arXiv subjects

Tahseen Rabbani

Publications and source records attributed to Tahseen Rabbani.

At least 19 recordsLinked to original sources

arXiv2026

Not All Prefills Are Equal: PPD Disaggregation for Multi-turn LLM Serving

Prefill-Decode (PD) disaggregation has become the standard architecture for modern LLM inference engines, which alleviates the interference of two distinctive workloads. With the growing demand for multi-turn interactions in chatbots and agentic systems, we re-examined PD in this case and found two fundamental inefficiencies: (1) every turn requires prefilling the new prompt and response from the last turn, and (2) repeated KV transfers between prefill and decode nodes saturate the bandwidth, leading to high latency and even service degradation. Our key insight is that not all prefill operations are equally disruptive: append-prefill, which processes only the new input tokens while reusing cached KV states, incurs an order-of-magnitude smaller decoding slowdown than full prefill. This motivates routing append-prefill to decode nodes locally. However, through comprehensive analysis, we show that no single fixed routing strategy satisfies all Service Level Objectives (SLOs) simultaneously. Based on this insight, we propose Prefill Prefill-capable Decode (PPD) disaggregation, a dynamic routing system that decides when to process Turn 2+ requests locally on decode nodes using cached KV states. PPD adapts to varying SLOs via configurable weights and seamlessly integrates with traditional PD deployments. With extensive evaluations, we show that PPD reduces Turn 2+ time-to-first-token (TTFT) by $\sim$68\% while maintaining competitive time-per-output-token (TPOT), effectively alleviating KV transfer congestion under high load. PPD provides a flexible and efficient paradigm for multi-turn LLM serving.

cs.NI↗

arXiv2026

Federation over Text: Insight Sharing for Multi-Agent Reasoning

We propose a federated learning-like framework, Federation over Text (FoT), that enables multiple clients solving different tasks to collectively generate a shared library of metacognitive insights by iteratively federating their local reasoning processes without sharing actual problem instances or task instructions. Instead of federation over gradients (e.g., as in distributed training), FoT operates at the semantic level without any gradient optimization or supervision signal. Iteratively, each client runs an LLM agent that does local thinking and self-improvement on their specific tasks independently, and shares reasoning traces with a central server, which aggregates and distills them into a cross-task (and cross-domain) insight library that existing and future agents can leverage to improve performance on related tasks. Experiments show that FoT improves reasoning effectiveness and efficiency across a wide range of challenging applications, including mathematical problem solving, cross-domain collaboration, real-world daily tasks, and machine learning research insight discovery. Specifically, it improves average performance scores by 25% while reducing the reasoning tokens by 4% across the first three applications. In the research insight discovery application, FoT is able to generate insights that cover over 80% of the major contributions in the subsequent papers.

cs.LG↗

arXiv2026

Mitigating Unintended Memorization with LoRA in Federated Learning for LLMs

Federated learning (FL) is a popular paradigm for collaborative training which avoids direct data exposure between clients. However, data privacy issues still remain: FL-trained large language models are capable of memorizing and completing phrases and sentences contained in training data when given their prefixes. Thus, it is possible for adversarial and honest- but-curious clients to recover training data of other participants simply through targeted prompting. In this work, we demonstrate that a popular and simple fine-tuning strategy, low-rank adaptation (LoRA), reduces memorization during FL by a factor of up to 10 without significant performance cost. We study this effect by performing fine-tuning tasks in high-risk domains such as medicine, law, and finance. We observe a reduction in memorization for a wide variety of model families, from 1B to 70B parameters. We find that LoRA can reduce memorization in centralized learning as well, and we compare how the memorization patterns differ. Furthermore, we study the effect of hyperparameters and show that LoRA can be combined with other privacy-preserving techniques such as gradient clipping and Gaussian noise, secure aggregation, and Goldfish loss to further improve record-level privacy while maintaining performance.

