Siddharth Singh

Aarti Basant, Abhijit Khairnar, Abhijit Paithankar et al. · nvidia

We introduce Nemotron-Nano-9B-v2, a hybrid Mamba-Transformer language model designed to increase throughput for reasoning workloads while achieving state-of-the-art accuracy compared to similarly-sized models. Nemotron-Nano-9B-v2 builds on the Nemotron-H architecture, in which the majority of the self-attention layers in the common Transformer architecture are replaced with Mamba-2 layers, to achieve improved inference speed when generating the long thinking traces needed for reasoning. We create Nemotron-Nano-9B-v2 by first pre-training a 12-billion-parameter model (Nemotron-Nano-12B-v2-Base) on 20 trillion tokens using an FP8 training recipe. After aligning Nemotron-Nano-12B-v2-Base, we employ the Minitron strategy to compress and distill the model with the goal of enabling inference on up to 128k tokens on a single NVIDIA A10G GPU (22GiB of memory, bfloat16 precision). Compared to existing similarly-sized models (e.g., Qwen3-8B), we show that Nemotron-Nano-9B-v2 achieves on-par or better accuracy on reasoning benchmarks while achieving up to 6x higher inference throughput in reasoning settings like 8k input and 16k output tokens. We are releasing Nemotron-Nano-9B-v2, Nemotron-Nano12B-v2-Base, and Nemotron-Nano-9B-v2-Base checkpoints along with the majority of our pre- and post-training datasets on Hugging Face.

Aaron Blakeman, Aaron Grattafiori, Aarti Basant et al. · nvidia

We present Nemotron 3 Nano 30B-A3B, a Mixture-of-Experts hybrid Mamba-Transformer language model. Nemotron 3 Nano was pretrained on 25 trillion text tokens, including more than 3 trillion new unique tokens over Nemotron 2, followed by supervised fine tuning and large-scale RL on diverse environments. Nemotron 3 Nano achieves better accuracy than our previous generation Nemotron 2 Nano while activating less than half of the parameters per forward pass. It achieves up to 3.3x higher inference throughput than similarly-sized open models like GPT-OSS-20B and Qwen3-30B-A3B-Thinking-2507, while also being more accurate on popular benchmarks. Nemotron 3 Nano demonstrates enhanced agentic, reasoning, and chat abilities and supports context lengths up to 1M tokens. We release both our pretrained Nemotron 3 Nano 30B-A3B Base and post-trained Nemotron 3 Nano 30B-A3B checkpoints on Hugging Face.

Aaron Blakeman, Aaron Grattafiori, Aarti Basant et al. · nvidia

We introduce the Nemotron 3 family of models - Nano, Super, and Ultra. These models deliver strong agentic, reasoning, and conversational capabilities. The Nemotron 3 family uses a Mixture-of-Experts hybrid Mamba-Transformer architecture to provide best-in-class throughput and context lengths of up to 1M tokens. Super and Ultra models are trained with NVFP4 and incorporate LatentMoE, a novel approach that improves model quality. The two larger models also include MTP layers for faster text generation. All Nemotron 3 models are post-trained using multi-environment reinforcement learning enabling reasoning, multi-step tool use, and support granular reasoning budget control. Nano, the smallest model, outperforms comparable models in accuracy while remaining extremely cost-efficient for inference. Super is optimized for collaborative agents and high-volume workloads such as IT ticket automation. Ultra, the largest model, provides state-of-the-art accuracy and reasoning performance. Nano is released together with its technical report and this white paper, while Super and Ultra will follow in the coming months. We will openly release the model weights, pre- and post-training software, recipes, and all data for which we hold redistribution rights.

Siddharth Singh, Olatunji Ruwase, Ammar Ahmad Awan et al.

Mixture-of-Experts (MoE) is a neural network architecture that adds sparsely activated expert blocks to a base model, increasing the number of parameters without impacting computational costs. However, current distributed deep learning frameworks are limited in their ability to train high-quality MoE models with large base models. In this work, we present DeepSpeed-TED, a novel, three-dimensional, hybrid parallel algorithm that combines data, tensor, and expert parallelism to enable the training of MoE models with 4 to 8x larger base models than the current state-of-the-art. We also describe memory optimizations in the optimizer step, and communication optimizations that eliminate unnecessary data movement. We implement our approach in DeepSpeed and achieve speedups of 26% over a baseline (i.e. without our communication optimizations) when training a 40 billion parameter MoE model (6.7 billion base model with 16 experts) on 128 V100 GPUs.

Siddharth Singh, Ritesh Kumar, Shyam Ratan et al.

