arXiv subjects

Saurabh Agarwal

Publications and source records attributed to Saurabh Agarwal.

At least 19 recordsLinked to original sources

arXiv2026

ACID: Adaptive Caching for vIDeo generation

Video diffusion models produce high-quality generations but remain slow at inference due to their sequential denoising procedure. Caching-based acceleration methods address this by reusing intermediate model outputs: leading dynamic approaches such as TeaCache, EasyCache, and DiCache accumulate a drift signal and skip expensive model evaluations when accumulated drift stays below a fixed threshold $τ$. This threshold controls an apparent tradeoff - raising it yields faster generation at the cost of visual quality, while lowering it preserves quality but sacrifices speed. We show this tradeoff is not fundamental; it is an artifact of holding $τ$ constant throughout denoising. We identify the existence of critical steps - timesteps where the drift signal changes rapidly - and show that applying a low threshold selectively at these steps while caching aggressively elsewhere recovers most of the quality of conservative caching at substantially higher inference speeds. Building on this insight, we propose ACID, a lightweight, training-free wrapper that monitors the rate of change of each method's existing drift signal to dynamically switch between a low and a high threshold. ACID is signal-agnostic and modular: it requires no retraining and plugs directly into existing dynamic caching methods without modifying their core mechanisms. Evaluated across three caching methods (TeaCache, EasyCache, DiCache) and three open-source video diffusion models (HunyuanVideo, Wan 2.1, CogVideoX), ACID consistently expands the Pareto frontier of visual quality versus inference speed beyond what any fixed threshold achieves. In particular, on TeaCache and HunyuanVideo, ACID achieves up to 2.16x speedup over the no-caching baseline, and up to 38% additional speedup over the conservative fixed-threshold baseline with negligible (<0.3 dB PSNR, <0.01 SSIM, <0.01 LPIPS) quality degradation.

cs.CV↗

arXiv2026

Sangam: Efficiently Serving Diffusion LLMs with the AR Stack

Diffusion language models (dLLMs) generate text by iteratively denoising a masked response and can commit multiple output positions per model invocation. Their bidirectional attention prevents exact autoregressive-style KV caching, since committing one position shifts the KV activations of all others. Approximate caching techniques such as Fast-dLLM and dKV-Cache refresh KV activations repeatedly and reuse them across intervening decodes, inducing a repeated prefill/decode structure. This makes AR serving mechanisms relevant to dLLMs, but not directly applicable. dLLM decodes are block-sized rather than token-sized, prefills recur, and bidirectional attention precludes the chunked prefill mechanism used for stall-free colocated serving. We present Sangam, a serving system for cached dLLM inference. Sangam introduces a deficit token-budget scheduler that admits in-flight decodes first, admits whole indivisible prefills only when the accumulated token budget allows, and carries unused budget forward. This achieves amortized stall-free scheduling. Disaggregated serving avoids prefill-decode interference but suffers from prefill/decode resource partitioning problem. Sangam adopts a hybrid serving strategy, overflowing prefills onto decode workers to relieve prefill under-provisioning, and uses the same deficit-budget scheduler to protect those workers' decodes from the overflow. We show that like AR serving, dLLM serving design space is governed by prefill-decode interference and prefill/decode partitioning. Colocated serving is most effective on decode-heavy workloads, cutting mean latency by 9-20% over hybrid execution on LLaDA-8B ShareGPT; while hybrid execution is most effective on prefill-heavy workloads, cutting mean latency by 8-20% over colocated execution on Dream-7B arXiv. Sangam is available at https://github.com/UT-InfraAI/sangam.

cs.DC↗

arXiv2026

Harmonia: End-to-End RAG Serving Optimization

Retrieval-Augmented Generation (RAG) improves the reliability of large language models by integrating external knowledge, but serving RAG pipelines efficiently is challenging because requests traverse heterogeneous components spanning LLM inference, databases, and CPU-side processing. We present Harmonia, an end-to-end RAG serving framework that addresses these bottlenecks through (i) a flexible pipeline specification interface for composing custom workflows, (ii) heterogeneity-aware deployment that provisions and configures components as a distributed inference system, and (iii) a closed-loop runtime controller that monitors load and execution progress and reduces SLO violations through request prioritization and auto-scaling. Across four RAG applications, Harmonia outperforms commercial alternatives, improving throughput by more than 2.04x while reducing SLO violations by up to 78.4 percent.

