Multi-modal Large Language Models (MLLMs) integrate visual and linguistic reasoning to address complex tasks such as image captioning and visual question answering. While MLLMs demonstrate remarkable versatility, MLLMs appears limited performance on special applications. But tuning MLLMs for downstream tasks encounters two key challenges: Task-Expert Specialization, where distribution shifts between pre-training and target datasets constrain target performance, and Open-World Stabilization, where catastrophic forgetting erases the model general knowledge. In this work, we systematically review recent advancements in MLLM tuning methodologies, classifying them into three paradigms: (I) Selective Tuning, (II) Additive Tuning, and (III) Reparameterization Tuning. Furthermore, we benchmark these tuning strategies across popular MLLM architectures and diverse downstream tasks to establish standardized evaluation analysis and systematic tuning principles. Finally, we highlight several open challenges in this domain and propose future research directions. To facilitate ongoing progress in this rapidly evolving field, we provide a public repository that continuously tracks developments: https://github.com/WenkeHuang/Awesome-MLLM-Tuning.
We present Pangu Ultra, a Large Language Model (LLM) with 135 billion parameters and dense Transformer modules trained on Ascend Neural Processing Units (NPUs). Although the field of LLM has been witnessing unprecedented advances in pushing the scale and capability of LLM in recent years, training such a large-scale model still involves significant optimization and system challenges. To stabilize the training process, we propose depth-scaled sandwich normalization, which effectively eliminates loss spikes during the training process of deep models. We pre-train our model on 13.2 trillion diverse and high-quality tokens and further enhance its reasoning capabilities during post-training. To perform such large-scale training efficiently, we utilize 8,192 Ascend NPUs with a series of system optimizations. Evaluations on multiple diverse benchmarks indicate that Pangu Ultra significantly advances the state-of-the-art capabilities of dense LLMs such as Llama 405B and Mistral Large 2, and even achieves competitive results with DeepSeek-R1, whose sparse model structure contains much more parameters. Our exploration demonstrates that Ascend NPUs are capable of efficiently and effectively training dense models with more than 100 billion parameters. Our model and system will be available for our commercial customers.
Large language model (LLM) inference serving systems are essential to various LLM-based applications. As demand for LLM services continues to grow, scaling these systems to handle high request rates while meeting latency Service-Level Objectives (SLOs), referred to as effective throughput, becomes critical. However, existing systems often struggle to improve effective throughput, primarily due to a significant decline in Time To First Token (TTFT) SLO attainment. We identify two major causes of this bottleneck: (1) memory-intensive KV cache that limits batch size expansion under GPU memory constraints, and (2) rigid batch composition enforced by the default First-Come-First-Serve scheduling policy. In this paper, we introduce Apt-Serve, a scalable framework designed to enhance effective throughput in LLM inference serving. Apt-Serve features a new hybrid cache scheme that combines KV cache with a memory-efficient hidden cache for reusable input hidden state vectors, allowing large batch sizes and improving request concurrency. Based on the hybrid cache, Apt-Serve employs an adaptive runtime scheduling mechanism that dynamically optimizes batch composition. We formally define the adaptive scheduling optimization problem and propose an efficient algorithm with theoretical guarantees. Extensive evaluations on three real-world datasets and LLMs ranging from 13B to 66B parameters demonstrate that Apt-Serve achieves up to 8.8x improvement in effective throughput compared to the state-of-the-art inference serving systems.
Deep reinforcement learning (DRL) has effectively enhanced gameplay experiences and game design across various game genres. However, few studies on fighting game agents have focused explicitly on enhancing player enjoyment, a critical factor for both developers and players. To address this gap and establish a practical baseline for designing enjoyability-focused agents, we propose a two-tier agent (TTA) system and conducted experiments in the classic fighting game Street Fighter II. The first tier of TTA employs a task-oriented network architecture, modularized reward functions, and hybrid training to produce diverse and skilled DRL agents. In the second tier of TTA, a Large Language Model Hyper-Agent, leveraging players' playing data and feedback, dynamically selects suitable DRL opponents. In addition, we investigate and model several key factors that affect the enjoyability of the opponent. The experiments demonstrate improvements from 64. 36% to 156. 36% in the execution of advanced skills over baseline methods. The trained agents also exhibit distinct game-playing styles. Additionally, we conducted a small-scale user study, and the overall enjoyment in the player's feedback validates the effectiveness of our TTA system.