cs.LG↗

arXiv2026

Asynchronous Heavy-Tailed Optimization

Heavy-tailed stochastic gradient noise, commonly observed in transformer models, can destabilize the optimization process. Recent works mainly focus on developing and understanding approaches to address heavy-tailed noise in the centralized or distributed, synchronous setting, leaving the interactions between such noise and asynchronous optimization underexplored. In this work, we investigate two communication schemes that handle stragglers with asynchronous updates in the presence of heavy-tailed gradient noise. We propose and theoretically analyze algorithmic modifications based on delay-aware learning rate scheduling and delay compensation to enhance the performance of asynchronous algorithms. Our convergence guarantees under heavy-tailed noise match the rate of the synchronous counterparts and improve delay tolerance compared with existing asynchronous approaches. Empirically, our approaches outperform prior synchronous and asynchronous methods in terms of accuracy/runtime trade-offs and are more robust to hyperparameters in both image and language tasks.

cs.LG↗

arXiv2025

Hold Onto That Thought: Assessing KV Cache Compression On Reasoning

Large language models (LLMs) have demonstrated remarkable performance on long-context tasks, but are often bottlenecked by memory constraints. Namely, the KV cache, which is used to significantly speed up attention computations, grows linearly with context length. A suite of compression algorithms has been introduced to alleviate cache growth by evicting unimportant tokens. However, several popular strategies are targeted towards the prefill phase, i.e., processing long prompt context, and their performance is rarely assessed on reasoning tasks requiring long decoding. In particular, short but complex prompts, such as those in benchmarks like GSM8K and MATH500, often benefit from multi-step reasoning and self-reflection, resulting in thinking sequences thousands of tokens long. In this work, we benchmark the performance of several popular compression strategies on long-reasoning tasks. For the non-reasoning Llama-3.1-8B-Instruct, we determine that no singular strategy fits all, and that performance is heavily influenced by dataset type. However, we discover that H2O and our decoding-enabled variant of SnapKV are dominant strategies for reasoning models, indicating the utility of heavy-hitter tracking for reasoning traces. We also find that eviction strategies at low budgets can produce longer reasoning traces, revealing a tradeoff between cache size and inference costs.

cs.CL↗

arXiv2025

DISCO: A Browser-Based Privacy-Preserving Framework for Distributed Collaborative Learning

Data is often impractical to share for a range of well considered reasons, such as concerns over privacy, intellectual property, and legal constraints. This not only fragments the statistical power of predictive models, but creates an accessibility bias, where accuracy becomes inequitably distributed to those who have the resources to overcome these concerns. We present DISCO: an open-source DIStributed COllaborative learning platform accessible to non-technical users, offering a means to collaboratively build machine learning models without sharing any original data or requiring any programming knowledge. DISCO's web application trains models locally directly in the browser, making our tool cross-platform out-of-the-box, including smartphones. The modular design of \disco offers choices between federated and decentralized paradigms, various levels of privacy guarantees and several approaches to weight aggregation strategies that allow for model personalization and bias resilience in the collaborative training. Code repository is available at https://github.com/epfml/disco and a showcase web interface at https://discolab.ai

cs.LG↗

arXiv2025

PRBench: Large-Scale Expert Rubrics for Evaluating High-Stakes Professional Reasoning

Frontier model progress is often measured by academic benchmarks, which offer a limited view of performance in real-world professional contexts. Existing evaluations often fail to assess open-ended, economically consequential tasks in high-stakes domains like Legal and Finance, where practical returns are paramount. To address this, we introduce Professional Reasoning Bench (PRBench), a realistic, open-ended, and difficult benchmark of real-world problems in Finance and Law. We open-source its 1,100 expert-authored tasks and 19,356 expert-curated criteria, making it, to our knowledge, the largest public, rubric-based benchmark for both legal and finance domains. We recruit 182 qualified professionals, holding JDs, CFAs, or 6+ years of experience, who contributed tasks inspired by their actual workflows. This process yields significant diversity, with tasks spanning 114 countries and 47 US jurisdictions. Our expert-curated rubrics are validated through a rigorous quality pipeline, including independent expert validation. Subsequent evaluation of 20 leading models reveals substantial room for improvement, with top scores of only 0.39 (Finance) and 0.37 (Legal) on our Hard subsets. We further catalog associated economic impacts of the prompts and analyze performance using human-annotated rubric categories. Our analysis shows that models with similar overall scores can diverge significantly on specific capabilities. Common failure modes include inaccurate judgments, a lack of process transparency and incomplete reasoning, highlighting critical gaps in their reliability for professional adoption.

