language models ((M)LLMs) offer strong reasoning abilities but suffer from high latency and cost. While fine-tuned (M)SLMs enable edge deployment, they often lose general capabilities and struggle with complex tasks. To address this, we propose EcoAgent, an Edge-Cloud cOllaborative multi-agent framework for mobile automation. EcoAgent features a closed-loop collaboration among a cloud-based Planning Agent and two edge-based agents: the Execution Agent for action execution and the Observation Agent for verifying outcomes. The Observation Agent uses a Pre-Understanding Module to compress screen images into concise text, reducing token usage. In case of failure, the Planning Agent retrieves screen history and replans via a Reflection Module. Experiments on AndroidWorld show that EcoAgent maintains high task success rates while significantly reducing MLLM token consumption, enabling efficient and practical mobile automation.

Fast-Slow Thinking for Large Vision-Language Model Reasoning

Recent advances in large vision-language models (LVLMs) have revealed an \textit{overthinking} phenomenon, where models generate verbose reasoning across all tasks regardless of questions. To address this issue, we present \textbf{FAST}, a novel \textbf{Fa}st-\textbf{S}low \textbf{T}hinking framework that dynamically adapts reasoning depth based on question characteristics. Through empirical analysis, we establish the feasibility of fast-slow thinking in LVLMs by investigating how response length and data distribution affect performance. We develop FAST-GRPO with three components: model-based metrics for question characterization, an adaptive thinking reward mechanism, and difficulty-aware KL regularization. Experiments across seven reasoning benchmarks demonstrate that FAST achieves state-of-the-art accuracy with over 10\% relative improvement compared to the base model, while reducing token usage by 32.7-67.3\% compared to previous slow-thinking approaches, effectively balancing reasoning length and accuracy.