Decoder-only large language model (LLM)-based embedding models are beginning to outperform BERT or T5-based embedding models in general-purpose text embedding tasks, including dense vector-based retrieval. In this work, we introduce the NV-Embed model, incorporating architectural designs, training procedures, and curated datasets to significantly enhance the performance of LLM as a versatile embedding model, while maintaining its simplicity and reproducibility.For model architecture, we propose a latent attention layer to obtain pooled embeddings, which consistently improves retrieval and downstream task accuracy compared to mean pooling or using the last <EOS> token embedding from LLMs. To enhance representation learning, we remove the causal attention mask of LLMs during contrastive training. For training algorithm, we introduce a two-stage contrastive instruction-tuning method. It first applies contrastive training with instructions on retrieval datasets, utilizing in-batch negatives and curated hard negative examples. At stage-2, it blends various non-retrieval into instruction tuning, which not only enhances non-retrieval task accuracy but also improves retrieval performance. For training data, we utilize the hard-negative mining, synthetic data generation and existing public available datasets to boost the performance of embedding model. By combining these techniques, our NV-Embed- v1 model secured the No.1 position on the Massive Text Embedding Benchmark (MTEB) (as of May 24, 2024), across 56 embedding tasks. NV-Embed-v2 has reclaimed and maintained the top spot on MTEB since August 30, 2024, demonstrating the sustained effectiveness of the proposed methods over time. Also, it achieved the highest scores in the Long Doc section and the second-highest scores in the QA section of the AIR Benchmark, which covers a range of out-of-domain information retrieval topics beyond those in MTEB. We further provide the analysis of model compression techniques for generalist embedding models. We open-source the model at: https://huggingface.co/nvidia/NV-Embed-v2 .

State-of-the-art retrieval models typically address a straightforward search scenario, in which retrieval tasks are fixed (e.g., finding a passage to answer a specific question) and only a single modality is supported for both queries and retrieved results. This paper introduces techniques for advancing information retrieval with multimodal large language models (MLLMs), enabling a broader search scenario, termed universal multimodal retrieval, where multiple modalities and diverse retrieval tasks are accommodated. To this end, we first study fine-tuning an MLLM as a bi-encoder retriever on 10 datasets with 16 retrieval tasks. Our empirical results show that the fine-tuned MLLM retriever is capable of understanding challenging queries, composed of both text and image, but it underperforms compared to a smaller CLIP retriever in cross-modal retrieval tasks due to the modality bias exhibited by MLLMs. To address the issue, we propose modality-aware hard negative mining to mitigate the modality bias exhibited by MLLM retrievers. Second, we propose continuously fine-tuning the universal multimodal retriever to enhance its text retrieval capability while preserving multimodal retrieval capability. As a result, our model, MM-Embed, achieves state-of-the-art performance on the multimodal retrieval benchmark M-BEIR, which spans multiple domains and tasks, while also surpassing the state-of-the-art text retrieval model, NV-Embed-v1, on the MTEB retrieval benchmark. Finally, we explore prompting the off-the-shelf MLLMs as zero-shot rerankers to refine the ranking of the candidates from the multimodal retriever. We find that, through prompt-and-reranking, MLLMs can further improve multimodal retrieval when the user queries (e.g., text-image composed queries) are more complex and challenging to understand. These findings also pave the way for advancing universal multimodal retrieval in the future. We release the model weights at: https://huggingface.co/nvidia/MM-Embed.

Large Language Models (LLMs) have demonstrated significant potential in code generation by following natural language instructions. Unfortunately, crucial real-world software engineering tasks, such as debugging or repository-level feature implementation, involve processing extensive contexts beyond current LLM context sizes and performing complex reasoning that is brittle using standard autoregressive decoding. Enhancing LLMs’ performance in these scenarios requires careful consideration of the contextual information provided to the model, optimizing how the model leverages that, and identifying tools that enable more effective navigation of the development environment. To address these challenges, we introduce Nemotron-CORTEXA, an agentic system built on a predefined scaffold that enhances LLMs’ ability to navigate and reason efficiently in complex software engineering contexts. Specifically, we develop a novel code embedding model that retrieves the most relevant files with greater precision, along with a localization agent that refines the granularity of the retrieval process. Additionally, we demonstrate that providing diverse contextual information and utilizing different prompt formats enable the model to identify and resolve issues more efficiently. We evaluate Nemotron-CORTEXA using SWE-bench, a benchmark derived from real-world GitHub issues. Compared to the widely used Agentless framework, Nemotron-CORTEXA achieves a higher issue resolution rate at a lower cost, highlighting its practical impact in addressing real-world software engineering challenges.

