Recent advancements in large language models (LLMs) showcase varied multilingual capabilities across tasks like translation, code generation, and reasoning. Previous assessments often limited their scope to fundamental natural language processing (NLP) or isolated capability-specific tasks. To alleviate this drawback, we aim to present a comprehensive multilingual multitask benchmark. First, we present a pipeline for selecting available and reasonable benchmarks from massive ones, addressing the oversight in previous work regarding the utility of these benchmarks, i.e., their ability to differentiate between models being evaluated. Leveraging this pipeline, we introduce P-MMEval, a large-scale benchmark covering effective fundamental and capability-specialized datasets. Furthermore, P-MMEval delivers consistent language coverage across various datasets and provides parallel samples. Finally, we conduct extensive experiments on representative multilingual model series to compare performances across models, analyze dataset effectiveness, examine prompt impacts on model performances, and explore the relationship between multilingual performances and factors such as tasks, model sizes, and languages. These insights offer valuable guidance for future research. The dataset is available at https://huggingface.co/datasets/Qwen/P-MMEval.

Recent advancements in Large Video-Language Models (LVLMs) have driven the development of benchmarks designed to assess cognitive abilities in video-based tasks. However, most existing benchmarks heavily rely on web-collected videos paired with human annotations or model-generated questions, which limit control over the video content and fall short in evaluating advanced cognitive abilities involving symbolic elements and abstract concepts. To address these limitations, we introduce VCBench, a controllable benchmark to assess LVLMs' cognitive abilities, involving symbolic and abstract concepts at varying difficulty levels. By generating video data with the Python-based engine, VCBench allows for precise control over the video content, creating dynamic, task-oriented videos that feature complex scenes and abstract concepts. Each task pairs with tailored question templates that target specific cognitive challenges, providing a rigorous evaluation test. Our evaluation reveals that even state-of-the-art (SOTA) models, such as Qwen2-VL-72B, struggle with simple video cognition tasks involving abstract concepts, with performance sharply dropping by 19% as video complexity rises. These findings reveal the current limitations of LVLMs in advanced cognitive tasks and highlight the critical role of VCBench in driving research toward more robust LVLMs for complex video cognition challenges.

Code-mixing(CM) or code-switching(CSW) refers to the juxtaposition of linguistic units from two or more languages during the conversation or sometimes even a single utterance. Code-mixing introduces unique challenges in daily life, such as syntactic mismatches and semantic blending, that are rarely encountered in monolingual settings. Large language models (LLMs) have revolutionized the field of natural language processing (NLP) by offering unprecedented capabilities in understanding human languages. However, the effectiveness of current state-of-the-art multilingual LLMs has not yet been fully explored in the CM scenario. To fill this gap, we first benchmark the performance of multilingual LLMs on various code-mixing NLP tasks. Then we propose to improve the multilingual LLMs' ability to understand code-mixing through reinforcement learning from human feedback (RLHF) and code-mixed machine translation tasks. Given the high-cost and time-consuming preference labeling procedure, we improve this by utilizing LLMs as annotators to perform the reinforcement learning from AI feedback (RLAIF). The experiments show the effectiveness of the proposed method.

Robotics tasks are highly compositional by nature. For example, to perform a high-level task like cleaning the table a robot must employ low-level capabilities of moving the effectors to the objects on the table, pick them up and then move them off the table one-by-one, while re-evaluating the consequently dynamic scenario in the process. Given that large vision language models (VLMs) have shown progress on many tasks that require high level, human-like reasoning, we ask the question: if the models are taught the requisite low-level capabilities, can they compose them in novel ways to achieve interesting high-level tasks like cleaning the table without having to be explicitly taught so? To this end, we present ClevrSkills - a benchmark suite for compositional reasoning in robotics. ClevrSkills is an environment suite developed on top of the ManiSkill2 simulator and an accompanying dataset. The dataset contains trajectories generated on a range of robotics tasks with language and visual annotations as well as multi-modal prompts as task specification. The suite includes a curriculum of tasks with three levels of compositional understanding, starting with simple tasks requiring basic motor skills. We benchmark multiple different VLM baselines on ClevrSkills and show that even after being pre-trained on large numbers of tasks, these models fail on compositional reasoning in robotics tasks.

Using Large Language Models (LLMs) to address critical societal problems requires adopting this novel technology into socio-technical systems. However, the complexity of such systems and the nature of LLMs challenge such a vision. It is unlikely that the solution to such challenges will come from the Artificial Intelligence (AI) community itself. Instead, the Systems Engineering approach is better equipped to facilitate the adoption of LLMs by prioritising the problems and their context before any other aspects. This paper introduces the challenges LLMs generate and surveys systems research efforts for engineering AI-based systems. We reveal how the systems engineering principles have supported addressing similar issues to the ones LLMs pose and discuss our findings to provide future directions for adopting LLMs.