cs.CL↗

arXiv2025

ResearchRubrics: A Benchmark of Prompts and Rubrics For Evaluating Deep Research Agents

Deep Research (DR) is an emerging agent application that leverages large language models (LLMs) to address open-ended queries. It requires the integration of several capabilities, including multi-step reasoning, cross-document synthesis, and the generation of evidence-backed, long-form answers. Evaluating DR remains challenging because responses are lengthy and diverse, admit many valid solutions, and often depend on dynamic information sources. We introduce ResearchRubrics, a standardized benchmark for DR built with over 2,800+ hours of human labor that pairs realistic, domain-diverse prompts with 2,500+ expert-written, fine-grained rubrics to assess factual grounding, reasoning soundness, and clarity. We also propose a new complexity framework for categorizing DR tasks along three axes: conceptual breadth, logical nesting, and exploration. In addition, we develop human and model-based evaluation protocols that measure rubric adherence for DR agents. We evaluate several state-of-the-art DR systems and find that even leading agents like Gemini's DR and OpenAI's DR achieve under 68% average compliance with our rubrics, primarily due to missed implicit context and inadequate reasoning about retrieved information. Our results highlight the need for robust, scalable assessment of deep research capabilities, to which end we release ResearchRubrics(including all prompts, rubrics, and evaluation code) to facilitate progress toward well-justified research assistants.

cs.AI↗

arXiv2025

Remote Labor Index: Measuring AI Automation of Remote Work

AIs have made rapid progress on research-oriented benchmarks of knowledge and reasoning, but it remains unclear how these gains translate into economic value and automation. To measure this, we introduce the Remote Labor Index (RLI), a broadly multi-sector benchmark comprising real-world, economically valuable projects designed to evaluate end-to-end agent performance in practical settings. AI agents perform near the floor on RLI, with the highest-performing agent achieving an automation rate of 2.5%. These results help ground discussions of AI automation in empirical evidence, setting a common basis for tracking AI impacts and enabling stakeholders to proactively navigate AI-driven labor automation.

cs.LG↗

arXiv2025

Access Paths for Efficient Ordering with Large Language Models

In this work, we present the \texttt{LLM ORDER BY} semantic operator as a logical abstraction and conduct a systematic study of its physical implementations. First, we propose several improvements to existing semantic sorting algorithms and introduce a semantic-aware external merge sort algorithm. Our extensive evaluation reveals that no single implementation offers universal optimality on all datasets. From our evaluations, we observe a general test-time scaling relationship between sorting cost and the ordering quality for comparison-based algorithms. Building on these insights, we design a budget-aware optimizer that utilizes heuristic rules, LLM-as-Judge evaluation, and consensus aggregation to dynamically select the near-optimal access path for LLM ORDER BY. In our extensive evaluations, our optimizer consistently achieves ranking accuracy on par with or superior to the best static methods across all benchmarks. We believe that this work provides foundational insights into the principled optimization of semantic operators essential for building robust, large-scale LLM-powered analytic systems.

cs.DB↗

arXiv2025

HashEvict: A Pre-Attention KV Cache Eviction Strategy using Locality-Sensitive Hashing

Transformer-based large language models (LLMs) use the key-value (KV) cache to significantly accelerate inference by storing the key and value embeddings of past tokens. However, this cache consumes significant GPU memory. In this work, we introduce HashEvict, an algorithm that uses locality-sensitive hashing (LSH) to compress the KV cache. HashEvict quickly locates tokens in the cache that are cosine dissimilar to the current query token. This is achieved by computing the Hamming distance between binarized Gaussian projections of the current token query and cached token keys, with a projection length much smaller than the embedding dimension. We maintain a lightweight binary structure in GPU memory to facilitate these calculations. Unlike existing compression strategies that compute attention to determine token retention, HashEvict makes these decisions pre-attention, thereby reducing computational costs. Additionally, HashEvict is dynamic - at every decoding step, the key and value of the current token replace the embeddings of a token expected to produce the lowest attention score. We demonstrate that HashEvict can compress the KV cache by 30%-70% while maintaining high performance across reasoning, multiple-choice, long-context retrieval and summarization tasks.