Despite recent progress in large-scale reinforcement learning (RL) for reasoning, the training recipe for building high-performing reasoning models remains elusive. Key implementation details of frontier models, such as DeepSeek-R1, including data curation strategies and RL training recipe, are often omitted. Moreover, recent research indicates distillation remains more effective than RL for smaller models. In this work, we demonstrate that large-scale RL can significantly enhance the reasoning capabilities of strong, small- and mid-sized models, achieving results that surpass those of state-of-the-art distillation-based models. We systematically study the RL training process through extensive ablations and propose a simple yet effective approach: first training on math-only prompts, then on code-only prompts. Notably, we find that math-only RL not only significantly enhances the performance of strong distilled models on math benchmarks (e.g., +14.6\% / +17.2\% on AIME 2025 for the 7B / 14B models), but also code reasoning tasks (e.g., +6.8\% / +5.8\% on LiveCodeBench for the 7B / 14B models). In addition, extended code-only RL further improves code benchmark performance while causing minimal degradation in math results. We develop a robust data curation pipeline to collect challenging prompts with high-quality, verifiable answers and test cases to enable verification-based RL across both domains. The dataset will be released to support open research. Finally, we identify key experimental insights, including curriculum learning with progressively increasing response lengths and the stabilizing effect of on-policy parameter updates. We find that RL not only elicits the foundational reasoning capabilities acquired during pretraining and supervised fine-tuning (SFT), but also pushes the limits of the model’s reasoning ability, enabling it to solve problems that were previously unsolvable.

In this work, we introduce ChatQA, a suite of models that outperform GPT-4 on retrieval-augmented generation (RAG) and conversational question answering (QA). To enhance generation, we propose a two-stage instruction tuning method that significantly boosts the performance of RAG. For effective retrieval, we introduce a dense retriever optimized for conversational QA, which yields results comparable to the alternative state-of-the-art query rewriting models, while substantially reducing deployment costs. We also present the ChatRAG Bench, which encompasses ten datasets covering comprehensive evaluations on RAG, table-related QA, arithmetic calculations, and scenarios involving unanswerable questions. Our ChatQA-1.0-70B (score: 54.14), built on Llama2, a weaker foundation model than GPT-4, can slightly outperform GPT-4-0613 (score: 53.90) and GPT-4-Turbo-2024-04-09 (score: 54.03) on the ChatRAG Bench, without relying on any synthetic data from OpenAI GPT models. Notably, Llama3-ChatQA-1.5-70B model surpasses the accuracy of GPT-4-Turbo-2024-04-09 by a margin. These results demonstrate the exceptional quality of the proposed ChatQA recipe. To advance research in this field, we open-sourced the model weights, instruction tuning data, ChatRAG Bench, and retriever for the community.

Standard frame-based cameras that sample light intensity frames are heavily impacted by motion blur for high-speed motion and fail to perceive scene accurately in high-dynamic range environments. Event-based cameras, on the other hand, overcome these limitations by asynchronously detecting the variation in individual pixel intensities. However, event cameras only capture pixels in motion, leading to sparse information. Hence, estimating the overall dense behavior of pixels is difficult. To address aforementioned issues associated with both sensors, we present Fusion-FlowNet, a sensor fusion framework for energy -efficient optical flow estimation. Fusion-FlowNet utilizes both frame- and event-based sensors, leveraging their complementary characteristics. Our proposed network architecture is also a fusion of Spiking Neural Net-works (SNNs) and Analog Neural Networks (ANNs) where each network is designed to simultaneously process asynchronous event streams and regular frame-based images, respectively. We perform end-to-end training using unsupervised learning to avoid expensive video annotations. Our method generalizes well across distinct environments (rapid motion and challenging lighting conditions) and demonstrates state-of-the-art optical flow prediction on the Multi-Vehicle Stereo Event Camera (MVSEC) dataset. Furthermore, the usage of SNNs in our architecture offers substantial savings in terms of the number of network parameters and computational energy cost.