Every time an Interactive Storytelling (IS) system gets a player input, it is facing the world-update problem. Classical approaches to this problem consist in mapping that input to known preprogrammed actions, what can severely constrain the free will of the player. When the expected experience has a strong focus on improvisation, like in Role-playing Games (RPGs), this problem is critical. In this paper we present PAYADOR, a different approach that focuses on predicting the outcomes of the actions instead of representing the actions themselves. To implement this approach, we ground a Large Language Model to a minimal representation of the fictional world, obtaining promising results. We make this contribution open-source, so it can be adapted and used for other related research on unleashing the co-creativity power of RPGs.
Blood cultures are often over ordered without clear justification, straining healthcare resources and contributing to inappropriate antibiotic use pressures worsened by the global shortage. In study of 135483 emergency department (ED) blood culture orders, we developed machine learning (ML) models to predict the risk of bacteremia using structured electronic health record (EHR) data and provider notes via a large language model (LLM). The structured models AUC improved from 0.76 to 0.79 with note embeddings and reached 0.81 with added diagnosis codes. Compared to an expert recommendation framework applied by human reviewers and an LLM-based pipeline, our ML approach offered higher specificity without compromising sensitivity. The recommendation framework achieved sensitivity 86%, specificity 57%, while the LLM maintained high sensitivity (96%) but over classified negatives, reducing specificity (16%). These findings demonstrate that ML models integrating structured and unstructured data can outperform consensus recommendations, enhancing diagnostic stewardship beyond existing standards of care.
This paper proposes a novel framework utilizing multi-modal large language models (MLLMs) for referring video object segmentation (RefVOS). Previous MLLM-based methods commonly struggle with the dilemma between "Ref" and "VOS": they either specialize in understanding a few key frames (global reasoning) or tracking objects on continuous frames (local reasoning), and rely on external VOS or frame selectors to mitigate the other end of the challenge. However, our framework GLUS shows that global and local consistency can be unified into a single video segmentation MLLM: a set of sparse "context frames" provides global information, while a stream of continuous "query frames" conducts local object tracking. This is further supported by jointly training the MLLM with a pre-trained VOS memory bank to simultaneously digest short-range and long-range temporal information. To improve the information efficiency within the limited context window of MLLMs, we introduce object contrastive learning to distinguish hard false-positive objects and a self-refined framework to identify crucial frames and perform propagation. By collectively integrating these insights, our GLUS delivers a simple yet effective baseline, achieving new state-of-the-art for MLLMs on the MeViS and Ref-Youtube-VOS benchmark. Our project page is at https://glus-video.github.io/.
Modern Large Language Model (LLM) systems typically rely on Retrieval Augmented Generation (RAG) which aims to gather context that is useful for response generation. These RAG systems typically optimize strictly towards retrieving context that is maximally relevant to the query. However, conventional theory suggests that retrieval systems which seek to maximize context relevance without any additional explicit criteria can create information bottlenecks. We reaffirm this finding in the modern age of LLM's by showing that in standard RAG pipelines, maximizing for context relevance alone can degrade downstream response quality. In response, we show evaluations of existing RAG methods which account for both context relevance and answer quality. These evaluations introduce a novel finding that existing RAG systems scale poorly with inference time compute usage when considering our combined metric. We introduce "RErank BEyond reLevance (REBEL)", which enables RAG systems to scale with inference-time compute via injection of multi-criteria optimization using Chain-of-Thought prompting (and optionally Multi-Turn dialogue). Ultimately, this enables a new performance/speed tradeoff curve, where RAG systems are able to achieve both higher relevance of retrieved contexts and superior answer quality as inference time increases. Code for the implementation of our method in llama-index can be found at the following PR: https://github.com/run-llama/llama_index/pull/17590. Code for running experiments using this llama-index implementation can be found at https://github.com/microsoft/REBEL.
