This paper introduces SkillComposer, a method for structured skill composition in LLM agents, which predicts an executable skill plan that jointly specifies the activated subset, count, and execution…

The paper proposes ReuseRL, a method that improves agent generalization in Reinforcement Learning by enforcing structural compressibility of successful agent trajectories into reusable skills.

Mellum 2 is an open-weight 12B Mixture-of-Experts (MoE) language model specialized for software engineering, achieving performance competitive with larger models while maintaining the efficiency of a…

This paper investigates the relationship between language model perplexity and word error rate in modern automatic speech recognition systems, studying the impact of external language models, encoder…

This paper proposes a method for creating scalable browser agents by cloning user interaction skills from human browsing data using natural language skills and a skill graph.

The paper proposes DAG-MoE, a novel sparse Mixture-of-Experts framework that replaces standard weighted-sum aggregation with structural aggregation to enhance model performance and enable multi-step r…

SkillTrojan introduces a novel backdoor attack targeting the composition of reusable skills in agent systems, demonstrating high attack success rates with minimal impact on normal system functionality…

The paper introduces SkillReact, a framework that measures compositional risk in agent skill ecosystems, finding that even if individual skills are safe, their combination can create significant, unad…

The paper introduces SkillReact, a framework that measures compositional risk in agent skill ecosystems, finding that even if individual skills are safe, their combination can create significant, expl…

The paper introduces a novel, transferable learned attack (LT-MIA) that detects a universal 'signature of memorization' in language models, achieving high accuracy across diverse model architectures (…

The paper introduces PortBERT, a family of RoBERTa-based language models for Portuguese, which achieves competitive performance while explicitly balancing efficiency and accuracy.

This paper compares the performance of recurrent decoders and selective state-space models (Mamba) in intracortical brain-to-text systems, and investigates the impact of output targets (phonetic vs. c…

dMoE proposes a block-level Mixture-of-Experts (MoE) framework for Diffusion Large Language Models (dLLMs) that aggregates token-level expert distributions into a unified block-level distribution, sig…

The paper introduces XLGoBench, a synthetic benchmark of algorithmic tasks designed to detect persistent cross-lingual skill gaps in large language models.

The paper introduces MASA, a model-aware skill alignment framework that adaptively rewrites general and task-specific skills for LLM agents, achieving superior performance across diverse backbones and…

The paper introduces a systematic framework to convert large Mixture-of-Experts (MoE) models into memory-efficient, fully dense architectures, achieving superior performance compared to traditional pr…

The paper introduces SkillCenter, a large open skill library for AI agents, with over 216,000 skills from various sources, and presents an end-to-end framework for acquiring, filtering, generating, gr…

The paper introduces a diagnostic framework to decompose multilingual LLM performance variance, showing that language identity and model-benchmark interactions are key drivers of performance gaps.

This paper proposes a method for language identification using compositional vectors and the centered log-ratio transformation, achieving robust accuracy and strong performance for longer sequences.

The paper proposes a unified framework for designing efficient and expressive token mixing layers by separating the direct and recurrent influences of inputs, allowing for a principled trade-off betwe…