Training large foundation models costs hundreds of millions of dollars, making deployment optimization critical. Current approaches require machine learning engineers to manually craft training recipes through error-prone trial-and-error on expensive compute clusters. To enable efficient exploration of training configurations, researchers have developed performance modeling systems. However, these systems force users to translate their workloads into custom specification languages, introducing a fundamental semantic gap between the actual workload and its representation. This gap creates an inherent tradeoff: systems must either support a narrow set of workloads to maintain usability, require complex specifications that limit practical adoption, or compromise prediction accuracy with simplified models. We present Maya, a performance modeling system that eliminates these tradeoffs through transparent device emulation. By operating at the narrow interface between training frameworks and accelerator devices, Maya can capture complete workload behavior without requiring code modifications or translations. Maya intercepts device API calls from unmodified training code to directly observe low-level operations, enabling accurate performance prediction while maintaining both ease of use and generality. Our evaluation shows Maya achieves less than 5% prediction error across diverse models and optimization strategies, identifying configurations that reduce training costs by up to 56% compared to existing approaches.

Segmentation is a fundamental task in computer vision, with prompt-driven methods gaining prominence due to their flexibility. The recent Segment Anything Model (SAM) has demonstrated powerful point-prompt segmentation capabilities, while text-based segmentation models offer rich semantic understanding. However, existing approaches rarely explore how to effectively combine these complementary modalities for optimal segmentation performance. This paper presents BiPrompt-SAM, a novel dual-modal prompt segmentation framework that fuses the advantages of point and text prompts through an explicit selection mechanism. Specifically, we leverage SAM's inherent ability to generate multiple mask candidates, combined with a semantic guidance mask from text prompts, and explicitly select the most suitable candidate based on similarity metrics. This approach can be viewed as a simplified Mixture of Experts (MoE) system, where the point and text modules act as distinct "experts," and the similarity scoring serves as a rudimentary "gating network." We conducted extensive evaluations on both the Endovis17 medical dataset and RefCOCO series natural image datasets. On Endovis17, BiPrompt-SAM achieved 89.55\% mDice and 81.46\% mIoU, comparable to state-of-the-art specialized medical segmentation models. On the RefCOCO series datasets, our method attained 87.1\%, 86.5\%, and 85.8\% IoU, significantly outperforming existing approaches. Experiments demonstrate that our explicit dual-selection method effectively combines the spatial precision of point prompts with the semantic richness of text prompts, particularly excelling in scenarios involving semantically complex objects, multiple similar objects, and partial occlusions. BiPrompt-SAM not only provides a simple yet effective implementation but also offers a new perspective on multi-modal prompt fusion.

Chomsky (1968: 3) greeted the rise of computing technology with skepticism, arguing that "the kinds of structures that are realizable in terms of [computational methods ] are simply not those that must be postulated to underlie the use of language . " 55 years later, Piantadosi (2023: 15) celebrated the release of ChatGPT by directing that same criticism toward generative linguistic s: "the success of large language models is a failure for generative theories because it goes against virtually all of the principles these theories have espoused . " Chesi ( forthcoming) may not agree with Piantadosi's criticisms, but he does take them as a harbinger of scientific crisis. The minimalist program, hampered by a lack of formal and empirical rigor, has failed to produce a comprehensive, self - consistent theory of syntax. ChatG PT's apparent linguistic competence, in tandem with the success of computational accounts of gradient acceptability and online phenomena, seem to suggest that "generative linguistics no longer dictates the agenda for future linguistic challenges" ( Chesi forthcoming: 2). In order to survive, Chesi warns, generativists need to make progress towards a theory that is based on precisely stated principles and evaluated on a common set of explananda . Chesi's target paper presents the current collision of the worlds as a debate about the intellectual merits of generativist theories. According to Chesi, the success of generativism depends on generativists' ability to resolve their deficits of rigor, so that they can parry the theoretical attacks that language model s have levied against core principles of minimalism. This response argues, contrary to Chesi's framing but consistent with current consensus in the history and sociology of science (Fleck 1935; Kuhn 1962; Mullin s 1975; Latour 1984; Law & Lodge 1984), that the generativist crisis described by Piantadosi and Chesi is social in nature, and cannot be averted by intellectual means.