Large Language Models (LLMs) have demonstrated significant reasoning capabilities in robotic systems. However, their deployment in multi-robot systems remains fragmented and struggles to handle complex task dependencies and parallel execution. This study introduces the DART-LLM (Dependency-Aware Multi-Robot Task Decomposition and Execution using Large Language Models) system, designed to address these challenges. DART-LLM utilizes LLMs to parse natural language instructions, decomposing them into multiple subtasks with dependencies to establish complex task sequences, thereby enhancing efficient coordination and parallel execution in multi-robot systems. The system includes the QA LLM module, Breakdown Function modules, Actuation module, and a Vision-Language Model (VLM)-based object detection module, enabling task decomposition and execution from natural language instructions to robotic actions. Experimental results demonstrate that DART-LLM excels in handling long-horizon tasks and collaborative tasks with complex dependencies. Even when using smaller models like Llama 3.1 8B, the system achieves good performance, highlighting DART-LLM's robustness in terms of model size. Please refer to the project website \url{https://wyd0817.github.io/project-dart-llm/} for videos and code.

Recent research increasingly focuses on training vision-language models (VLMs) with long, detailed image captions. However, small-scale VLMs often struggle to balance the richness of these captions with the risk of hallucinating content during fine-tuning. In this paper, we explore how well VLMs adapt to such captions. To quantify caption quality, we propose Decomposed NLI (DNLI), an evaluation framework that breaks down generated captions into individual propositions, assessing each in isolation. This fine-grained analysis reveals a critical balance between capturing descriptive details and preventing hallucinations. Our findings show that simply reducing caption complexity or employing standard data curation techniques does not effectively resolve this issue. To tackle this challenge, we introduce Knowledge Adapted (KnowAda) fine-tuning, a data-centric approach that automatically adapts training data with the model's existing knowledge and visual understanding. KnowAda minimizes hallucinations while preserving high descriptiveness. We validate this approach across several small-scale VLMs (up to 7B parameters) and dense caption datasets, demonstrating that KnowAda effectively balances hallucination reduction and descriptiveness. Our results show that KnowAda outperforms various baselines in both automatic metrics and human evaluations. We will release our code and models.

As language models grow ever larger, so do their vocabularies. This has shifted the memory footprint of LLMs during training disproportionately to one single layer: the cross-entropy in the loss computation. Cross-entropy builds up a logit matrix with entries for each pair of input tokens and vocabulary items and, for small models, consumes an order of magnitude more memory than the rest of the LLM combined. We propose Cut Cross-Entropy (CCE), a method that computes the cross-entropy loss without materializing the logits for all tokens into global memory. Rather, CCE only computes the logit for the correct token and evaluates the log-sum-exp over all logits on the fly. We implement a custom kernel that performs the matrix multiplications and the log-sum-exp reduction over the vocabulary in flash memory, making global memory consumption for the cross-entropy computation negligible. This has a dramatic effect. Taking the Gemma 2 (2B) model as an example, CCE reduces the memory footprint of the loss computation from 24 GB to 1 MB, and the total training-time memory consumption of the classifier head from 28 GB to 1 GB. To improve the throughput of CCE, we leverage the inherent sparsity of softmax and propose to skip elements of the gradient computation that have a negligible (i.e., below numerical precision) contribution to the gradient. Experiments demonstrate that the dramatic reduction in memory consumption is accomplished without sacrificing training speed or convergence.

Mixture-of-Experts (MoE) architectures have recently gained popularity in enabling efficient scaling of large language models. However, we uncover a fundamental tension: while MoEs are designed for selective expert activation, production serving requires request batching, which forces the activation of all experts and negates MoE's efficiency benefits during the decode phase. We present Lynx, a system that enables efficient MoE inference through dynamic, batch-aware expert selection. Our key insight is that expert importance varies significantly across tokens and inference phases, creating opportunities for runtime optimization. Lynx leverages this insight through a lightweight framework that dynamically reduces active experts while preserving model accuracy. Our evaluations show that Lynx achieves up to 1.55x reduction in inference latency while maintaining negligible accuracy loss from baseline model across complex code generation and mathematical reasoning tasks.

Text classification is a fundamental task in natural language processing (NLP), and large language models (LLMs) have demonstrated their capability to perform this task across various domains. However, the performance of LLMs heavily depends on the quality of their input prompts. Recent studies have also shown that LLMs exhibit remarkable results in code-related tasks. To leverage the capabilities of LLMs in text classification, we propose the Code Completion Prompt (CoCoP) method, which transforms the text classification problem into a code completion task. CoCoP significantly improves text classification performance across diverse datasets by utilizing LLMs' code-completion capability. For instance, CoCoP enhances the accuracy of the SST2 dataset by more than 20%. Moreover, when CoCoP integrated with LLMs specifically designed for code-related tasks (code models), such as CodeLLaMA, this method demonstrates better or comparable performance to few-shot learning techniques while using only one-tenth of the model size. The source code of our proposed method will be available to the public upon the acceptance of the paper.