cs.DC↗

arXiv2026

CUCo: An Agentic Framework for Compute and Communication Co-design

Computation and communication in distributed LLM training and inference are traditionally optimized in isolation; expert-crafted systems such as DeepEP, FLUX, and TokenWeave show the potential of co-design but require deep systems expertise and hardware-specific tuning; CUCo is an agentic framework that automates compute-communication co-design of CUDA kernels by combining a structured design-space formalization with a correctness-first fast-path agent for reliable baselines and an evolution-driven slow-path agent for high-performance strategies, achieving up to 1.57x speedup across four multi-GPU workloads and discovering a two-stream overlap strategy on a DeepSeek-V3 MoE layer that hides dispatch behind local compute at an LLM inference cost under $10 per workload.

cs.DC↗

arXiv2026

RL-ACRGNet: Reinforcement Learning-Based Chest Radiology Report Generation Network

Medical imaging interpretation is a foundational pillar of modern clinical diagnostics, yet the manual generation of radiology reports remains a time-consuming process prone to interpretation inconsistencies. Within the field of medical AI, automating these descriptions through deep learning promises to streamline clinical workflows and standardise diagnostic output. However, accurate disease detection and precise report generation remain significant challenges due to limitations in capturing fine-grained visual features and ensuring clinical coherence. To address these issues, we propose RL-ACRGNet, an improved encoder-decoder model that integrates a pre-trained DenseNet encoder with a multilevel LSTM decoder within an off-policy reinforcement learning framework. Using a dual-network approach to refine visual-semantic embeddings through a metric-based reward mechanism, we demonstrate that RL-ACRGNet consistently outperforms state-of-the-art baselines on the IU-Xray dataset, achieving quantitative improvements in BLEU-4 (0.47%), METEOR (0.17%) and ROUGE-L (0.518). Furthermore, comprehensive evaluations on the large-scale MIMIC-CXR data set confirm the robust generalisation of the model and its ability to generate high-quality, clinically relevant reports

cs.AI↗

arXiv2026

Nalar: An agent serving framework

LLM-driven agentic applications increasingly automate complex, multi-step tasks, but serving them efficiently remains challenging due to heterogeneous components, dynamic and model-driven control flow, long-running state, and unpredictable latencies. Nalar is a ground-up agent-serving framework that cleanly separates workflow specification from execution while providing the runtime visibility and control needed for robust performance. Nalar preserves full Python expressiveness, using lightweight auto-generated stubs that turn agent and tool invocations into futures carrying dependency and context metadata. A managed state layer decouples logical state from physical placement, enabling safe reuse, migration, and consistent retry behavior. A two-level control architecture combines global policy computation with local event-driven enforcement to support adaptive routing, scheduling, and resource management across evolving workflows. Together, these mechanisms allow Nalar to deliver scalable, efficient, and policy-driven serving of heterogeneous agentic applications without burdening developers with orchestration logic. Across three agentic workloads, Nalar cuts tail latency by 34--74\%, achieves up to $2.9\times$ speedups, sustains 80 RPS where baselines fail, and scales to 130K futures with sub-500 ms control overhead.

cs.DC↗

arXiv2026

Software-Defined Agentic Serving

As multi-agent LLM pipelines grow in complexity, existing serving paradigms fail to adapt to the dynamic serving conditions. We argue that agentic serving systems should be programmable and system-aware, unlike existing serving which statically encode the parameters. In this work, we propose a new SDN-inspired agentic serving framework that helps control the key attributes of communication based on runtime state. This architecture enables serving-efficient, responsive agent systems and paves the way for high-level intent-driven agentic serving.

