arXiv subjects

Dhruv Deshmukh

Publications and source records attributed to Dhruv Deshmukh.

4 recordsLinked to original sources

arXiv2026

GAUSS: A Workload-Aware Analytical and Monte-Carlo Simulator for LLM Inference Serving

Large language model (LLM) inference has become an operational workload that fleets must provision, monitor, and bill against latency service-level objectives (SLOs). A single throughput number cannot answer the questions operators actually face, because generative serving couples two phases with very different service characteristics, continuous batching reshapes the effective service process, and user-visible tail latency can degrade long before aggregate throughput saturates. We present GAUSS, a workload-aware analytical and Monte-Carlo simulator that predicts the full distributions of time-to-first-token (TTFT) and time-between-tokens (TBT), together with the sustainable request rate, for a given workload, engine, and accelerator. GAUSS pairs calibrated, phase-specific latency surfaces for prefill and decode with a Lindley-recurrence model of queueing under stochastic arrivals and estimates performance by ensemble Monte-Carlo sampling rather than by replaying a single trace. This lets GAUSS reason about workloads, routing policies, and hardware that have never been deployed, and to capture the latency tails that trace-driven simulators systematically under-sample. We describe the model and show how it supports operating-point selection, traffic segmentation, and shortest-remaining-processing-time scheduling capacity questions that trace-driven simulators cannot address.

cs.AI↗

arXiv2026

Kascade: A Practical Sparse Attention Method for Long-Context LLM Inference

Attention is the dominant source of latency during long-context LLM inference, an increasingly popular workload with reasoning models and RAG. We propose Kascade, a training-free sparse attention method that leverages known observations such as 1) post-softmax attention is intrinsically sparse, and 2) the identity of high-weight keys is stable across nearby layers. Kascade computes exact Top-k indices in a small set of anchor layers, then reuses those indices in intermediate reuse layers. The anchor layers are selected algorithmically, via a dynamic-programming objective that maximizes cross-layer similarity over a development set, allowing easy deployment across models. The method incorporates efficient implementation constraints (e.g. tile-level operations), across both prefill and decode attention. The Top-k selection and reuse in Kascade is head-aware and we show in our experiments that this is critical for high accuracy. Kascade achieves up to 4.1x speedup in decode attention and 2.2x speedup in prefill attention over FlashAttention-3 baseline on H100 GPUs while closely matching dense attention accuracy on long-context benchmarks such as LongBench and AIME-24.

cs.LG↗

arXiv2026

Observation of period doubling and higher multiplicities in a driven single-spin system

One of the prime features of quantum systems strongly driven by external time-periodic fields is the subharmonic response with integer multiples of the drive period $k\, T_d$ due to long-lived interference. Here, we demonstrate experimentally, based on a careful theoretical analysis, period doubling and higher multiplicities ($k=2,\ldots 5$) for one of the most fundamental systems, namely, an individual spin $1/2$. Nitrogen-vacancy centers in diamond support sufficiently stable coherent dynamics owing to long coherence times and allow for optical addressability of their spin states. This allows to monitor coherent period $k$-tupling oscillations over a broad set of driving parameters in the vicinity of the ideal manifolds. In this domain, superimposed low-frequency modulations serve as unique proxy for the approach toward period $k$-tupling.

quant-ph↗

arXiv2023

Entropy Aware Training for Fast and Accurate Distributed GNN

Several distributed frameworks have been developed to scale Graph Neural Networks (GNNs) on billion-size graphs. On several benchmarks, we observe that the graph partitions generated by these frameworks have heterogeneous data distributions and class imbalance, affecting convergence, and resulting in lower performance than centralized implementations. We holistically address these challenges and develop techniques that reduce training time and improve accuracy. We develop an Edge-Weighted partitioning technique to improve the micro average F1 score (accuracy) by minimizing the total entropy. Furthermore, we add an asynchronous personalization phase that adapts each compute-host's model to its local data distribution. We design a class-balanced sampler that considerably speeds up convergence. We implemented our algorithms on the DistDGL framework and observed that our training techniques scale much better than the existing training approach. We achieved a (2-3x) speedup in training time and 4\% improvement on average in micro-F1 scores on 5 large graph benchmarks compared to the standard baselines.

cs.LG↗