Arctic Inference with Shift Parallelism: Fast and Efficient Open Source Inference System for Enterprise AI

Inference is now the dominant AI workload, yet existing systems force trade-offs between latency, throughput, and cost. Arctic Inference, an open-source vLLM plugin from Snowflake AI Research, introduces Shift Parallelism, a dynamic parallelism strategy that adapts to real-world traffic while integrating speculative decoding, SwiftKV compute reduction, and optimized embedding inference. It achieves up to 3.4 times faster request completion, 1.75 times faster generation, and 1.6M tokens/sec per GPU for embeddings, outperforming both latency- and throughput-optimized deployments. Already powering Snowflake Cortex AI, Arctic Inference delivers state-of-the-art, cost-effective inference for enterprise AI and is now available to the community.

FreeLoRA: Enabling Training-Free LoRA Fusion for Autoregressive Multi-Subject Personalization

Subject-driven image generation plays a crucial role in applications such as virtual try-on and poster design. Existing approaches typically fine-tune pretrained generative models or apply LoRA-based adaptations for individual subjects. However, these methods struggle with multi-subject personalization, as combining independently adapted modules often requires complex re-tuning or joint optimization. We present FreeLoRA, a simple and generalizable framework that enables training-free fusion of subject-specific LoRA modules for multi-subject personalization. Each LoRA module is adapted on a few images of a specific subject using a Full Token Tuning strategy, where it is applied across all tokens in the prompt to encourage weakly supervised token-content alignment. At inference, we adopt Subject-Aware Inference, activating each module only on its corresponding subject tokens. This enables training-free fusion of multiple personalized subjects within a single image, while mitigating overfitting and mutual interference between subjects. Extensive experiments show that FreeLoRA achieves strong performance in both subject fidelity and prompt consistency.

Exploring the Potential of Metacognitive Support Agents for Human-AI Co-Creation

Despite the potential of generative AI (GenAI) design tools to enhance design processes, professionals often struggle to integrate AI into their workflows. Fundamental cognitive challenges include the need to specify all design criteria as distinct parameters upfront (intent formulation) and designers' reduced cognitive involvement in the design process due to cognitive offloading, which can lead to insufficient problem exploration, underspecification, and limited ability to evaluate outcomes. Motivated by these challenges, we envision novel metacognitive support agents that assist designers in working more reflectively with GenAI. To explore this vision, we conducted exploratory prototyping through a Wizard of Oz elicitation study with 20 mechanical designers probing multiple metacognitive support strategies. We found that agent-supported users created more feasible designs than non-supported users, with differing impacts between support strategies. Based on these findings, we discuss opportunities and tradeoffs of metacognitive support agents and considerations for future AI-based design tools.

AWP: Activation-Aware Weight Pruning and Quantization with Projected Gradient Descent

To address the enormous size of Large Language Models (LLMs), model compression methods, such as quantization and pruning, are often deployed, especially on edge devices. In this work, we focus on layer-wise post-training quantization and pruning. Drawing connections between activation-aware weight pruning and sparse approximation problems, and motivated by the success of Iterative Hard Thresholding (IHT), we propose a unified method for Activation-aware Weight pruning and quantization via Projected gradient descent (AWP). Our experiments demonstrate that AWP outperforms state-of-the-art LLM pruning and quantization methods. Theoretical convergence guarantees of the proposed method for pruning are also provided.

Fostering Video Reasoning via Next-Event Prediction

Next-token prediction serves as the foundational learning task enabling reasoning in LLMs. But what should the learning task be when aiming to equip MLLMs with temporal reasoning capabilities over video inputs? Existing tasks such as video question answering often rely on annotations from humans or much stronger MLLMs, while video captioning tends to entangle temporal reasoning with spatial information. To address this gap, we propose next-event prediction (NEP), a learning task that harnesses future video segments as a rich, self-supervised signal to foster temporal reasoning. We segment each video into past and future frames: the MLLM takes the past frames as input and predicts a summary of events derived from the future frames, thereby encouraging the model to reason temporally in order to complete the task. To support this task, we curate V1-33K, a dataset comprising 33,000 automatically extracted video segments spanning diverse real-world scenarios. We further explore a range of video instruction-tuning strategies to study their effects on temporal reasoning. To evaluate progress, we introduce FutureBench to assess coherence in predicting unseen future events. Experiments validate that NEP offers a scalable and effective training paradigm for fostering temporal reasoning in MLLMs.