As demand for Large Language Models (LLMs) and AI agents rapidly grows, optimizing systems for efficient LLM inference becomes critical. While significant efforts have focused on system-level engineering, little is explored from a mathematical modeling and queuing perspective. In this paper, we aim to develop the queuing fundamentals for large language model (LLM) inference, bridging the gap between the queueing theory and LLM system communities. In particular, we study the throughput aspect in LLM inference systems. We prove that a large class of 'work-conserving' scheduling algorithms can achieve maximum throughput for individual inference LLM engine, highlighting 'work-conserving' as a key design principle in practice. In a network of LLM agents, work-conserving scheduling alone is insufficient, particularly when facing specific workload structures and multi-class workflows that require more sophisticated scheduling strategies. Evaluations of real-world systems show that Orca and Sarathi-serve are throughput-optimal, reassuring practitioners, while FasterTransformer and vanilla vLLM are not maximally stable and should be used with caution. Our results highlight the substantial benefits that the queueing community can offer in improving LLM inference systems and call for more interdisciplinary development.
Image restoration (IR) often faces various complex and unknown degradations in real-world scenarios, such as noise, blurring, compression artifacts, and low resolution, etc. Training specific models for specific degradation may lead to poor generalization. To handle multiple degradations simultaneously, All-in-One models might sacrifice performance on certain types of degradation and still struggle with unseen degradations during training. Existing IR agents rely on multimodal large language models (MLLM) and a time-consuming rolling-back selection strategy neglecting image quality. As a result, they may misinterpret degradations and have high time and computational costs to conduct unnecessary IR tasks with redundant order. To address these, we propose a Quality-Driven agent (Q-Agent) via Chain-of-Thought (CoT) restoration. Specifically, our Q-Agent consists of robust degradation perception and quality-driven greedy restoration. The former module first fine-tunes MLLM, and uses CoT to decompose multi-degradation perception into single-degradation perception tasks to enhance the perception of MLLMs. The latter employs objective image quality assessment (IQA) metrics to determine the optimal restoration sequence and execute the corresponding restoration algorithms. Experimental results demonstrate that our Q-Agent achieves superior IR performance compared to existing All-in-One models.
This paper proposes DeepGreen, an Large Language Model Driven (LLM-Driven) system for detecting corporate green-washing behaviour. Utilizing dual-layer LLM analysis, DeepGreen preliminarily identifies potential green keywords in financial statements and then assesses their implementation degree via iterative semantic analysis of LLM. A core variable GreenImplement is derived from the ratio from the two layers' output. We extract 204 financial statements of 68 companies from A-share market over three years, comprising 89,893 words, and analyse them through DeepGreen. Our analysis, supported by violin plots and K-means clustering, reveals insights and validates the variable against the Huazheng ESG rating. It offers a novel perspective for regulatory agencies and investors, serving as a proactive monitoring tool that complements traditional methods.Empirical tests show that green implementation can significantly boost the asset return rate of companies, but there is heterogeneity in scale. Small and medium-sized companies have limited contribution to asset return via green implementation, so there is a stronger motivation for green-washing.
Model Predictive Control (MPC) is a powerful control strategy widely utilized in domains like energy management, building control, and autonomous systems. However, its effectiveness in real-world settings is challenged by the need to incorporate context-specific predictions and expert instructions, which traditional MPC often neglects. We propose InstructMPC, a novel framework that addresses this gap by integrating real-time human instructions through a Large Language Model (LLM) to produce context-aware predictions for MPC. Our method employs a Language-to-Distribution (L2D) module to translate contextual information into predictive disturbance trajectories, which are then incorporated into the MPC optimization. Unlike existing context-aware and language-based MPC models, InstructMPC enables dynamic human-LLM interaction and fine-tunes the L2D module in a closed loop with theoretical performance guarantees, achieving a regret bound of $O(\sqrt{T\log T})$ for linear dynamics when optimized via advanced fine-tuning methods such as Direct Preference Optimization (DPO) using a tailored loss function.