cs.DC↗

arXiv2025

SuperGen: An Efficient Ultra-high-resolution Video Generation System with Sketching and Tiling

Diffusion models have recently achieved remarkable success in generative tasks (e.g., image and video generation), and the demand for high-quality content (e.g., 2K/4K videos) is rapidly increasing across various domains. However, generating ultra-high-resolution videos on existing standard-resolution (e.g., 720p) platforms remains challenging due to the excessive re-training requirements and prohibitively high computational and memory costs. To this end, we introduce SUPERGEN, an efficient tile-based framework for ultra-high-resolution video generation. SUPERGEN features a novel training-free algorithmic innovation with tiling to successfully support a wide range of resolutions without additional training efforts while significantly reducing both memory footprint and computational complexity. Moreover, SUPERGEN incorporates a tile-tailored, adaptive, region-aware caching strategy that accelerates video generation by exploiting redundancy across denoising steps and spatial regions. SUPERGEN also integrates cache-guided, communication-minimized tile parallelism for enhanced throughput and minimized latency. Evaluations show that SUPERGEN maximizes performance gains while achieving high output quality across various benchmarks.

cs.LG↗

arXiv2025

Tesserae: Scalable Placement Policies for Deep Learning Workloads

Training deep learning (DL) models has become a dominant workload in data-centers and improving resource utilization is a key goal of DL cluster schedulers. In order to do this, schedulers typically incorporate placement policies that govern where jobs are placed on the cluster. Existing placement policies are either designed as ad-hoc heuristics or incorporated as constraints within a complex optimization problem and thus either suffer from suboptimal performance or poor scalability. Our key insight is that many placement constraints can be formulated as graph matching problems and based on that we design novel placement policies for minimizing job migration overheads and job packing. We integrate these policies into Tesserae and describe how our design leads to a scalable and effective GPU cluster scheduler. Our experimental results show that Tesserae improves average JCT by up to 1.62x and the Makespan by up to 1.15x compared with the existing schedulers.

cs.DC↗

arXiv2025

Decoding Speculative Decoding

Speculative Decoding is a widely used technique to speed up inference for Large Language Models (LLMs) without sacrificing quality. When performing inference, speculative decoding uses a smaller draft model to generate speculative tokens and then uses the target LLM to verify those draft tokens. The speedup provided by speculative decoding heavily depends on the choice of the draft model. In this work, we perform a detailed study comprising over 350 experiments with LLaMA-65B and OPT-66B using speculative decoding and delineate the factors that affect the performance gain provided by speculative decoding. Our experiments indicate that the performance of speculative decoding depends heavily on the latency of the draft model, and the draft model's capability in language modeling does not correlate strongly with its performance in speculative decoding. Based on these insights we explore a new design space for draft models and design hardware-efficient draft models for speculative decoding. Our newly designed draft model can provide 111% higher throughput than existing draft models and our approach generalizes further to all LLaMA models (1/2/3.1) and supervised fine-tuned models.

cs.LG↗

arXiv2024

SYMPHONY: Improving Memory Management for LLM Inference Workloads

Large Language Models (LLMs) are increasingly being deployed in applications such as chatbots, code editors, and conversational agents. A key feature of LLMs is their ability to engage in multi-turn interactions with humans or external tools, enabling a wide range of tasks. Each new request in a multi-turn interaction depends on the intermediate state, specifically the key-value (K,V) caches, from previous requests in the ongoing interaction. Existing serving engines either recompute the K,V caches or offload them to main memory. Profiling reveals that recomputation can result in over 99% of processed tokens being redundant. On the other hand, offloading K,V caches from GPU memory makes inference serving stateful, leading to load imbalances across the cluster. To address these challenges, we developed SYMPHONY. SYMPHONY leverages the observation that multi-turn work loads provide additional hints that allow K,V caches to be migrated off the critical serving path. By utilizing these hints, SYMPHONY dynamically migrates K,V caches to enable finegrained scheduling of inference requests. Our experiments demonstrate that SYMPHONY can handle over 8x the number of requests compared to state-of-the-art baselines, with a similar latency profile.