In the proposed demo, we will present a new software - Linguistic Field Data Management and Analysis System - LiFE (https://github.com/kmi-linguistics/life) - an open-source, web-based linguistic data management and analysis application that allows for systematic storage, management, sharing and usage of linguistic data collected from the field. The application allows users to store lexical items, sentences, paragraphs, audio-visual content with rich glossing / annotation; generate interactive and print dictionaries; and also train and use natural language processing tools and models for various purposes using this data. Since its a web-based application, it also allows for seamless collaboration among multiple persons and sharing the data, models, etc with each other. The system uses the Python-based Flask framework and MongoDB in the backend and HTML, CSS and Javascript at the frontend. The interface allows creation of multiple projects that could be shared with the other users. At the backend, the application stores the data in RDF format so as to allow its release as Linked Data over the web using semantic web technologies - as of now it makes use of the OntoLex-Lemon for storing the lexical data and Ligt for storing the interlinear glossed text and then internally linking it to the other linked lexicons and databases such as DBpedia and WordNet. Furthermore it provides support for training the NLP systems using scikit-learn and HuggingFace Transformers libraries as well as make use of any model trained using these libraries - while the user interface itself provides limited options for tuning the system, an externally-trained model could be easily incorporated within the application; similarly the dataset itself could be easily exported into a standard machine-readable format like JSON or CSV that could be consumed by other programs and pipelines.

Siddharth Singh, Abhinav Bhatele

Parallel training of neural networks at scale is challenging due to significant overheads arising from communication. Recently, deep learning researchers have developed a variety of pruning algorithms that are capable of pruning (i.e. setting to zero) 80-90% of the parameters in a neural network to yield sparse subnetworks that equal the accuracy of the unpruned parent network. In this work, we propose a novel approach that exploits these sparse subnetworks to optimize the memory utilization and communication in two popular algorithms for parallel deep learning namely -- data and inter-layer parallelism. We integrate our approach into AxoNN, a highly scalable framework for parallel deep learning that relies on data and inter-layer parallelism, and demonstrate the reduction in communication time and memory utilization. On 512 NVIDIA V100 GPUs, our optimizations reduce the memory consumption of a 2.7 billion parameter model by 74%, and the total communication time by 40%, thus providing an overall speedup of 34% over AxoNN, 32% over DeepSpeed-3D and 46% over Sputnik, a sparse matrix computation baseline.

Ritesh Kumar, Siddharth Singh, Shyam Ratan et al.

In this paper we discuss an in-progress work on the development of a speech corpus for four low-resource Indo-Aryan languages -- Awadhi, Bhojpuri, Braj and Magahi using the field methods of linguistic data collection. The total size of the corpus currently stands at approximately 18 hours (approx. 4-5 hours each language) and it is transcribed and annotated with grammatical information such as part-of-speech tags, morphological features and Universal dependency relationships. We discuss our methodology for data collection in these languages, most of which was done in the middle of the COVID-19 pandemic, with one of the aims being to generate some additional income for low-income groups speaking these languages. In the paper, we also discuss the results of the baseline experiments for automatic speech recognition system in these languages.

Ifrah Idrees, Siddharth Singh, Kerui Xu et al.

For social robots like Astro which interact with and adapt to the daily movements of users within the home, realistic simulation of human activity is needed for feature development and testing. This paper presents a framework for simulating daily human activity patterns in home environments at scale, supporting manual configurability of different personas or activity patterns, variation of activity timings, and testing on multiple home layouts. We introduce a method for specifying day-to-day variation in schedules and present a bidirectional constraint propagation algorithm for generating schedules from templates. We validate the expressive power of our framework through a use case scenario analysis and demonstrate that our method can be used to generate data closely resembling human behavior from three public datasets and a self-collected dataset. Our contribution supports systematic testing of social robot behaviors at scale, enables procedural generation of synthetic datasets of human movement in different households, and can help minimize bias in training data, leading to more robust and effective robots for home environments.

Cunyang Wei, Siddharth Singh, Aishwarya Sarkar et al.