Large Language Models (LLMs) have demonstrated remarkable capabilities in code generation. However, the quality of the generated code is heavily dependent on the structure and composition of the prompts used. Crafting high-quality prompts is a challenging task that requires significant knowledge and skills of prompt engineering. To advance the automation support for the prompt engineering for LLM-based code generation, we propose a novel solution Diffusion-Driven Prompt Tuning (DDPT) that learns how to generate optimal prompt embedding from Gaussian Noise to automate the prompt engineering for code generation. We evaluate the feasibility of diffusion-based optimization and abstract the optimal prompt embedding as a directional vector toward the optimal embedding. We use the code generation loss given by the LLMs to help the diffusion model capture the distribution of optimal prompt embedding during training. The trained diffusion model can build a path from the noise distribution to the optimal distribution at the sampling phrase, the evaluation result demonstrates that DDPT helps improve the prompt optimization for code generation.
Software requirements expressed in natural language (NL) frequently suffer from verbosity, ambiguity, and inconsistency. This creates a range of challenges, including selecting an appropriate architecture for a system and assessing different architectural alternatives. Relying on human expertise to accomplish the task of mapping NL requirements to architecture is time-consuming and error-prone. This paper proposes ARLO, an approach that automates this task by leveraging (1) a set of NL requirements for a system, (2) an existing standard that specifies architecturally relevant software quality attributes, and (3) a readily available Large Language Model (LLM). Specifically, ARLO determines the subset of NL requirements for a given system that is architecturally relevant and maps that subset to a tailorable matrix of architectural choices. ARLO applies integer linear programming on the architectural-choice matrix to determine the optimal architecture for the current requirements. We demonstrate ARLO's efficacy using a set of real-world examples. We highlight ARLO's ability (1) to trace the selected architectural choices to the requirements and (2) to isolate NL requirements that exert a particular influence on a system's architecture. This allows the identification, comparative assessment, and exploration of alternative architectural choices based on the requirements and constraints expressed therein.
This paper presents Thanos, a novel weight-pruning algorithm designed to reduce the memory footprint and enhance the computational efficiency of large language models (LLMs) by removing redundant weights while maintaining accuracy. Thanos introduces a block-wise pruning strategy with adaptive masks that dynamically adjust to weight importance, enabling flexible sparsity patterns and structured formats, such as $n:m$ sparsity, optimized for hardware acceleration. Experimental evaluations demonstrate that Thanos achieves state-of-the-art performance in structured pruning and outperforms existing methods in unstructured pruning. By providing an efficient and adaptable approach to model compression, Thanos offers a practical solution for deploying large models in resource-constrained environments.
This paper introduces the art project The Dream Within Huang Long Cave, an AI-driven interactive and immersive narrative experience. The project offers new insights into AI technology, artistic practice, and psychoanalysis. Inspired by actual geographical landscapes and familial archetypes, the work combines psychoanalytic theory and computational technology, providing an artistic response to the concept of the non-existence of the Big Other. The narrative is driven by a combination of a large language model (LLM) and a realistic digital character, forming a virtual agent named YELL. Through dialogue and exploration within a cave automatic virtual environment (CAVE), the audience is invited to unravel the language puzzles presented by YELL and help him overcome his life challenges. YELL is a fictional embodiment of the Big Other, modeled after the artist's real father. Through a cross-temporal interaction with this digital father, the project seeks to deconstruct complex familial relationships. By demonstrating the non-existence of the Big Other, we aim to underscore the authenticity of interpersonal emotions, positioning art as a bridge for emotional connection and understanding within family dynamics.