cs.DC↗

arXiv2024

LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding

We present LayerSkip, an end-to-end solution to speed-up inference of large language models (LLMs). First, during training we apply layer dropout, with low dropout rates for earlier layers and higher dropout rates for later layers, and an early exit loss where all transformer layers share the same exit. Second, during inference, we show that this training recipe increases the accuracy of early exit at earlier layers, without adding any auxiliary layers or modules to the model. Third, we present a novel self-speculative decoding solution where we exit at early layers and verify and correct with remaining layers of the model. Our proposed self-speculative decoding approach has less memory footprint than other speculative decoding approaches and benefits from shared compute and activations of the draft and verification stages. We run experiments on different Llama model sizes on different types of training: pretraining from scratch, continual pretraining, finetuning on specific data domain, and finetuning on specific task. We implement our inference solution and show speedups of up to 2.16x on summarization for CNN/DM documents, 1.82x on coding, and 2.0x on TOPv2 semantic parsing task. We open source our code and checkpoints at https://github.com/facebookresearch/LayerSkip.

cs.CL↗

arXiv2024

Everything Everywhere All at Once: LLMs can In-Context Learn Multiple Tasks in Superposition

Large Language Models (LLMs) have demonstrated remarkable in-context learning (ICL) capabilities. In this study, we explore a surprising phenomenon related to ICL: LLMs can perform multiple, computationally distinct ICL tasks simultaneously, during a single inference call, a capability we term "task superposition". We provide empirical evidence of this phenomenon across various LLM families and scales and show that this phenomenon emerges even if we train the model to in-context learn one task at a time. We offer theoretical explanations that this capability is well within the expressive power of transformers. We also explore how LLMs internally compose task vectors during superposition. Furthermore, we show that larger models can solve more ICL tasks in parallel, and better calibrate their output distribution. Our findings offer insights into the latent capabilities of LLMs, further substantiate the perspective of "LLMs as superposition of simulators", and raise questions about the mechanisms enabling simultaneous task execution.

cs.LG↗

arXiv2024

CHAI: Clustered Head Attention for Efficient LLM Inference

Large Language Models (LLMs) with hundreds of billions of parameters have transformed the field of machine learning. However, serving these models at inference time is both compute and memory intensive, where a single request can require multiple GPUs and tens of Gigabytes of memory. Multi-Head Attention is one of the key components of LLMs, which can account for over 50% of LLMs memory and compute requirement. We observe that there is a high amount of redundancy across heads on which tokens they pay attention to. Based on this insight, we propose Clustered Head Attention (CHAI). CHAI combines heads with a high amount of correlation for self-attention at runtime, thus reducing both memory and compute. In our experiments, we show that CHAI is able to reduce the memory requirements for storing K,V cache by up to 21.4% and inference time latency by up to 1.73x without any fine-tuning required. CHAI achieves this with a maximum 3.2% deviation in accuracy across 3 different models (i.e. OPT-66B, LLAMA-7B, LLAMA-33B) and 5 different evaluation datasets.

cs.LG↗

arXiv2024

MultiFusionNet: Multilayer Multimodal Fusion of Deep Neural Networks for Chest X-Ray Image Classification

Chest X-ray imaging is a critical diagnostic tool for identifying pulmonary diseases. However, manual interpretation of these images is time-consuming and error-prone. Automated systems utilizing convolutional neural networks (CNNs) have shown promise in improving the accuracy and efficiency of chest X-ray image classification. While previous work has mainly focused on using feature maps from the final convolution layer, there is a need to explore the benefits of leveraging additional layers for improved disease classification. Extracting robust features from limited medical image datasets remains a critical challenge. In this paper, we propose a novel deep learning-based multilayer multimodal fusion model that emphasizes extracting features from different layers and fusing them. Our disease detection model considers the discriminatory information captured by each layer. Furthermore, we propose the fusion of different-sized feature maps (FDSFM) module to effectively merge feature maps from diverse layers. The proposed model achieves a significantly higher accuracy of 97.21% and 99.60% for both three-class and two-class classifications, respectively. The proposed multilayer multimodal fusion model, along with the FDSFM module, holds promise for accurate disease classification and can also be extended to other disease classifications in chest X-ray images.