Graph neural networks (GNNs) are widely used for learning on graph datasets derived from various real-world scenarios. Learning from extremely large graphs requires distributed training, and mini-batching with sampling is a popular approach for parallelizing GNN training. Existing distributed mini-batch approaches have significant performance bottlenecks due to expensive sampling methods and limited scaling when using data parallelism. In this work, we present ScaleGNN, a 4D parallel framework for scalable mini-batch GNN training that combines communication-free distributed sampling, 3D parallel matrix multiplication (PMM), and data parallelism. ScaleGNN introduces a uniform vertex sampling algorithm, enabling each process (GPU device) to construct its local mini-batch, i.e., subgraph partitions without any inter-process communication. 3D PMM enables scaling mini-batch training to much larger GPU counts than vanilla data parallelism with significantly lower communication overheads. We also present additional optimizations to overlap sampling with training, reduce communication overhead by sending data in lower precision, kernel fusion, and communication-computation overlap. We evaluate ScaleGNN on five graph datasets and demonstrate strong scaling up to 2048 GPUs on Perlmutter, 2048 GCDs on Frontier, and 1024 GPUs on Tuolumne. On Perlmutter, ScaleGNN achieves 3.5x end-to-end training speedup over the SOTA baseline on ogbn-products.

Siddharth Singh, Zachary Sating, Abhinav Bhatele

Despite their better convergence properties compared to first-order optimizers, second-order optimizers for deep learning have been less popular due to their significant computational costs. The primary efficiency bottleneck in such optimizers is matrix inverse calculations in the preconditioning step, which are expensive to compute on GPUs. In this paper, we introduce Jorge, a second-order optimizer that promises the best of both worlds -- rapid convergence benefits of second-order methods, and high computational efficiency typical of first-order methods. We address the primary computational bottleneck of computing matrix inverses by completely eliminating them using an approximation of the preconditioner computation. This makes Jorge extremely efficient on GPUs in terms of wall-clock time. Further, we describe an approach to determine Jorge's hyperparameters directly from a well-tuned SGD baseline, thereby significantly minimizing tuning efforts. Our empirical evaluations demonstrate the distinct advantages of using Jorge, outperforming state-of-the-art optimizers such as SGD, AdamW, and Shampoo across multiple deep learning models, both in terms of sample efficiency and wall-clock time.

Prajwal Singhania, Siddharth Singh, Lannie Dalton Hough et al.

As large language models (LLMs) continue to grow in size, distributed inference has become increasingly important. Model-parallel strategies must now efficiently scale not only across multiple GPUs but also across multiple nodes. In this work, we present a detailed performance study of multi-node distributed inference using LLMs on GPU-based supercomputers. We conduct experiments with several state-of-the-art inference engines alongside YALIS, a research-oriented prototype engine designed for controlled experimentation. We analyze the strong-scaling behavior of different model-parallel schemes and identify key bottlenecks. Since all-reduce operations are a common performance bottleneck, we develop NVRAR, a hierarchical all-reduce algorithm based on recursive doubling with NVSHMEM. NVRAR achieves up to 1.9x-3.6x lower latency than NCCL for message sizes between 128 KB and 2 MB on HPE Slingshot and InfiniBand interconnects. Integrated into YALIS, NVRAR achieves up to a 1.72x reduction in end-to-end batch latency for the Llama 3.1 405B model in multi-node decode-heavy workloads using tensor parallelism.

Rahul Anand, Siddharth Singh, Dileep A D et al.

Vision foundation models have demonstrated strong transferability across diverse visual recognition tasks and are increasingly considered for biometric applications. Their suitability for iris Presentation Attack Detection (PAD), particularly under realistic open-set operating conditions, remains insufficiently examined. This work presents a systematic failure analysis of general-purpose vision foundation models for open-set iris PAD using periocular imagery. Five representative foundation models are evaluated under three open-set protocols that explicitly separate different sources of distribution shift: unseen Presentation Attack Instruments (PAIs), unseen datasets captured with different sensors and cross-spectral transfer from near-infrared (NIR) to visible spectrum (VIS) imagery. Both frozen feature representations and parameter-efficient task adaptation using Low-Rank Adaptation (LoRA) are assessed within a unified experimental framework. The results indicate that foundation models can transfer across datasets with similar sensing characteristics, but fail to generalise reliably to unseen attack instruments and degrade sharply under cross-spectral evaluation. While LoRA improves performance in certain cross-dataset settings, it frequently amplifies failure under attack-level and spectral shifts. Additional validation experiments using segmented iris inputs, full backbone fine-tuning, joint cross-dataset and cross-PAI shifts, and reverse VIS to NIR transfer further confirm that these failures are not simply artefacts of periocular input, weak adaptation, or one-directional spectral evaluation. These findings show that strong closed-set or cross-dataset performance should not be treated as evidence of robust open-set security, and highlight the need for PAD representations that maintain sensitivity to presentation artefacts while remaining stable under realistic deployment variation.

Sean McLeish, John Kirchenbauer, David Yu Miller et al.

