Avinash Kumar

Sourish Wawdhane, Avinash Kumar, Poulami Das

Mixture-of-Expert (MoE) models enable efficient inference by employing smaller experts and activating only a subset of them per token. MoE serving engines distribute experts across multiple GPUs and route tokens to appropriate GPUs at inference time based on experts activated. They process tokens in lock-step fashion, where tokens within a batch must finish processing before proceeding to the next layer. This synchronization barrier acts as a critical bottleneck because the performance of MoE models is limited by the straggler GPU that finishes last. Stragglers emerge when too many heavily used experts are placed on the same GPU or the slowest GPU. While prior works place experts that balance token loads across GPUs, they all overlook GPU variability and often place highly used experts on the slowest GPUs. We propose GEM, GPU-variability-aware Expert Mapping, a framework for GPU variability-aware expert to GPU mapping for MoE models. GEM exploits two insights. First, we must place experts such that each GPU receives non-uniform token loads based on their variability and they all finish processing a layer at about the same time. Our studies show that there are two types of experts: consistent that are used most of the time and temporal that are often used together for the remaining time. Our second insight is that we must place simultaneously used consistent and temporal experts on different GPUs and avoid placing them on slower GPUs to reduce slowdown. GEM gathers the variability profile of GPUs for each model and task and uses the token load distributions per task to map experts to GPUs. Our experiments show that GEM improves end-to-end latency by 7.9% on average and by up to 16.5% compared to the baseline.

Avinash Kumar, Anish Kumar, Sumit Sharma et al.

Current practice in parameter space exploration in euclidean space is dominated by randomized sampling or design of experiment methods. The biggest issue with these methods is not keeping track of what part of parameter space has been explored and what has not. In this context, we utilize the geometric learning of explored data space using modern machine learning methods to keep track of already explored regions and samples from the regions that are unexplored. For this purpose, we use a modified version of a robust random-cut forest along with other heuristic-based approaches. We demonstrate our method and its progression in two-dimensional Euclidean space but it can be extended to any dimension since the underlying method is generic.

Avinash Kumar, Sujay Sanghavi, Poulami Das

Speculative decoding accelerates LLM inference by using a fast draft model to generate tokens and a more accurate target model to verify them. Its performance depends on the $\textit{acceptance length}$, or number of draft tokens accepted by the target. Our studies show that the acceptance length of even state-of-the-art speculators, like DFlash, EAGLE-3 and PARD degrade with generation length, reaching values close to 1 (i.e. no speedup) within just a few thousand output tokens, making speculators ineffective for long-response tasks. Acceptance lengths decline because most speculators are trained offline on short sequences, but are forced to match the target model on much longer outputs at inference, well beyond their training distribution. To address this issue, we propose $\textit{Test-Time Speculation (TTS)}$, an online distillation approach that continuously adapts the speculator at test-time. TTS leverages the key insight that the token verification step already invokes the target model for each draft token, providing the training signal needed to adapt the draft at no additional cost. Treating the draft as the student and the target as a teacher, TTS adjusts the draft over several speculation rounds, with each update improving the draft's accuracy as generation proceeds. Our results across multiple models from the Qwen-3, Qwen-3.5, and Llama3.1 families show that TTS improves acceptance lengths over state-of-the-art speculators by up to $72\%$ and $41\%$ on average, with the benefits scaling with increased generation lengths.

Ravi Ghadia, Avinash Kumar, Gaurav Jain et al.

Autoregressive Transformers rely on Key-Value (KV) caching to accelerate inference. However, the linear growth of the KV cache with context length leads to excessive memory consumption and bandwidth constraints. This bottleneck is particularly problematic in real-time applications -- such as chatbots and interactive assistants -- where low latency and high memory efficiency are critical. Existing methods drop distant tokens or compress states in a lossy manner, sacrificing accuracy by discarding vital context or introducing bias. We propose MorphKV, an inference-time technique that maintains a constant-sized KV cache while preserving accuracy. MorphKV balances long-range dependencies and local coherence during text generation. It eliminates early-token bias while retaining high-fidelity context by adaptively ranking tokens through correlation-aware selection. Unlike heuristic retention or lossy compression, MorphKV iteratively refines the KV cache via lightweight updates guided by attention patterns of recent tokens. This approach captures inter-token correlation with greater accuracy, crucial for tasks like content creation and code generation. Our studies on long-response tasks show 52.9$\%$ memory savings and 18.2$\%$ higher accuracy on average compared to state-of-the-art prior works, enabling efficient real-world deployment.