SARGes: Semantically Aligned Reliable Gesture Generation via Intent Chain Nan Gao 1, Yihua Bao 2, Dongdong Weng 2, Jiayi Zhao 2, Jia Li 1, Y an Zhou 3, Pengfei Wan 3, and Di Zhang 3 Abstract -- Co-speech gesture generation enhances human-computer interaction realism through speech-synchronized gesture synthesis. However, generating semantically meaningful gestures remains a challenging problem. We propose SARGes, a novel framework that leverages large language models (LLMs) to parse speech content and generate reliable semantic gesture labels, which subsequently guide the synthesis of meaningful co-speech gestures. First, we constructed a comprehensive co-speech gesture ethogram and developed an LLM-based intent chain reasoning mechanism that systematically parses and decomposes gesture semantics into structured inference steps following ethogram criteria, effectively guiding LLMs to generate context-aware gesture labels. Subsequently, we constructed an intent chain-annotated text-to-gesture label dataset and trained a lightweight gesture label generation model, which then guides the generation of credible and semantically coherent co-speech gestures. Experimental results demonstrate that SARGes achieves highly semantically-aligned gesture labeling (50.2% accuracy) with efficient single-pass inference (0.4 seconds). The proposed method provides an interpretable intent reasoning pathway for semantic gesture synthesis. I. INTRODUCTION Gestures in human communication significantly enhance semantic transmission, emotional expression, and conversational flow [1].

Recent advancements in large multimodal models have led to the emergence of remarkable generalist capabilities in digital domains, yet their translation to physical agents such as robots remains a significant challenge. This report introduces a new family of AI models purposefully designed for robotics and built upon the foundation of Gemini 2.0. We present Gemini Robotics, an advanced Vision-Language-Action (VLA) generalist model capable of directly controlling robots. Gemini Robotics executes smooth and reactive movements to tackle a wide range of complex manipulation tasks while also being robust to variations in object types and positions, handling unseen environments as well as following diverse, open vocabulary instructions. We show that with additional fine-tuning, Gemini Robotics can be specialized to new capabilities including solving long-horizon, highly dexterous tasks, learning new short-horizon tasks from as few as 100 demonstrations and adapting to completely novel robot embodiments. This is made possible because Gemini Robotics builds on top of the Gemini Robotics-ER model, the second model we introduce in this work. Gemini Robotics-ER (Embodied Reasoning) extends Gemini's multimodal reasoning capabilities into the physical world, with enhanced spatial and temporal understanding. This enables capabilities relevant to robotics including object detection, pointing, trajectory and grasp prediction, as well as multi-view correspondence and 3D bounding box predictions. We show how this novel combination can support a variety of robotics applications. We also discuss and address important safety considerations related to this new class of robotics foundation models. The Gemini Robotics family marks a substantial step towards developing general-purpose robots that realizes AI's potential in the physical world.

Recent advancements in Large Language Models (LLMs) have significantly enhanced interactions between users and models. These advancements concurrently underscore the need for rigorous safety evaluations due to the manifestation of social biases, which can lead to harmful societal impacts. Despite these concerns, existing benchmarks may overlook the intrinsic weaknesses of LLMs, which can generate biased responses even with simple adversarial instructions. To address this critical gap, we introduce a new benchmark, Fairness Benchmark in LLM under Extreme Scenarios (FLEX), designed to test whether LLMs can sustain fairness even when exposed to prompts constructed to induce bias. To thoroughly evaluate the robustness of LLMs, we integrate prompts that amplify potential biases into the fairness assessment. Comparative experiments between FLEX and existing benchmarks demonstrate that traditional evaluations may underestimate the inherent risks in models. This highlights the need for more stringent LLM evaluation benchmarks to guarantee safety and fairness.