Scaling laws are typically fit using a family of models with a narrow range of frozen hyperparameter choices. In this work we study scaling laws using multiple architectural shapes and hyperparameter choices, highlighting their impact on resulting prescriptions. As a primary artifact of our research, we release the Gemstones: an open-source scaling law dataset, consisting of over 4000 checkpoints from transformers with up to 2 billion parameters and diverse architectural shapes; including ablations over learning rate and cooldown. Our checkpoints enable more complex studies of scaling, such as analyzing the relationship between width and depth. By examining our model suite, we find that the prescriptions of scaling laws can be highly sensitive to the experimental design process and the specific model checkpoints used during fitting.

Siddharth Singh, Prajwal Singhania, Aditya Ranjan et al.

Training and fine-tuning large language models (LLMs) with hundreds of billions to trillions of parameters requires tens of thousands of GPUs, and a highly scalable software stack. In this work, we present a novel four-dimensional hybrid parallel algorithm implemented in a highly scalable, portable, open-source framework called AxoNN. We describe several performance optimizations in AxoNN to improve matrix multiply kernel performance, overlap non-blocking collectives with computation, and performance modeling to choose performance optimal configurations. These have resulted in unprecedented scaling and peak flop/s (bf16) for training of GPT-style transformer models on Perlmutter (620.1 Petaflop/s), Frontier (1.381 Exaflop/s) and Alps (1.423 Exaflop/s). While the abilities of LLMs improve with the number of trainable parameters, so do privacy and copyright risks caused by memorization of training data, which can cause disclosure of sensitive or private information at inference time. We highlight this side effect of scale through experiments that explore "catastrophic memorization", where models are sufficiently large to memorize training data in a single pass, and present an approach to prevent it. As part of this study, we demonstrate fine-tuning of a 405-billion parameter LLM using AxoNN on Frontier.

Aman Chaturvedi, Daniel Nichols, Siddharth Singh et al.

Large Language Model (LLM) based coding tools have been tremendously successful as software development assistants, yet they are often designed for general purpose programming tasks and perform poorly for more specialized domains such as high performance computing. Creating specialized models and tools for these domains is crucial towards gaining the benefits of LLMs in areas such as HPC. While previous work has explored HPC-specific models, LLMs still struggle to generate parallel code and it is not at all clear what hurdles are still holding back these LLMs and what must be done to overcome them. In this work, we conduct an in-depth study along the many axes of fine-tuning a specialized HPC LLM in order to better understand the challenges. Based on our findings we fine-tune and evaluate a specialized HPC LLM that is shown to be the best performing open-source code LLM for parallel code generation to date.

Siddharth Singh, Soumee Guha, Qing Chang et al.

Learning based multi-robot path planning methods struggle to scale or generalize to changes, particularly variations in the number of robots during deployment. Most existing methods are trained on a fixed number of robots and may tolerate a reduced number during testing, but typically fail when the number increases. Additionally, training such methods for a larger number of agents can be both time consuming and computationally expensive. However, analytical methods can struggle to scale computationally or handle dynamic changes in the environment. In this work, we propose to leverage a diffusion model based planner capable of handling dynamically varying number of agents. Our approach is trained on a limited number of agents and generalizes effectively to larger numbers of agents during deployment. Results show that integrating a single shared diffusion model based planner with dedicated inter-agent attention computation and temporal convolution enables a train small deploy-large paradigm with good accuracy. We validate our method across multiple scenarios and compare the performance with existing multi-agent reinforcement learning techniques and heuristic control based methods.

Jonas Geiping, Sean McLeish, Neel Jain et al.

We study a novel language model architecture that is capable of scaling test-time computation by implicitly reasoning in latent space. Our model works by iterating a recurrent block, thereby unrolling to arbitrary depth at test-time. This stands in contrast to mainstream reasoning models that scale up compute by producing more tokens. Unlike approaches based on chain-of-thought, our approach does not require any specialized training data, can work with small context windows, and can capture types of reasoning that are not easily represented in words. We scale a proof-of-concept model to 3.5 billion parameters and 800 billion tokens. We show that the resulting model can improve its performance on reasoning benchmarks, sometimes dramatically, up to a computation load equivalent to 50 billion parameters.