Avinash Kumar, Sujay Sanghavi, Poulami Das

Speculative decoding accelerates LLM inference by using a smaller draft model to speculate tokens that a larger target model verifies. Verification is often the bottleneck (e.g. verification is $4\times$ slower than token generation when a 3B model speculates for a 70B target model), but most prior works focus only on accelerating drafting. $\textit{``Intermediate"}$ verification reduces verification time by discarding inaccurate draft tokens early, but existing methods incur substantial training overheads in incorporating the intermediate verifier, increase the memory footprint to orchestrate the intermediate verification step, and compromise accuracy by relying on approximate heuristics. We propose $\underline{\textit{Hi}}\textit{erarchical }\underline{\textit{Spec}}\textit{ulative Decoding (HiSpec)}$, a framework for high-throughput speculative decoding that exploits $\textit{early-exit (EE) models}$ for low-overhead intermediate verification. EE models allow tokens to exit early by skipping layer traversal and are explicitly trained so that hidden states at selected layers can be interpreted, making them uniquely suited for intermediate verification without drastically increasing compute and memory overheads. To improve resource-efficiency even further, we design a methodology that enables HiSpec to re-use key-value caches and hidden states between the draft, intermediate verifier, and target models. To maintain accuracy, HiSpec periodically validates the draft tokens accepted by the intermediate verifier against the target model. Our evaluations using various representative benchmarks and models show that HiSpec improves throughput by 1.28$\times$ on average and by up to 2.01$\times$ compared to the baseline single-layer speculation without compromising accuracy.

Sujaya Maiyya, Shantanu Sharma, Avinash Kumar

Managing personal health data is a challenge in today's fragmented and institution-centric healthcare ecosystem. Individuals often lack meaningful control over their medical records, which are scattered across incompatible systems and formats. This vision paper presents Health+, a user-centric, multimodal health data management system that empowers individuals (including those with limited technical expertise) to upload, query, and share their data across modalities (e.g., text, images, reports). Rather than aiming for institutional overhaul, Health+ emphasizes individual agency by providing intuitive interfaces and intelligent recommendations for data access and sharing. At the system level, it tackles the complexity of storing, integrating, and securing heterogeneous health records, ensuring both efficiency and privacy. By unifying multimodal data and prioritizing patients, Health+ lays the foundation for a more connected, interpretable, and user-controlled health information ecosystem.

Shvetank Prakash, Andrew Cheng, Arya Tschand et al.

The field of computer architecture, which bridges high-level software abstractions and low-level hardware implementations, remains absent from current large language model (LLM) evaluations. To this end, we present QuArch (pronounced 'quark'), the first benchmark designed to facilitate the development and evaluation of LLM knowledge and reasoning capabilities specifically in computer architecture. QuArch provides a comprehensive collection of 2,671 expert-validated question-answer (QA) pairs covering various aspects of computer architecture, including processor design, memory systems, and interconnection networks. Our evaluation reveals that while frontier models possess domain-specific knowledge, they struggle with skills that require higher-order thinking in computer architecture. Frontier model accuracies vary widely (from 34% to 72%) on these advanced questions, highlighting persistent gaps in architectural reasoning across analysis, design, and implementation QAs. By holistically assessing fundamental skills, QuArch provides a foundation for building and measuring LLM capabilities that can accelerate innovation in computing systems. With over 140 contributors from 40 institutions, this benchmark represents a community effort to set the standard for architectural reasoning in LLM evaluation.

Avinash Kumar, Kyeolhee Kang, Ammar ul Hassan et al.

Handwritten fonts have a distinct expressive character, but they are often difficult to read due to unclear or inconsistent handwriting. FontFusionGAN (FFGAN) is a novel method for improving handwritten fonts by combining them with printed fonts. Our method implements generative adversarial network (GAN) to generate font that mix the desirable features of handwritten and printed fonts. By training the GAN on a dataset of handwritten and printed fonts, it can generate legible and visually appealing font images. We apply our method to a dataset of handwritten fonts and demonstrate that it significantly enhances the readability of the original fonts while preserving their unique aesthetic. Our method has the potential to improve the readability of handwritten fonts, which would be helpful for a variety of applications including document creation, letter writing, and assisting individuals with reading and writing difficulties. In addition to addressing the difficulties of font creation for languages with complex character sets, our method is applicable to other text-image-related tasks, such as font attribute control and multilingual font style transfer.

Abdul Sami, Avinash Kumar, Irfanullah Memon et al.