Foundation models have demonstrated strong reasoning and generalization capabilities in driving-related tasks, including scene understanding, planning, and control. However, they still face challenges in hallucinations, uncertainty, and long inference latency. While existing foundation models have general knowledge of avoiding collisions, they often lack transportation-specific safety knowledge. To overcome these limitations, we introduce LetsPi, a physics-informed, dual-phase, knowledge-driven framework for safe, human-like trajectory planning. To prevent hallucinations and minimize uncertainty, this hybrid framework integrates Large Language Model (LLM) reasoning with physics-informed social force dynamics. LetsPi leverages the LLM to analyze driving scenes and historical information, providing appropriate parameters and target destinations (goals) for the social force model, which then generates the future trajectory. Moreover, the dual-phase architecture balances reasoning and computational efficiency through its Memory Collection phase and Fast Inference phase. The Memory Collection phase leverages the physics-informed LLM to process and refine planning results through reasoning, reflection, and memory modules, storing safe, high-quality driving experiences in a memory bank. Surrogate safety measures and physics-informed prompt techniques are introduced to enhance the LLM's knowledge of transportation safety and physical force, respectively. The Fast Inference phase extracts similar driving experiences as few-shot examples for new scenarios, while simplifying input-output requirements to enable rapid trajectory planning without compromising safety. Extensive experiments using the HighD dataset demonstrate that LetsPi outperforms baseline models across five safety metrics.See PDF for project Github link.
The human face plays a central role in social communication, necessitating the use of performant computer vision tools for human-centered applications. We propose Face-LLaVA, a multimodal large language model for face-centered, in-context learning, including facial expression and attribute recognition. Additionally, Face-LLaVA is able to generate natural language descriptions that can be used for reasoning. Leveraging existing visual databases, we first developed FaceInstruct-1M, a face-centered database for instruction tuning MLLMs for face processing. We then developed a novel face-specific visual encoder powered by Face-Region Guided Cross-Attention that integrates face geometry with local visual features. We evaluated the proposed method across nine different datasets and five different face processing tasks, including facial expression recognition, action unit detection, facial attribute detection, age estimation and deepfake detection. Face-LLaVA achieves superior results compared to existing open-source MLLMs and competitive performance compared to commercial solutions. Our model output also receives a higher reasoning rating by GPT under a zero-shot setting across all the tasks. Both our dataset and model wil be released at https://face-llava.github.io to support future advancements in social AI and foundational vision-language research.
Large language model-based agents are becoming increasingly popular as a low-cost mechanism to provide personalized, conversational advice, and have demonstrated impressive capabilities in relatively simple scenarios, such as movie recommendations. But how do these agents perform in complex high-stakes domains, where domain expertise is essential and mistakes carry substantial risk? This paper investigates the effectiveness of LLM-advisors in the finance domain, focusing on three distinct challenges: (1) eliciting user preferences when users themselves may be unsure of their needs, (2) providing personalized guidance for diverse investment preferences, and (3) leveraging advisor personality to build relationships and foster trust. Via a lab-based user study with 64 participants, we show that LLM-advisors often match human advisor performance when eliciting preferences, although they can struggle to resolve conflicting user needs. When providing personalized advice, the LLM was able to positively influence user behavior, but demonstrated clear failure modes. Our results show that accurate preference elicitation is key, otherwise, the LLM-advisor has little impact, or can even direct the investor toward unsuitable assets. More worryingly, users appear insensitive to the quality of advice being given, or worse these can have an inverse relationship. Indeed, users reported a preference for and increased satisfaction as well as emotional trust with LLMs adopting an extroverted persona, even though those agents provided worse advice.
The computational complexity of large language model (LLM) inference significantly constrains their deployment efficiency on edge devices. In contrast, small language models offer faster decoding and lower resource consumption but often suffer from degraded response quality and heightened susceptibility to hallucinations. To address this trade-off, collaborative decoding, in which a large model assists in generating critical tokens, has emerged as a promising solution. This paradigm leverages the strengths of both model types by enabling high-quality inference through selective intervention of the large model, while maintaining the speed and efficiency of the smaller model. In this work, we present a novel collaborative decoding inference system that allows small models to perform on-device inference while selectively consulting a cloud-based large model for critical token generation. Remarkably, the system achieves a 60% performance gain on CommonsenseQA using only a 0.5B model on an M1 MacBook, with under 7% of tokens generation uploaded to the large model in the cloud.