Siddharth Singh, Ifrah Idrees, Abraham Dauhajre

Advancements in foundation models have catalyzed research in Embodied AI to develop interactive agents capable of environmental reasoning and interaction. Developing such agents requires diverse, large-scale datasets. Prior frameworks generate synthetic data for long-term human-robot interactions but fail to model the bidirectional influence between human behavior and household environments. Our proposed generative framework creates household datasets at scale through loosely coupled generation of long-term human-robot interactions and environments. Human personas influence environment generation, while environment schematics and semantics shape human-robot interactions. The generated 3D data includes rich static context such as object and environment semantics, and temporal context capturing human and agent behaviors over extended periods. Our flexible tool allows users to define dataset characteristics via natural language prompts, enabling configuration of environment and human activity data through natural language specifications. The tool creates variations of user-defined configurations, enabling scalable data generation. We validate our framework through statistical evaluation using multi-modal embeddings and key metrics: cosine similarity, mutual information gain, intervention analysis, and iterative improvement validation. Statistical comparisons show good alignment with real-world datasets (HOMER) with cosine similarity (0.60), while synthetic datasets (Wang et al.) show moderate alignment (0.27). Intervention analysis across age, organization, and sleep pattern changes shows statistically significant effects (p < 0.001) with large effect sizes (Cohen's d = 0.51-1.12), confirming bidirectional coupling translates persona traits into measurable environmental and behavioral differences. These contributions enable development and testing of household smart devices at scale.

Shalmoli Ghosh, Prajwal Singhania, Siddharth Singh et al.

Online forums and social media platforms are increasingly being used to discuss topics of varying polarities where different people take different stances. Several methodologies for automatic stance detection from text have been proposed in literature. To our knowledge, there has not been any systematic investigation towards their reproducibility, and their comparative performances. In this work, we explore the reproducibility of several existing stance detection models, including both neural models and classical classifier-based models. Through experiments on two datasets -- (i)~the popular SemEval microblog dataset, and (ii)~a set of health-related online news articles -- we also perform a detailed comparative analysis of various methods and explore their shortcomings. Implementations of all algorithms discussed in this paper are available at https://github.com/prajwal1210/Stance-Detection-in-Web-and-Social-Media.

Ritesh Kumar, Ojaswee Bhalla, Madhu Vanthi et al.

In this paper, we discuss the development of an annotation schema to build datasets for evaluating the offline harm potential of social media texts. We define "harm potential" as the potential for an online public post to cause real-world physical harm (i.e., violence). Understanding that real-world violence is often spurred by a web of triggers, often combining several online tactics and pre-existing intersectional fissures in the social milieu, to result in targeted physical violence, we do not focus on any single divisive aspect (i.e., caste, gender, religion, or other identities of the victim and perpetrators) nor do we focus on just hate speech or mis/dis-information. Rather, our understanding of the intersectional causes of such triggers focuses our attempt at measuring the harm potential of online content, irrespective of whether it is hateful or not. In this paper, we discuss the development of a framework/annotation schema that allows annotating the data with different aspects of the text including its socio-political grounding and intent of the speaker (as expressed through mood and modality) that together contribute to it being a trigger for offline harm. We also give a comparative analysis and mapping of our framework with some of the existing frameworks.

Siddharth Singh, Mahua Singh, Abhinav Bhatele

We evaluate the current state of collective communication on GPU-based supercomputers for large language model (LLM) training at scale. Existing libraries such as RCCL and Cray-MPICH exhibit critical limitations on systems such as Frontier -- Cray-MPICH underutilizes network and compute resources, while RCCL suffers from severe scalability issues. To address these challenges, we introduce PCCL, a communication library with highly optimized implementations of all-gather and reduce-scatter operations tailored for distributed deep learning workloads. PCCL is designed to maximally utilize all available network and compute resources and to scale efficiently to thousands of GPUs. It achieves substantial performance improvements, delivering 6-33x speedups over RCCL and 28-70x over Cray-MPICH for all-gather on 2048 GCDs of Frontier. These gains translate directly to end-to-end performance: in large-scale GPT-3-style training, PCCL provides up to 60% and 40% speedups over RCCL for 7B and 13B parameter models, respectively.

Aditya K. Ranjan, Siddharth Singh, Cunyang Wei et al.

Graph neural networks (GNNs) leverage the connectivity and structure of real-world graphs to learn intricate properties and relationships between nodes. Many real-world graphs exceed the memory capacity of a GPU due to their sheer size, and training GNNs on such graphs requires techniques such as mini-batch sampling to scale. The alternative approach of distributed full-graph training suffers from high communication overheads and load imbalance due to the irregular structure of graphs. We propose a three-dimensional (3D) parallel approach for full-graph training that tackles these issues and scales to billion-edge graphs. In addition, we introduce optimizations such as a double permutation scheme for load balancing, and a performance model to predict the optimal 3D configuration of our parallel implementation -- Plexus. We evaluate Plexus on six different graph datasets and show scaling results on up to 2048 GPUs of Perlmutter, and 1024 GPUs of Frontier. Plexus achieves unprecedented speedups of 2.3-12.5x over prior state of the art, and a reduction in time-to-solution by 5.2-8.7x on Perlmutter and 7.0-54.2x on Frontier.

