arXiv subjects

Siddharth Singh

Publications and source records attributed to Siddharth Singh.

At least 37 records · Page 2Linked to original sources

arXiv2025

Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach

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.

cs.LG↗

arXiv2025

Democratizing AI: Open-source Scalable LLM Training on GPU-based Supercomputers

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.

cs.LG↗

arXiv2024

HPC-Coder-V2: Studying Code LLMs Across Low-Resource Parallel Languages

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.

cs.DC↗

arXiv2024

Energy participation ratio analysis for very anharmonic superconducting circuits

Superconducting circuits are being employed for large-scale quantum devices, and a pertinent challenge is to perform accurate numerical simulations of device parameters. One of the most advanced methods for analyzing superconducting circuit designs is the energy participation ratio (EPR) method, which constructs quantum Hamiltonians based on the energy distribution extracted from classical electromagnetic simulations. In the EPR approach, we extract linear terms from finite element simulations and add nonlinear terms using the energy participation ratio extracted from the classical simulations. However, the EPR method relies on a low-order expansion of nonlinear terms, which is prohibitive for accurately describing highly anharmonic circuits. An example of such a circuit is the fluxonium qubit, which has recently attracted increasing attention due to its high lifetimes and low error rates. In this work, we extend the EPR approach to effectively address highly nonlinear superconducting circuits, and, as a proof of concept, we apply our approach to a fluxonium qubit. Specifically, we design, fabricate, and experimentally measure a fluxonium qubit coupled to a readout resonator. We compare the measured frequencies of both the qubit and the resonator to those extracted from the EPR analysis, and we find an excellent agreement. Furthermore, we compare the dispersive shift as a function of external flux obtained from experiments with our EPR analysis and a simpler lumped element model. Our findings reveal that the EPR results closely align with the experimental data, providing more accurate estimations compared to the simplified lumped element simulations.

quant-ph↗

arXiv2024

Improved fluxonium readout through dynamic flux pulsing

The ability to perform rapid, high fidelity readout of a qubit state is an important requirement for quantum algorithms and, in particular, for enabling operations such as mid-circuit measurements and measurement-based feedback for error correction schemes on large quantum processors. The growing interest in fluxonium qubits, due to their long coherence times and high anharmonicity, merits further attention to reducing the readout duration and measurement errors. We find that this can be accomplished by exploiting the flux tunability of fluxonium qubits. In this work, we experimentally demonstrate flux-pulse-assisted readout, as proposed in Phys. Rev. Applied 22, 014079 (https://doi.org/10.1103/PhysRevApplied.22.014079), in a setup without a quantum-limited parametric amplifier. Increasing the dispersive shift magnitude by almost 20% through flux pulsing, we achieve an assignment fidelity of 94.3% with an integration time of 280 ns. The readout performance is limited by state initialization, but we find that the limit imposed only by the signal-to-noise ratio corresponds to an assignment fidelity of 99.9% with a 360 ns integration time. We also verify these results through simple semi-classical simulations. These results constitute the fastest reported readout of a fluxonium qubit, with the prospect of further improvement by incorporation of a parametric amplifier in the readout chain to enhance measurement efficiency.

quant-ph↗

arXiv2024

Loki: Low-rank Keys for Efficient Sparse Attention

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.

cs.LG↗

arXiv2024

Be like a Goldfish, Don't Memorize! Mitigating Memorization in Generative LLMs

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.

cs.CL↗

arXiv2024

Semantic Search Evaluation

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.

cs.IR↗

arXiv2024

Lie Theory Based Optimization for Unified State Planning of Mobile Manipulators

Mobile manipulators are finding use in numerous practical applications. The current issues with mobile manipulation are the large state space owing to the mobile base and the challenge of modeling high degree of freedom systems. It is critical to devise fast and accurate algorithms that generate smooth motion plans for such mobile manipulators. Existing techniques attempt to solve this problem but focus on separating the motion of the base and manipulator. We propose an approach using Lie theory to find the inverse kinematic constraints by converting the kinematic model, created using screw coordinates, between its Lie group and vector representation. An optimization function is devised to solve for the desired joint states of the entire mobile manipulator. This allows the motion of the mobile base and manipulator to be planned and applied in unison resulting in a smooth and accurate motion plan. The performance of the proposed state planner is validated on simulated mobile manipulators in an analytical experiment. Our solver is available with further derivations and results at https://github.com/peleito/slithers.

cs.RO↗

arXiv2024

A Hidden Convexity of Nonlinear Elasticity

A technique for developing convex dual variational principles for the governing PDE of nonlinear elastostatics and elastodynamics is presented. This allows the definition of notions of a variational dual solution and a dual solution corresponding to the PDEs of nonlinear elasticity, even when the latter arise as formal Euler-Lagrange equations corresponding to non-quasiconvex elastic energy functionals whose energy minimizers do not exist. This is demonstrated rigorously in the case of elastostatics for the Saint-Venant Kirchhoff material (in all dimensions), where the existence of variational dual solutions is also proven. The existence of a variational dual solution for the incompressible neo-Hookean material in 2-d is also shown. Stressed and unstressed elastostatic and elastodynamic solutions in 1 space dimension corresponding to a non-convex, double-well energy are computed using the dual methodology. In particular, we show the stability of a dual elastodynamic equilibrium solution for which there are regions of non-vanishing length with negative elastic stiffness, i.e.~non-hyperbolic regions, for which the corresponding primal problem is ill-posed and demonstrates an explosive `Hadamard instability;' this appears to have implications for the modeling of physically observed softening behavior in macroscopic mechanical response.

math.AP↗

arXiv2024

A 4D Hybrid Algorithm to Scale Parallel Training to Thousands of GPUs

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.

cs.LG↗

arXiv2024

HarmPot: An Annotation Framework for Evaluating Offline Harm Potential of Social Media Text

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.

cs.CL↗

arXiv2023

A Framework for Realistic Simulation of Daily Human Activity

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.

cs.RO↗

arXiv2023

Jorge: Approximate Preconditioning for GPU-efficient Second-order Optimization

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.

cs.LG↗

arXiv2023

Modeling of experimentally observed topological defects inside bulk polycrystals

A rigorous methodology is developed for computing elastic fields generated by experimentally observed defect structures within grains in a polycrystal that has undergone tensile extension. An example application is made using a near-field High Energy X-ray Diffraction Microscope measurement of a zirconium sample that underwent $13.6\%$ tensile extension from an initially well-annealed state. (Sub)grain boundary features are identified with apparent disclination line defects in them. The elastic fields of these features identified from the experiment are calculated.

cond-mat.mtrl-sci↗

arXiv2023

AxoNN: An asynchronous, message-driven parallel framework for extreme-scale deep learning

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.

cs.LG↗

arXiv2023

Exploiting Sparsity in Pruned Neural Networks to Optimize Large Model Training

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.

cs.LG↗

arXiv2023

A Hybrid Tensor-Expert-Data Parallelism Approach to Optimize Mixture-of-Experts Training

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.

cs.LG↗