Spiking neural networks (SNNs) operating with sparse binary signals (spikes) implemented on event-driven hardware can potentially be more energy -efficient than traditional artificial neural networks (ANNs). However, SNNs perform computations over time, and the neuron activation function does not have a well-defined derivative leading to unique training challenges. In this paper, we discuss the various spike representations and training mechanisms for deep SNN s. Additionally, we review applications that go beyond classification, like gesture recognition, motion estimation, and sequential learning. The unique features of SNNs, such as high activation sparsity and spike-based computations, can be leveraged in hardware implementations for energy-efficient processing. To that effect, we discuss various SNN implementations, both using digital ASICs as well as analog in-memory computing primitives. Finally, we present an outlook on future applications and open research areas for both SNN algorithms and hardware implementations.

Optical flow can be leveraged in robotic systems for obstacle detection where low latency solutions are critical in highly dynamic settings. While event-based cameras have changed the dominant paradigm of sending by encoding stimuli into spike trails, offering low bandwidth and latency, events are still processed with traditional convolutional networks in GPUs defeating, thus, the promise of efficient low capacity low power processing that inspired the design of event sensors. In this work, we introduce a shallow spiking neural network for the computation of optical flow consisting of Leaky Integrate and Fire neurons.Optical flow is predicted as the synthesis of motion orientation selective channels. Learning is accomplished by Back-propapagation Through Time. We present promising results on events recorded in real "in the wild" scenes that has the capability to use only a small fraction of the energy consumed in CNNs deployed on GPUs.

Spiking Neural Networks (SNNs), widely known as the third generation of neural networks, encode input information temporally using sparse spiking events, which can be harnessed to achieve higher computational efficiency for cognitive tasks. However, considering the rapid strides in accuracy enabled by state-of-the-art Analog Neural Networks (ANNs), SNN training algorithms are much less mature, leading to accuracy gap between SNNs and ANNs. In this paper, we propose different SNN training methodologies, varying in degrees of biofidelity, and evaluate their efficacy on complex image recognition datasets. First, we present biologically plausible Spike Timing Dependent Plasticity (STDP) based deterministic and stochastic algorithms for unsupervised representation learning in SNNs. Our analysis on the CIFAR-10 dataset indicates that STDP-based learning rules enable the convolutional layers to self-learn low-level input features using fewer training examples. However, STDP-based learning is limited in applicability to shallow SNNs (≤4 layers) while yielding considerably lower than state-of-the-art accuracy. In order to scale the SNNs deeper and improve the accuracy further, we propose conversion methodology to map off-the-shelf trained ANN to SNN for energy-efficient inference. We demonstrate 69.96% accuracy for VGG16-SNN on ImageNet. However, ANN-to-SNN conversion leads to high inference latency for achieving the best accuracy. In order to minimize the inference latency, we propose spike-based error backpropagation algorithm using differentiable approximation for the spiking neuron. Our preliminary experiments on CIFAR-10 show that spike-based error backpropagation effectively captures temporal statistics to reduce the inference latency by up to 8× compared to converted SNNs while yielding comparable accuracy

In this era of machine learning models, their functionality is being threatened by adversarial attacks. In the face of this struggle for making artificial neural networks robust, finding a model, resilient to these attacks, is very important. In this work, we present, for the first time, a comprehensive analysis of the behavior of more bio-plausible networks, namely Spiking Neural Network (SNN) under state-of-the-art adversarial tests. We perform a comparative study of the accuracy degradation between conventional VGG-9 Artificial Neural Network (ANN) and equivalent spiking network with CIFAR-10 dataset in both whitebox and blackbox setting for different types of single-step and multi-step FGSM (Fast Gradient Sign Method) attacks. We demonstrate that SNNs tend to show more resiliency compared to ANN under blackbox attack scenario. Additionally, we find that SNN robustness is largely dependent on the corresponding training mechanism. We observe that SNNs trained by spike-based backpropagation are more adversarially robust than the ones obtained by ANN-to-SNN conversion rules in several whitebox and blackbox scenarios. Finally, we also propose a simple, yet, effective framework for crafting adversarial attacks from SNNs. Our results suggest that attacks crafted from SNNs following our proposed method are much stronger than those crafted from ANNs.

This paper proposes a novel algorithm to identify three inertial parameters: sprung mass, yaw moment of inertia, and longitudinal position of the center of gravity. A four-wheel nonlinear vehicle model with roll dynamics and a correlation between the inertial parameters is used for a dual unscented Kalman filter to simultaneously identify the inertial parameters and the vehicle state. A local observability analysis on the nonlinear vehicle model is used to activate and deactivate different modes of the proposed algorithm. Extensive CarSim simulations and experimental tests show the performance and robustness of the proposed approach on a flat road with a constant tire–road friction coefficient.