Chujie Zheng, Jeffrey Wang, Shuqian Albee Zhang et al.

We propose a novel method for evaluating the performance of a content search system that measures the semantic match between a query and the results returned by the search system. We introduce a metric called "on-topic rate" to measure the percentage of results that are relevant to the query. To achieve this, we design a pipeline that defines a golden query set, retrieves the top K results for each query, and sends calls to GPT 3.5 with formulated prompts. Our semantic evaluation pipeline helps identify common failure patterns and goals against the metric for relevance improvements.

Abhimanyu Hans, Yuxin Wen, Neel Jain et al.

Large language models can memorize and repeat their training data, causing privacy and copyright risks. To mitigate memorization, we introduce a subtle modification to the next-token training objective that we call the goldfish loss. During training, randomly sampled subsets of tokens are excluded from the loss computation. These dropped tokens are not memorized by the model, which prevents verbatim reproduction of a complete chain of tokens from the training set. We run extensive experiments training billion-scale Llama-2 models, both pre-trained and trained from scratch, and demonstrate significant reductions in extractable memorization with little to no impact on downstream benchmarks.

Prajwal Singhania, Siddharth Singh, Shwai He et al.

Inference on large language models (LLMs) can be expensive in terms of the compute and memory costs involved, especially when long sequence lengths are used. In particular, the self-attention mechanism used in LLM inference contributes significantly to these costs, which has sparked an interest in approximating the self-attention computation to reduce such costs. In this work, we propose to approximate self-attention by focusing on the dimensionality of key vectors computed in the attention block. Our analysis reveals that key vectors lie in a significantly lower-dimensional space, consistently across several datasets and models. Exploiting this observation, we propose Loki, a novel sparse attention method that ranks and selects tokens in the KV-cache based on attention scores computed in low-dimensional space. Our evaluations show that Loki is able to speed up the attention computation due to reduced data movement (load/store) and compute costs while maintaining the efficacy of the models better than other popular approximation methods.

Siddharth Singh, Prajwal Singhania, Aditya K. Ranjan et al.

Heavy communication, in particular, collective operations, can become a critical performance bottleneck in scaling the training of billion-parameter neural networks to large-scale parallel systems. This paper introduces a four-dimensional (4D) approach to optimize communication in parallel training. This 4D approach is a hybrid of 3D tensor and data parallelism, and is implemented in the AxoNN framework. In addition, we employ two key strategies to further minimize communication overheads. First, we aggressively overlap expensive collective operations (reduce-scatter, all-gather, and all-reduce) with computation. Second, we develop an analytical model to identify high-performing configurations within the large search space defined by our 4D algorithm. This model empowers practitioners by simplifying the tuning process for their specific training workloads. When training an 80-billion parameter GPT on 1024 GPUs of Perlmutter, AxoNN surpasses Megatron-LM, a state-of-the-art framework, by a significant 26%. Additionally, it achieves a significantly high 57% of the theoretical peak FLOP/s or 182 PFLOP/s in total.

Ritesh Kumar, Enakshi Nandi, Laishram Niranjana Devi et al.

In this paper, we discuss the development of a multilingual dataset annotated with a hierarchical, fine-grained tagset marking different types of aggression and the "context" in which they occur. The context, here, is defined by the conversational thread in which a specific comment occurs and also the "type" of discursive role that the comment is performing with respect to the previous comment. The initial dataset, being discussed here (and made available as part of the ComMA@ICON shared task), consists of a total 15,000 annotated comments in four languages - Meitei, Bangla, Hindi, and Indian English - collected from various social media platforms such as YouTube, Facebook, Twitter and Telegram. As is usual on social media websites, a large number of these comments are multilingual, mostly code-mixed with English. The paper gives a detailed description of the tagset being used for annotation and also the process of developing a multi-label, fine-grained tagset that can be used for marking comments with aggression and bias of various kinds including gender bias, religious intolerance (called communal bias in the tagset), class/caste bias and ethnic/racial bias. We also define and discuss the tags that have been used for marking different the discursive role being performed through the comments, such as attack, defend, etc. We also present a statistical analysis of the dataset as well as results of our baseline experiments with developing an automatic aggression identification system using the dataset developed.

Daniel Nichols, Siddharth Singh, Shu-Huai Lin et al.