Human-object interaction (HOI) synthesis is important for various applications, ranging from virtual reality to robotics. However, acquiring 3D HOI data is challenging due to its complexity and high cost, limiting existing methods to the narrow diversity of object types and interaction patterns in training datasets. This paper proposes a novel zero-shot HOI synthesis framework without relying on end-to-end training on currently limited 3D HOI datasets. The core idea of our method lies in leveraging extensive HOI knowledge from pre-trained Multimodal Models. Given a text description, our system first obtains temporally consistent 2D HOI image sequences using image or video generation models, which are then uplifted to 3D HOI milestones of human and object poses. We employ pre-trained human pose estimation models to extract human poses and introduce a generalizable category-level 6-DoF estimation method to obtain the object poses from 2D HOI images. Our estimation method is adaptive to various object templates obtained from text-to-3D models or online retrieval. A physics-based tracking of the 3D HOI kinematic milestone is further applied to refine both body motions and object poses, yielding more physically plausible HOI generation results. The experimental results demonstrate that our method is capable of generating open-vocabulary HOIs with physical realism and semantic diversity.

Text-to-SQL demands precise reasoning to convert natural language questions into structured queries. While large language models (LLMs) excel in many reasoning tasks, their ability to leverage Chain-of-Thought (CoT) reasoning for text-to-SQL remains underexplored. We identify critical limitations: zero-shot CoT offers minimal gains, and Direct Preference Optimization (DPO) applied without CoT yields marginal improvements. We propose ExCoT, a novel framework that iteratively optimizes open-source LLMs by combining CoT reasoning with off-policy and on-policy DPO, relying solely on execution accuracy as feedback. This approach eliminates the need for reward models or human-annotated preferences. Our experimental results demonstrate significant performance gains: ExCoT improves execution accuracy on BIRD dev set from 57.37% to 68.51% and on Spider test set from 78.81% to 86.59% for LLaMA-3 70B, with Qwen-2.5-Coder demonstrating similar improvements. Our best model achieves state-of-the-art performance in the single-model setting on both BIRD and Spider datasets, notably achieving 68.53% on the BIRD test set.

We introduce LogQuant, a groundbreaking 2-bit quantization technique for KV Cache in large language model (LLM) inference, delivering substantial memory savings while preserving superior performance. Previous methods either assume that later tokens are more important or attempt to predict important tokens based on earlier attention patterns. Both approaches, however, can result in performance bottlenecks or frequent mispredictions. LogQuant takes a different approach. By applying a log-based filtering mechanism, it selectively compresses the KV Cache across the entire context, achieving better performance with the same or even reduced memory footprint compared to existing methods. In benchmark tests, it enhances throughput by 25% and boosts batch size by 60% without increasing memory consumption. For challenging tasks such as Math and Code Completion, LogQuant improves accuracy by 40% to 200% at the same compression ratio, outperforming comparable techniques.LogQuant integrates effortlessly with popular inference frameworks like Python's transformers library. Implementation can be available in https://github.com/Concyclics/LogQuantKV.

In this challenge, we explored text-based food hazard prediction with long tail distributed classes. The task was divided into two subtasks: (1) predicting whether a web text implies one of ten food-hazard categories and identifying the associated food category, and (2) providing a more fine-grained classification by assigning a specific label to both the hazard and the product. Our findings highlight that large language model-generated synthetic data can be highly effective for oversampling long-tail distributions. Furthermore, we find that fine-tuned encoder-only, encoder-decoder, and decoder-only systems achieve comparable maximum performance across both subtasks. During this challenge, we gradually released (under CC BY-NC-SA 4.0) a novel set of 6,644 manually labeled food-incident reports.