Automatic font generation (AFG) is the process of creating a new font using only a few examples of the style images. Generating fonts for complex languages like Korean and Chinese, particularly in handwritten styles, presents significant challenges. Traditional AFGs, like Generative adversarial networks (GANs) and Variational Auto-Encoders (VAEs), are usually unstable during training and often face mode collapse problems. They also struggle to capture fine details within font images. To address these problems, we present a diffusion-based AFG method which generates high-quality, diverse Korean font images using only a single reference image, focusing on handwritten and printed styles. Our approach refines noisy images incrementally, ensuring stable training and visually appealing results. A key innovation is our text encoder, which processes phonetic representations to generate accurate and contextually correct characters, even for unseen characters. We used a pre-trained style encoder from DG FONT to effectively and accurately encode the style images. To further enhance the generation quality, we used perceptual loss that guides the model to focus on the global style of generated images. Experimental results on over 2000 Korean characters demonstrate that our model consistently generates accurate and detailed font images and outperforms benchmark methods, making it a reliable tool for generating authentic Korean fonts across different styles.

Avinash Kumar, Shashank Nag, Jason Clemons et al.

Early-Exit Large Language Models (EE-LLMs) enable high throughput inference by allowing tokens to exit early at intermediate layers. However, their throughput is limited by the computational and memory savings. Existing EE-LLM frameworks rely on a single model and therefore, their token generation latencies are bottlenecked by tokens that do not exit early and traverse additional layers. Moreover, early exits are only known at runtime and depend on the request. Therefore, these frameworks load the weights of all model layers even though large portions remain unused when tokens exit early. The lack of memory savings limit us from scaling the batch sizes. We propose $\textit{HELIOS}$, a framework that improves both token generation latency and batch sizes to enable high-throughput in EE-LLMs. HELIOS exploits two insights. $\textit{First}$, early exits are often complimentary across models, tokens that do not exit early on one model often take an early-exit on another. HELIOS employs multiple models and dynamically switches between them to collectively maximize the number of tokens that exit early, and minimize token generation latencies. $\textit{Second}$, even when a predicted token does not exit early due to poor confidence, it often remains unchanged even after additional layer traversal. HELIOS greedily allows such tokens to exit early and only loads the weights of the most likely to be used layers, yielding memory savings which is then re-purposed to increase batch sizes. HELIOS employs real-time profiling to accurately identify the early-exit distributions, and adaptively switches between models by tracking tokens in real-time to minimize the performance degradation caused by greedy model loading and exiting. Our evaluations show that HELIOS achieves $1.48\times$ higher throughput and $15.14\times$ larger batch size compared to existing EE-LLM frameworks.

Rohit Sharma, Shanu Kumar, Avinash Kumar

Pre-trained transformer models such as BERT have shown massive gains across many text classification tasks. However, these models usually need enormous labeled data to achieve impressive performances. Obtaining labeled data is often expensive and time-consuming, whereas collecting unlabeled data using some heuristics is relatively much cheaper for any task. Therefore, this paper proposes a method that encapsulates reinforcement learning-based text generation and semi-supervised adversarial learning approaches in a novel way to improve the model's performance. Our method READ, Reinforcement-based Adversarial learning, utilizes an unlabeled dataset to generate diverse synthetic text through reinforcement learning, improving the model's generalization capability using adversarial learning. Our experimental results show that READ outperforms the existing state-of-art methods on multiple datasets.

Avinash Kumar, Tushar Jain

This paper revisits the problem of optimal control law design for linear systems using the global optimal control framework introduced by Vadim Krotov. Krotov's approach is based on the idea of total decomposition of the original optimal control problem (OCP) with respect to time, by an $ad$ $hoc$ choice of the so-called Krotov's function or solving function, thereby providing sufficient conditions for the existence of global solution based on another optimization problem, which is completely equivalent to the original OCP. It is well known that the solution of this equivalent optimization problem is obtained using an iterative method. In this paper, we propose suitable Krotov's functions for linear quadratic OCP and subsequently, show that by imposing convexity condition on this equivalent optimization problem, there is no need to compute an iterative solution. We also give some key insights into the solution procedure of the linear quadratic OCP using the proposed methodology in contrast to the celebrated Calculus of Variations (CoV) and Hamilton-Jacobi-Bellman (HJB) equation based approach.

Tobias Johannink, Shikhar Bahl, Ashvin Nair et al.

Conventional feedback control methods can solve various types of robot control problems very efficiently by capturing the structure with explicit models, such as rigid body equations of motion. However, many control problems in modern manufacturing deal with contacts and friction, which are difficult to capture with first-order physical modeling. Hence, applying control design methodologies to these kinds of problems often results in brittle and inaccurate controllers, which have to be manually tuned for deployment. Reinforcement learning (RL) methods have been demonstrated to be capable of learning continuous robot controllers from interactions with the environment, even for problems that include friction and contacts. In this paper, we study how we can solve difficult control problems in the real world by decomposing them into a part that is solved efficiently by conventional feedback control methods, and the residual which is solved with RL. The final control policy is a superposition of both control signals. We demonstrate our approach by training an agent to successfully perform a real-world block assembly task involving contacts and unstable objects.