The field of deep learning has witnessed a remarkable shift towards extremely compute- and memory-intensive neural networks. These newer larger models have enabled researchers to advance state-of-the-art tools across a variety of fields. This phenomenon has spurred the development of algorithms for distributed training of neural networks over a larger number of hardware accelerators. In this paper, we discuss and compare current state-of-the-art frameworks for large scale distributed deep learning. First, we survey current practices in distributed learning and identify the different types of parallelism used. Then, we present empirical results comparing their performance on large image and language training tasks. Additionally, we address their statistical efficiency and memory consumption behavior. Based on our results, we discuss algorithmic and implementation portions of each framework which hinder performance.

Siddharth Singh, Abhinav Bhatele

In the last few years, the memory requirements to train state-of-the-art neural networks have far exceeded the DRAM capacities of modern hardware accelerators. This has necessitated the development of efficient algorithms to train these neural networks in parallel on large-scale GPU-based clusters. Since computation is relatively inexpensive on modern GPUs, designing and implementing extremely efficient communication in these parallel training algorithms is critical for extracting the maximum performance. This paper presents AxoNN, a parallel deep learning framework that exploits asynchrony and message-driven execution to schedule neural network operations on each GPU, thereby reducing GPU idle time and maximizing hardware efficiency. By using the CPU memory as a scratch space for offloading data periodically during training, AxoNN is able to reduce GPU memory consumption by four times. This allows us to increase the number of parameters per GPU by four times, thus reducing the amount of communication and increasing performance by over 13%. When tested against large transformer models with 12-100 billion parameters on 48-384 NVIDIA Tesla V100 GPUs, AxoNN achieves a per-GPU throughput of 49.4-54.78% of theoretical peak and reduces the training time by 22-37 days (15-25% speedup) as compared to the state-of-the-art.

Brandon Leung, Siddharth Singh, Arik Horodniceanu

Deep learning-based object reconstruction algorithms have shown remarkable improvements over classical methods. However, supervised learning based methods perform poorly when the training data and the test data have different distributions. Indeed, most current works perform satisfactorily on the synthetic ShapeNet dataset, but dramatically fail in when presented with real world images. To address this issue, unsupervised domain adaptation can be used transfer knowledge from the labeled synthetic source domain and learn a classifier for the unlabeled real target domain. To tackle this challenge of single view 3D reconstruction in the real domain, we experiment with a variety of domain adaptation techniques inspired by the maximum mean discrepancy (MMD) loss, Deep CORAL, and the domain adversarial neural network (DANN). From these findings, we additionally propose a novel architecture which takes advantage of the fact that in this setting, target domain data is unsupervised with regards to the 3D model but supervised for class labels. We base our framework off a recent network called pix2vox. Results are performed with ShapeNet as the source domain and domains within the Object Dataset Domain Suite (ODDS) dataset as the target, which is a real world multiview, multidomain image dataset. The domains in ODDS vary in difficulty, allowing us to assess notions of domain gap size. Our results are the first in the multiview reconstruction literature using this dataset.

Georgios Georgakis, Bernadette Bucher, Karl Schmeckpeper et al.

We consider the problem of object goal navigation in unseen environments. Solving this problem requires learning of contextual semantic priors, a challenging endeavour given the spatial and semantic variability of indoor environments. Current methods learn to implicitly encode these priors through goal-oriented navigation policy functions operating on spatial representations that are limited to the agent's observable areas. In this work, we propose a novel framework that actively learns to generate semantic maps outside the field of view of the agent and leverages the uncertainty over the semantic classes in the unobserved areas to decide on long term goals. We demonstrate that through this spatial prediction strategy, we are able to learn semantic priors in scenes that can be leveraged in unknown environments. Additionally, we show how different objectives can be defined by balancing exploration with exploitation during searching for semantic targets. Our method is validated in the visually realistic environments of the Matterport3D dataset and show improved results on object goal navigation over competitive baselines.

Shiladitya Bhattacharya, Siddharth Singh, Ritesh Kumar et al.

In this paper, we discuss the development of a multilingual annotated corpus of misogyny and aggression in Indian English, Hindi, and Indian Bangla as part of a project on studying and automatically identifying misogyny and communalism on social media (the ComMA Project). The dataset is collected from comments on YouTube videos and currently contains a total of over 20,000 comments. The comments are annotated at two levels - aggression (overtly aggressive, covertly aggressive, and non-aggressive) and misogyny (gendered and non-gendered). We describe the process of data collection, the tagset used for annotation, and issues and challenges faced during the process of annotation. Finally, we discuss the results of the baseline experiments conducted to develop a classifier for misogyny in the three languages.

Pinkesh Badjatiya, Mausoom Sarkar, Abhishek Sinha et al.