cs.LG↗

arXiv2024

Balancing Label Imbalance in Federated Environments Using Only Mixup and Artificially-Labeled Noise

Clients in a distributed or federated environment will often hold data skewed towards differing subsets of labels. This scenario, referred to as heterogeneous or non-iid federated learning, has been shown to significantly hinder model training and performance. In this work, we explore the limits of a simple yet effective augmentation strategy for balancing skewed label distributions: filling in underrepresented samples of a particular label class using pseudo-images. While existing algorithms exclusively train on pseudo-images such as mixups of local training data, our augmented client datasets consist of both real and pseudo-images. In further contrast to other literature, we (1) use a DP-Instahide variant to reduce the decodability of our image encodings and (2) as a twist, supplement local data using artificially labeled, training-free 'natural noise' generated by an untrained StyleGAN. These noisy images mimic the power spectra patterns present in natural scenes which, together with mixup images, help homogenize label distribution among clients. We demonstrate that small amounts of augmentation via mixups and natural noise markedly improve label-skewed CIFAR-10 and MNIST training.

cs.LG↗

arXiv2024

A Linear Time and Space Local Point Cloud Geometry Encoder via Vectorized Kernel Mixture (VecKM)

We propose VecKM, a local point cloud geometry encoder that is descriptive and efficient to compute. VecKM leverages a unique approach by vectorizing a kernel mixture to represent the local point cloud. Such representation's descriptiveness is supported by two theorems that validate its ability to reconstruct and preserve the similarity of the local shape. Unlike existing encoders downsampling the local point cloud, VecKM constructs the local geometry encoding using all neighboring points, producing a more descriptive encoding. Moreover, VecKM is efficient to compute and scalable to large point cloud inputs: VecKM reduces the memory cost from $(n^2+nKd)$ to $(nd+np)$; and reduces the major runtime cost from computing $nK$ MLPs to $n$ MLPs, where $n$ is the size of the point cloud, $K$ is the neighborhood size, $d$ is the encoding dimension, and $p$ is a marginal factor. The efficiency is due to VecKM's unique factorizable property that eliminates the need of explicitly grouping points into neighbors. In the normal estimation task, VecKM demonstrates not only 100x faster inference speed but also highest accuracy and strongest robustness. In classification and segmentation tasks, integrating VecKM as a preprocessing module achieves consistently better performance than the PointNet, PointNet++, and point transformer baselines, and runs consistently faster by up to 10 times.

cs.CV↗

arXiv2024

Sketch-GNN: Scalable Graph Neural Networks with Sublinear Training Complexity

Graph Neural Networks (GNNs) are widely applied to graph learning problems such as node classification. When scaling up the underlying graphs of GNNs to a larger size, we are forced to either train on the complete graph and keep the full graph adjacency and node embeddings in memory (which is often infeasible) or mini-batch sample the graph (which results in exponentially growing computational complexities with respect to the number of GNN layers). Various sampling-based and historical-embedding-based methods are proposed to avoid this exponential growth of complexities. However, none of these solutions eliminates the linear dependence on graph size. This paper proposes a sketch-based algorithm whose training time and memory grow sublinearly with respect to graph size by training GNNs atop a few compact sketches of graph adjacency and node embeddings. Based on polynomial tensor-sketch (PTS) theory, our framework provides a novel protocol for sketching non-linear activations and graph convolution matrices in GNNs, as opposed to existing methods that sketch linear weights or gradients in neural networks. In addition, we develop a locality-sensitive hashing (LSH) technique that can be trained to improve the quality of sketches. Experiments on large-graph benchmarks demonstrate the scalability and competitive performance of our Sketch-GNNs versus their full-size GNN counterparts.