TuneComp: Joint Fine-tuning and Compression for Large Foundation Models

To reduce model size during post-training, compression methods, including knowledge distillation, low-rank approximation, and pruning, are often applied after fine-tuning the model. However, sequential fine-tuning and compression sacrifices performance, while creating a larger than necessary model as an intermediate step. In this work, we aim to reduce this gap, by directly constructing a smaller model while guided by the downstream task. We propose to jointly fine-tune and compress the model by gradually distilling it to a pruned low-rank structure. Experiments demonstrate that joint fine-tuning and compression significantly outperforms other sequential compression methods.

RLJP: Legal Judgment Prediction via First-Order Logic Rule-enhanced with Large Language Models

Legal Judgment Prediction (LJP) is a pivotal task in legal AI. Existing semantic-enhanced LJP models integrate judicial precedents and legal knowledge for high performance. But they neglect legal reasoning logic, a critical component of legal judgments requiring rigorous logical analysis. Although some approaches utilize legal reasoning logic for high-quality predictions, their logic rigidity hinders adaptation to case-specific logical frameworks, particularly in complex cases that are lengthy and detailed. This paper proposes a rule-enhanced legal judgment prediction framework based on first-order logic (FOL) formalism and comparative learning (CL) to develop an adaptive adjustment mechanism for legal judgment logic and further enhance performance in LJP. Inspired by the process of human exam preparation, our method follows a three-stage approach: first, we initialize judgment rules using the FOL formalism to capture complex reasoning logic accurately; next, we propose a Confusion-aware Contrastive Learning (CACL) to dynamically optimize the judgment rules through a quiz consisting of confusable cases; finally, we utilize the optimized judgment rules to predict legal judgments. Experimental results on two public datasets show superior performance across all metrics. The code is publicly available{https://anonymous.4open.science/r/RLJP-FDF1}.

$μ$-MoE: Test-Time Pruning as Micro-Grained Mixture-of-Experts

To tackle the huge computational demand of large foundation models, activation-aware compression techniques without retraining have been introduced. However, since these rely on calibration data, domain shift may arise for unknown downstream tasks. With a computationally efficient calibration, activation-aware pruning can be executed for every prompt adaptively, yet achieving reduced complexity at inference. We formulate it as a mixture of micro-experts, called $\mu$-MoE. Several experiments demonstrate that $\mu$-MoE can dynamically adapt to task/prompt-dependent structured sparsity on the fly.

LatentLLM: Attention-Aware Joint Tensor Compression

Modern foundation models such as large language models (LLMs) and large multi-modal models (LMMs) require a massive amount of computational and memory resources. We propose a new framework to convert such LLMs/LMMs into a reduced-dimension latent structure. Our method extends a local activation-aware tensor decomposition to a global attention-aware joint tensor de-composition. Our framework can significantly improve the model accuracy over the existing model compression methods when reducing the latent dimension to realize computationally/memory-efficient LLMs/LLMs. We show the benefit on several benchmark including multi-modal reasoning tasks.

Prosody-Adaptable Audio Codecs for Zero-Shot Voice Conversion via In-Context Learning

Recent advances in discrete audio codecs have significantly improved speech representation modeling, while codec language models have enabled in-context learning for zero-shot speech synthesis. Inspired by this, we propose a voice conversion (VC) model within the VALLE-X framework, leveraging its strong in-context learning capabilities for speaker adaptation. To enhance prosody control, we introduce a prosody-aware audio codec encoder (PACE) module, which isolates and refines prosody from other sources, improving expressiveness and control. By integrating PACE into our VC model, we achieve greater flexibility in prosody manipulation while preserving speaker timbre. Experimental evaluation results demonstrate that our approach outperforms baseline VC systems in prosody preservation, timbre consistency, and overall naturalness, surpassing baseline VC systems.