We present a Japanese domain-specific language model for the pharmaceutical field, developed through continual pretraining on 2 billion Japanese pharmaceutical tokens and 8 billion English biomedical tokens. To enable rigorous evaluation, we introduce three new benchmarks: YakugakuQA, based on national pharmacist licensing exams; NayoseQA, which tests cross-lingual synonym and terminology normalization; and SogoCheck, a novel task designed to assess consistency reasoning between paired statements. We evaluate our model against both open-source medical LLMs and commercial models, including GPT-4o. Results show that our domain-specific model outperforms existing open models and achieves competitive performance with commercial ones, particularly on terminology-heavy and knowledge-based tasks. Interestingly, even GPT-4o performs poorly on SogoCheck, suggesting that cross-sentence consistency reasoning remains an open challenge. Our benchmark suite offers a broader diagnostic lens for pharmaceutical NLP, covering factual recall, lexical variation, and logical consistency. This work demonstrates the feasibility of building practical, secure, and cost-effective language models for Japanese domain-specific applications, and provides reusable evaluation resources for future research in pharmaceutical and healthcare NLP. Our model, codes, and datasets are released at https://github.com/EQUES-Inc/pharma-LLM-eval.

Large Language Models (LLMs), despite their remarkable capabilities, are hampered by hallucinations. A particularly challenging variant, knowledge overshadowing, occurs when one piece of activated knowledge inadvertently masks another relevant piece, leading to erroneous outputs even with high-quality training data. Current understanding of overshadowing is largely confined to inference-time observations, lacking deep insights into its origins and internal mechanisms during model training. Therefore, we introduce PhantomCircuit, a novel framework designed to comprehensively analyze and detect knowledge overshadowing. By innovatively employing knowledge circuit analysis, PhantomCircuit dissects the function of key components in the circuit and how the attention pattern dynamics contribute to the overshadowing phenomenon and its evolution throughout the training process. Extensive experiments demonstrate PhantomCircuit's effectiveness in identifying such instances, offering novel insights into this elusive hallucination and providing the research community with a new methodological lens for its potential mitigation.

Generative models of 3D molecular structure play a rapidly growing role in the design and simulation of molecules. Diffusion models currently dominate the space of 3D molecule generation, while autoregressive models have trailed behind. In this work, we present Quetzal, a simple but scalable autoregressive model that builds molecules atom-by-atom in 3D. Treating each molecule as an ordered sequence of atoms, Quetzal combines a causal transformer that predicts the next atom's discrete type with a smaller Diffusion MLP that models the continuous next-position distribution. Compared to existing autoregressive baselines, Quetzal achieves substantial improvements in generation quality and is competitive with the performance of state-of-the-art diffusion models. In addition, by reducing the number of expensive forward passes through a dense transformer, Quetzal enables significantly faster generation speed, as well as exact divergence-based likelihood computation. Finally, without any architectural changes, Quetzal natively handles variable-size tasks like hydrogen decoration and scaffold completion. We hope that our work motivates a perspective on scalability and generality for generative modelling of 3D molecules.

While Multimodal Large Language Models (MLLMs) excel at general vision-language tasks, visuospatial cognition - reasoning about spatial layouts, relations, and dynamics - remains a significant challenge. Existing models often lack the necessary architectural components and specialized training data for fine-grained spatial understanding. We introduce ViCA2 (Visuospatial Cognitive Assistant 2), a novel MLLM designed to enhance spatial reasoning. ViCA2 features a dual vision encoder architecture integrating SigLIP for semantics and Hiera for spatial structure, coupled with a token ratio control mechanism for efficiency. We also developed ViCA-322K, a new large-scale dataset with over 322,000 spatially grounded question-answer pairs for targeted instruction tuning. On the challenging VSI-Bench benchmark, our ViCA2-7B model achieves a state-of-the-art average score of 56.8, significantly surpassing larger open-source models (e.g., LLaVA-NeXT-Video-72B, 40.9) and leading proprietary models (Gemini-1.5 Pro, 45.4). This demonstrates the effectiveness of our approach in achieving strong visuospatial intelligence with a compact model. We release ViCA2, its codebase, and the ViCA-322K dataset to facilitate further research.