cs.LG↗

arXiv2024

WAVES: Benchmarking the Robustness of Image Watermarks

In the burgeoning age of generative AI, watermarks act as identifiers of provenance and artificial content. We present WAVES (Watermark Analysis Via Enhanced Stress-testing), a benchmark for assessing image watermark robustness, overcoming the limitations of current evaluation methods. WAVES integrates detection and identification tasks and establishes a standardized evaluation protocol comprised of a diverse range of stress tests. The attacks in WAVES range from traditional image distortions to advanced, novel variations of diffusive, and adversarial attacks. Our evaluation examines two pivotal dimensions: the degree of image quality degradation and the efficacy of watermark detection after attacks. Our novel, comprehensive evaluation reveals previously undetected vulnerabilities of several modern watermarking algorithms. We envision WAVES as a toolkit for the future development of robust watermarks. The project is available at https://wavesbench.github.io/

cs.CV↗

arXiv2024

Calibrated Dataset Condensation for Faster Hyperparameter Search

Dataset condensation can be used to reduce the computational cost of training multiple models on a large dataset by condensing the training dataset into a small synthetic set. State-of-the-art approaches rely on matching the model gradients between the real and synthetic data. However, there is no theoretical guarantee of the generalizability of the condensed data: data condensation often generalizes poorly across hyperparameters/architectures in practice. This paper considers a different condensation objective specifically geared toward hyperparameter search. We aim to generate a synthetic validation dataset so that the validation-performance rankings of the models, with different hyperparameters, on the condensed and original datasets are comparable. We propose a novel hyperparameter-calibrated dataset condensation (HCDC) algorithm, which obtains the synthetic validation dataset by matching the hyperparameter gradients computed via implicit differentiation and efficient inverse Hessian approximation. Experiments demonstrate that the proposed framework effectively maintains the validation-performance rankings of models and speeds up hyperparameter/architecture search for tasks on both images and graphs.

cs.LG↗

arXiv2024

conv_einsum: A Framework for Representation and Fast Evaluation of Multilinear Operations in Convolutional Tensorial Neural Networks

Modern ConvNets continue to achieve state-of-the-art results over a vast array of vision and image classification tasks, but at the cost of increasing parameters. One strategy for compactifying a network without sacrificing much expressive power is to reshape it into a tensorial neural network (TNN), which is a higher-order tensorization of its layers, followed by a factorization, such as a CP-decomposition, which strips a weight down to its critical basis components. Passes through TNNs can be represented as sequences of multilinear operations (MLOs), where the evaluation path can greatly affect the number of floating point operations (FLOPs) incurred. While functions such as the popular einsum can evaluate simple MLOs such as contractions, existing implementations cannot process multi-way convolutions, resulting in scant assessments of how optimal evaluation paths through tensorized convolutional layers can improve training speed. In this paper, we develop a unifying framework for representing tensorial convolution layers as einsum-like strings and a meta-algorithm conv_einsum which is able to evaluate these strings in a FLOPs-minimizing manner. Comprehensive experiments, using our open-source implementation, over a wide range of models, tensor decompositions, and diverse tasks, demonstrate that conv_einsum significantly increases both computational and memory-efficiency of convolutional TNNs.

cs.LG↗

arXiv2023

Comfetch: Federated Learning of Large Networks on Constrained Clients via Sketching

Federated learning (FL) is a popular paradigm for private and collaborative model training on the edge. In centralized FL, the parameters of a global architecture (such as a deep neural network) are maintained and distributed by a central server/controller to clients who transmit model updates (gradients) back to the server based on local optimization. While many efforts have focused on reducing the communication complexity of gradient transmission, the vast majority of compression-based algorithms assume that each participating client is able to download and train the current and full set of parameters, which may not be a practical assumption depending on the resource constraints of smaller clients such as mobile devices. In this work, we propose a simple yet effective novel algorithm, Comfetch, which allows clients to train large networks using reduced representations of the global architecture via the count sketch, which reduces local computational and memory costs along with bi-directional communication complexity. We provide a nonconvex convergence guarantee and experimentally demonstrate that it is possible to learn large models, such as a deep convolutional network, through federated training on their sketched counterparts. The resulting global models exhibit competitive test accuracy over CIFAR10/100 classification when compared against un-compressed model training.

cs.LG↗