eess.IV↗

arXiv2023

Blox: A Modular Toolkit for Deep Learning Schedulers

Deep Learning (DL) workloads have rapidly increased in popularity in enterprise clusters and several new cluster schedulers have been proposed in recent years to support these workloads. With rapidly evolving DL workloads, it is challenging to quickly prototype and compare scheduling policies across workloads. Further, as prior systems target different aspects of scheduling (resource allocation, placement, elasticity etc.), it is also challenging to combine these techniques and understand the overall benefits. To address these challenges we propose Blox, a modular toolkit which allows developers to compose individual components and realize diverse scheduling frameworks. We identify a set of core abstractions for DL scheduling, implement several existing schedulers using these abstractions, and verify the fidelity of these implementations by reproducing results from prior research. We also highlight how we can evaluate and compare existing schedulers in new settings: different workload traces, higher cluster load, change in DNN workloads and deployment characteristics. Finally, we showcase Blox's extensibility by composing policies from different schedulers, and implementing novel policies with minimal code changes. Blox is available at \url{https://github.com/msr-fiddle/blox}.

cs.DC↗

arXiv2023

BagPipe: Accelerating Deep Recommendation Model Training

Deep learning based recommendation models (DLRM) are widely used in several business critical applications. Training such recommendation models efficiently is challenging because they contain billions of embedding-based parameters, leading to significant overheads from embedding access. By profiling existing systems for DLRM training, we observe that around 75\% of the iteration time is spent on embedding access and model synchronization. Our key insight in this paper is that embedding access has a specific structure which can be used to accelerate training. We observe that embedding accesses are heavily skewed, with around 1\% of embeddings representing more than 92\% of total accesses. Further, we observe that during offline training we can lookahead at future batches to determine exactly which embeddings will be needed at what iteration in the future. Based on these insights, we develop Bagpipe, a system for training deep recommendation models that uses caching and prefetching to overlap remote embedding accesses with the computation. We design an Oracle Cacher, a new component that uses a lookahead algorithm to generate optimal cache update decisions while providing strong consistency guarantees against staleness. We also design a logically replicated, physically partitioned cache and show that our design can reduce synchronization overheads in a distributed setting. Finally, we propose a disaggregated system architecture and show that our design can enable low-overhead fault tolerance. Our experiments using three datasets and four models show that Bagpipe provides a speed up of up to 5.6x compared to state of the art baselines, while providing the same convergence and reproducibility guarantees as synchronous training.

cs.DC↗

arXiv2023

Cuttlefish: Low-Rank Model Training without All the Tuning

Recent research has shown that training low-rank neural networks can effectively reduce the total number of trainable parameters without sacrificing predictive accuracy, resulting in end-to-end speedups. However, low-rank model training necessitates adjusting several additional factorization hyperparameters, such as the rank of the factorization at each layer. In this paper, we tackle this challenge by introducing Cuttlefish, an automated low-rank training approach that eliminates the need for tuning factorization hyperparameters. Cuttlefish leverages the observation that after a few epochs of full-rank training, the stable rank (i.e., an approximation of the true rank) of each layer stabilizes at a constant value. Cuttlefish switches from full-rank to low-rank training once the stable ranks of all layers have converged, setting the dimension of each factorization to its corresponding stable rank. Our results show that Cuttlefish generates models up to 5.6 times smaller than full-rank models, and attains up to a 1.2 times faster end-to-end training process while preserving comparable accuracy. Moreover, Cuttlefish outperforms state-of-the-art low-rank model training methods and other prominent baselines. The source code for our implementation can be found at: https://github.com/hwang595/Cuttlefish.

cs.LG↗