Jina-VLM: Small Multilingual Vision Language Model

LeMat-GenBench has been introduced as a unified evaluation framework for generative models of crystalline materials, addressing the challenges posed by the lack of standardized metrics in the field. This framework includes an open-source evaluation suite and a public leaderboard on Hugging Face, benchmarking 12 recent generative models and revealing insights into the trade-offs between stability, novelty, and diversity in model performance.

Large-scale Mixture of Experts (MoE) Large Language Models (LLMs) have emerged as leading open-weight models, but their random expert selection mechanism leads to significant data movement overhead. A recent study conducted comprehensive profiling across four state-of-the-art MoE models, revealing insights that can enhance future serving systems and reduce bottlenecks in multi-unit LLM serving.

A new approach called Semantic Soft Bootstrapping (SSB) has been proposed to enhance long context reasoning in large language models (LLMs) without relying on reinforcement learning. This self-distillation technique allows the model to act as both teacher and student, improving its reasoning capabilities by providing varied semantic contexts during training.

A recent study has introduced a taxonomy for cataloguing Open-source Pre-Trained Models (PTMs) from Hugging Face, specifically tailored to Software Engineering (SE) tasks. This classification encompasses 147 SE tasks, aiming to enhance the identification and reuse of models for software development activities. The research involved a comprehensive five-phase methodology, including data collection and validation processes.

A recent study has demonstrated the potential of small language models (SLMs) to effectively support multimodal search and recommendation tasks, utilizing a framework that integrates upside-down reinforcement learning and synthetic data distillation from larger models like Llama-3. The 100M-parameter GPT-2 model achieved relevance and diversity scores comparable to larger counterparts while significantly reducing inference latency and memory overhead.

A novel multilingual training-free approach to remote sensing image captioning has been proposed, utilizing retrieval-augmented prompting to generate captions without the need for large annotated datasets. This method employs a domain-adapted SigLIP2 encoder to retrieve relevant captions and examples, which are processed by a language model in both image-blind and image-aware setups.

The recent introduction of DESIGNER, a design-logic-guided reasoning data synthesis pipeline, aims to enhance the capabilities of large language models (LLMs) in tackling complex, multidisciplinary questions. By leveraging extensive raw documents, DESIGNER generates high-difficulty questions that challenge LLMs' reasoning abilities across various disciplines.

Recent research has introduced ReMindView-Bench, a benchmark designed to evaluate how Vision-Language Models (VLMs) construct and maintain spatial mental models across multiple viewpoints. This initiative addresses the challenges VLMs face in achieving geometric coherence and cross-view consistency in spatial reasoning tasks, which are crucial for understanding 3D environments.