A recent trend in LLMs is developing recurrent sub-quadratic models that improve long-context processing efficiency. We investigate leading large long-context models, focusing on how their fixed-size recurrent memory affects their performance. Our experiments reveal that, even when these models are trained for extended contexts, their use of long contexts remains underutilized. Specifically, we demonstrate that a chunk-based inference procedure, which identifies and processes only the most relevant portion of the input can mitigate recurrent memory failures and be effective for many long-context tasks: On LongBench, our method improves the overall performance of Falcon3-Mamba-Inst-7B by 14%, Falcon-Mamba-Inst-7B by 28%, RecurrentGemma-IT-9B by 50%, and RWKV6-Finch-7B by 51%. Surprisingly, this simple approach also leads to state-of-the-art results in the challenging LongBench v2 benchmark, showing competitive performance with equivalent size Transformers. Furthermore, our findings raise questions about whether recurrent models genuinely exploit long-range dependencies, as our single-chunk strategy delivers stronger performance - even in tasks that presumably require cross-context relations.

This paper presents a comprehensive evaluation of lightweight deep learning models for image classification, emphasizing their suitability for deployment in resource-constrained environments such as low-memory devices. Five state-of-the-art architectures - MobileNetV3 Small, ResNet18, SqueezeNet, EfficientNetV2-S, and ShuffleNetV2 - are benchmarked across three diverse datasets: CIFAR-10, CIFAR-100, and Tiny ImageNet. The models are assessed using four key performance metrics: classification accuracy, inference time, floating-point operations (FLOPs), and model size. Additionally, we investigate the impact of hyperparameter tuning, data augmentation, and training paradigms by comparing pretrained models with scratch-trained counterparts, focusing on MobileNetV3 Small. Our findings reveal that transfer learning significantly enhances model accuracy and computational efficiency, particularly for complex datasets like Tiny ImageNet. EfficientNetV2 consistently achieves the highest accuracy, while MobileNetV3 offers the best balance between accuracy and efficiency, and SqueezeNet excels in inference speed and compactness. This study highlights critical trade-offs between accuracy and efficiency, offering actionable insights for deploying lightweight models in real-world applications where computational resources are limited. By addressing these challenges, this research contributes to optimizing deep learning systems for edge computing and mobile platforms.

We introduce the Llama-Nemotron series of models, an open family of heterogeneous reasoning models that deliver exceptional reasoning capabilities, inference efficiency, and an open license for enterprise use. The family comes in three sizes -- Nano (8B), Super (49B), and Ultra (253B) -- and performs competitively with state-of-the-art reasoning models such as DeepSeek-R1 while offering superior inference throughput and memory efficiency. In this report, we discuss the training procedure for these models, which entails using neural architecture search from Llama 3 models for accelerated inference, knowledge distillation, and continued pretraining, followed by a reasoning-focused post-training stage consisting of two main parts: supervised fine-tuning and large scale reinforcement learning. Llama-Nemotron models are the first open-source models to support a dynamic reasoning toggle, allowing users to switch between standard chat and reasoning modes during inference. To further support open research and facilitate model development, we provide the following resources: 1. We release the Llama-Nemotron reasoning models -- LN-Nano, LN-Super, and LN-Ultra -- under the commercially permissive NVIDIA Open Model License Agreement. 2. We release the complete post-training dataset: Llama-Nemotron-Post-Training-Dataset. 3. We also release our training codebases: NeMo, NeMo-Aligner, and Megatron-LM.