In social dilemma situations, individual rationality leads to sub-optimal group outcomes. Several human engagements can be modeled as a sequential (multi-step) social dilemmas. However, in contrast to humans, Deep Reinforcement Learning agents trained to optimize individual rewards in sequential social dilemmas converge to selfish, mutually harmful behavior. We introduce a status-quo loss (SQLoss) that encourages an agent to stick to the status quo, rather than repeatedly changing its policy. We show how agents trained with SQLoss evolve cooperative behavior in several social dilemma matrix games. To work with social dilemma games that have visual input, we propose GameDistill. GameDistill uses self-supervision and clustering to automatically extract cooperative and selfish policies from a social dilemma game. We combine GameDistill and SQLoss to show how agents evolve socially desirable cooperative behavior in the Coin Game.

Sudeep Dasari, Frederik Ebert, Stephen Tian et al.

Robot learning has emerged as a promising tool for taming the complexity and diversity of the real world. Methods based on high-capacity models, such as deep networks, hold the promise of providing effective generalization to a wide range of open-world environments. However, these same methods typically require large amounts of diverse training data to generalize effectively. In contrast, most robotic learning experiments are small-scale, single-domain, and single-robot. This leads to a frequent tension in robotic learning: how can we learn generalizable robotic controllers without having to collect impractically large amounts of data for each separate experiment? In this paper, we propose RoboNet, an open database for sharing robotic experience, which provides an initial pool of 15 million video frames, from 7 different robot platforms, and study how it can be used to learn generalizable models for vision-based robotic manipulation. We combine the dataset with two different learning algorithms: visual foresight, which uses forward video prediction models, and supervised inverse models. Our experiments test the learned algorithms' ability to work across new objects, new tasks, new scenes, new camera viewpoints, new grippers, or even entirely new robots. In our final experiment, we find that by pre-training on RoboNet and fine-tuning on data from a held-out Franka or Kuka robot, we can exceed the performance of a robot-specific training approach that uses 4x-20x more data. For videos and data, see the project webpage: https://www.robonet.wiki/

Tsung-Ting Kuo, Jina Huh, Jihoon Kim et al.

Objective. Annotation is expensive but essential for clinical note review and clinical natural language processing (cNLP). However, the extent to which computer-generated pre-annotation is beneficial to human annotation is still an open question. Our study introduces CLEAN (CLinical note rEview and ANnotation), a pre-annotation-based cNLP annotation system to improve clinical note annotation of data elements, and comprehensively compares CLEAN with the widely-used annotation system Brat Rapid Annotation Tool (BRAT). Materials and Methods. CLEAN includes an ensemble pipeline (CLEAN-EP) with a newly developed annotation tool (CLEAN-AT). A domain expert and a novice user/annotator participated in a comparative usability test by tagging 87 data elements related to Congestive Heart Failure (CHF) and Kawasaki Disease (KD) cohorts in 84 public notes. Results. CLEAN achieved higher note-level F1-score (0.896) over BRAT (0.820), with significant difference in correctness (P-value < 0.001), and the mostly related factor being system/software (P-value < 0.001). No significant difference (P-value 0.188) in annotation time was observed between CLEAN (7.262 minutes/note) and BRAT (8.286 minutes/note). The difference was mostly associated with note length (P-value < 0.001) and system/software (P-value 0.013). The expert reported CLEAN to be useful/satisfactory, while the novice reported slight improvements. Discussion. CLEAN improves the correctness of annotation and increases usefulness/satisfaction with the same level of efficiency. Limitations include untested impact of pre-annotation correctness rate, small sample size, small user size, and restrictedly validated gold standard. Conclusion. CLEAN with pre-annotation can be beneficial for an expert to deal with complex annotation tasks involving numerous and diverse target data elements.

Adarsh Modh, Siddharth Singh, A. V. S. Sai Bhargav Kumar et al.

In this paper, we present a novel control law for longitudinal speed control of autonomous vehicles. The key contributions of the proposed work include the design of a control law that reactively integrates the longitudinal surface gradient of road into its operation. In contrast to the existing works, we found that integrating the path gradient into the control framework improves the speed tracking efficacy. Since the control law is implemented over a shrinking domain scheme, it minimizes the integrated error by recomputing the control inputs at every discretized step and consequently provides less reaction time. This makes our control law suitable for motion planning frameworks that are operating at high frequencies. Furthermore, our work is implemented using a generalized vehicle model and can be easily extended to other classes of vehicles. The performance of gradient aware-shrinking domain based controller is implemented and tested on a stock electric vehicle on which a number of sensors are mounted. Results from the tests show the robustness of our control law for speed tracking on a terrain with varying gradient while also considering stringent time constraints imposed by the planning framework.