Large Language Models (LLMs) have shown remarkable capabilities across various domains, yet they struggle with knowledge-intensive tasks in areas that demand factual accuracy, e.g. industrial automation and healthcare. Key limitations include their tendency to hallucinate, lack of source traceability (provenance), and challenges in timely knowledge updates. Combining language models with knowledge graphs (GraphRAG) offers promising avenues for overcoming these deficits. However, a major challenge lies in creating such a knowledge graph in the first place. Here, we propose a novel approach that combines LLMs with a tripartite knowledge graph representation, which is constructed by connecting complex, domain-specific objects via a curated ontology of corresponding, domain-specific concepts to relevant sections within chunks of text through a concept-anchored pre-analysis of source documents starting from an initial lexical graph. Subsequently, we formulate LLM prompt creation as an unsupervised node classification problem allowing for the optimization of information density, coverage, and arrangement of LLM prompts at significantly reduced lengths. An initial experimental evaluation of our approach on a healthcare use case, involving multi-faceted analyses of patient anamneses given a set of medical concepts as well as a series of clinical guideline literature, indicates its potential to optimize information density, coverage, and arrangement of LLM prompts while significantly reducing their lengths, which, in turn, may lead to reduced costs as well as more consistent and reliable LLM outputs.

Ophthalmic diseases pose a significant global health burden. However, traditional diagnostic methods and existing monocular image-based deep learning approaches often overlook the pathological correlations between the two eyes. In practical medical robotic diagnostic scenarios, paired retinal images (binocular fundus images) are frequently required as diagnostic evidence. To address this, we propose DMS-Net-a dual-modal multi-scale siamese network for binocular retinal image classification. The framework employs a weight-sharing siamese ResNet-152 architecture to concurrently extract deep semantic features from bilateral fundus images. To tackle challenges like indistinct lesion boundaries and diffuse pathological distributions, we introduce the OmniPool Spatial Integrator Module (OSIM), which achieves multi-resolution feature aggregation through multi-scale adaptive pooling and spatial attention mechanisms. Furthermore, the Calibrated Analogous Semantic Fusion Module (CASFM) leverages spatial-semantic recalibration and bidirectional attention mechanisms to enhance cross-modal interaction, aggregating modality-agnostic representations of fundus structures. To fully exploit the differential semantic information of lesions present in bilateral fundus features, we introduce the Cross-Modal Contrastive Alignment Module (CCAM). Additionally, to enhance the aggregation of lesion-correlated semantic information, we introduce the Cross-Modal Integrative Alignment Module (CIAM). Evaluation on the ODIR-5K dataset demonstrates that DMS-Net achieves state-of-the-art performance with an accuracy of 82.9%, recall of 84.5%, and a Cohen's kappa coefficient of 83.2%, showcasing robust capacity in detecting symmetrical pathologies and improving clinical decision-making for ocular diseases. Code and the processed dataset will be released subsequently.

Effective reasoning is crucial to solving complex mathematical problems. Recent large language models (LLMs) have boosted performance by scaling test-time computation through long chain-of-thought reasoning. However, transformer-based models are inherently limited in extending context length due to their quadratic computational complexity and linear memory requirements. In this paper, we introduce a novel hybrid linear RNN reasoning model, M1, built on the Mamba architecture, which allows memory-efficient inference. Our approach leverages a distillation process from existing reasoning models and is further enhanced through RL training. Experimental results on the AIME and MATH benchmarks show that M1 not only outperforms previous linear RNN models but also matches the performance of state-of-the-art Deepseek R1 distilled reasoning models at a similar scale. We also compare our generation speed with a highly performant general purpose inference engine, vLLM, and observe more than a 3x speedup compared to a same size transformer. With throughput speedup, we are able to achieve higher accuracy compared to DeepSeek R1 distilled transformer reasoning models under a fixed generation time budget using self-consistency voting. Overall, we introduce a hybrid Mamba reasoning model and provide a more effective approach to scaling test-time generation using self-consistency or long chain of thought reasoning.