arXiv subjects

Kaoutar El Maghraoui

Publications and source records attributed to Kaoutar El Maghraoui.

At least 19 recordsLinked to original sources

arXiv2026

Router Sensitivity Under Lightweight Fine-Tuning Identifies Prunable Experts in Mixture-of-Experts Models

Mixture-of-Experts (MoE) models decouple parameter count from per-token compute, but deployment still requires hosting every expert in memory. Recent theory shows that experts whose router weights change least during fine-tuning can be pruned with provable accuracy preservation, yet the guarantee assumes full fine-tuning. We show that the signal can be elicited through a brief parameter-efficient adaptation. We fine-tune with a lightweight adapter, rank experts by the induced router change, and prune the least-changed experts in one shot. On Mixtral-8$\times$7B-Instruct, router-only LoRA trains 0.002% of parameters and retains 27.54% MMLU-Pro accuracy with half the experts removed, against roughly 16% for magnitude and random pruning. Signal quality improves monotonically with adapter size, reaching 28.76%, and declines as adaptation spreads beyond the router. Under their shared budget, IA3 reaches 28.04% while Houlsby reaches 25.39%. The criterion transfers to Qwen1.5-MoE fine-tuned for mathematical reasoning, retaining 49.7% mean accuracy over eleven benchmarks with half the experts removed. Structural pruning reduces memory by 49% and per-token latency by 37%. Lightweight router sensitivity therefore makes provably motivated, task-conditioned expert pruning practical at scale.

cs.LG↗

arXiv2026

Post-Grokking Collapse at the Representation-Readout Interface in Muon-Trained Transformers

Muon-trained modular-arithmetic transformers can lose accuracy while retaining linearly decodable task information. Adjacent swaps localize five captured unnormalized failures to AdamW readout updates. Multiplying the actual readout displacement by the large feature mean produces a class-dependent logit offset shared across inputs that nearly reproduces each failure. Training-only decoders recover 98.20-100% held-out accuracy. Correcting cross-entropy derivative errors stabilizes five matched branches through step 100,000; four prospective accurate-CE RMS runs fail through embedding updates.

cs.AI↗

arXiv2026

MaskCoFT: Masked Co-Adaptive Fine-Tuning for Memory-Efficient MoE Inference

Mixture-of-experts (MoE) language models often exceed the memory of a single GPU. Expert offloading keeps most experts in host memory and loads them on demand, so decoding speed depends on how many experts each token must fetch. Caching and prefetching reduce this cost only as far as the routing allows. Router-only fine-tuning can reshape the routing to reuse experts, but it keeps the experts frozen, so they cannot adapt to the tokens the new routing sends them. We propose MaskCoFT, a masked co-adaptive fine-tuning method that trains routers and experts together with the cross-entropy loss alone. During fine-tuning, a learnable binary mask restricts the Top-K routing of each layer to a subset of experts, and the experts adapt to the tokens redirected to them. At inference, the learned mask becomes a soft prior that re-ranks experts, so every expert remains selectable. We simulate a GPU cache of 4 experts per layer for Mixtral-8x7B and 12 for DeepSeek-V2-Lite. MaskCoFT cuts expert fetches per token by 23.7% and 10.1% relative to the base model. In real offloading system serving, it lowers the time per output token by up to 16.4% and 5.5%, respectively. Its average accuracy over nine benchmarks stays above the base model by 0.92 and 0.53 points.

cs.LG↗

arXiv2026

PHMForge: Evaluating LLM Agents on Industrial Prognostics through MCP-Native, Algorithm-Grounded Tools

LLM agents are beginning to invoke industrial asset-management tools through the Model Context Protocol (MCP), yet whether they can act reliably on this substrate for safety-critical \emph{Prognostics and Health Management (PHM)} is unanswered. Prior benchmarks conflate protocol fluency with reasoning, instrumentation failures with agent failures, and tool use with tool retrieval. We introduce \textbf{PHMForge}, an evaluation environment that closes each conflation. PHMForge ships 99 SME-authored scenarios across eight industrial asset classes spanning rotating equipment, aero-engines, and lithium-ion cells, on public datasets including NASA PCoE, served through 39 MCP-native tools wrapping published PHM algorithms (C-MAPSS, ISO~10816, Arrhenius capacity-fade models, time-series foundation models). Krippendorff's $α\in [0.74,\,0.82]$ on a 30-scenario stratified rotating-equipment/aero-engine sample; the battery extension is single-rater. Across three agentic frameworks and six LLM backbones, the strongest configuration reaches \textbf{80.8\% pass@1}, with the residual gap concentrated in orchestration and tool-sequencing errors. Crucially, an architectural ablation shows that replacing MCP execution with text-based Retrieval-Augmented Generation (RAG) over telemetry-equivalent evidence collapses Remaining Useful Life \emph{pass-all-3} from \textbf{100\% to 20\%} (5/5 vs.\ 1/5) on the battery class, exposing the structural limits of static retrieval for prognostic computation. Trajectory decomposition shows orchestration errors dominate failures across backbones, while schema-invalid tool calls concentrate in smaller open-weight models. Frontier LLMs are stronger at calling tools than at planning when to call them. PHMForge is open-sourced with deterministic evaluators, a public leaderboard, and a datasheet.

cs.AI↗

arXiv2026

Beyond Static Leaderboards: Predictive Validity for the Evaluation of LLM Agents

Agent benchmarks are growing fast, but no single benchmark touches more than four or five of the dimensions that deployment exposes. This paper aggregates the largest coordinated deep-dive of one MCP-based industrial-agent benchmark to date: fourteen parallel implementation studies covering new asset classes (including a multi-modal visual extension), alternative orchestrations, retrieval strategies, reasoning modes, infrastructure optimizations, and evaluation-methodology probes. Consolidating those studies with seven prior agent benchmarks, we argue that aggregate-score leaderboards systematically underspecify deployed-agent evaluation. Rankings derived from aggregate scores do not transfer to out-of-distribution settings; recent public-to-hidden competition retrospectives provide direct empirical evidence of this rank instability. We propose ranking configurations by predictive validity, the correlation between in-sample and out-of-sample rank, rather than in-sample mean, and report a twelve-tier measurement apparatus that exposes the deployment-relevant dimensions HELM and its agent-era successors collapse. The position is operationalized through three falsifiable out-of-distribution criteria with explicit thresholds; existing evidence partly supports it but is too thin to confirm. We close with a pre-registered pilot design and a field-level vision for what the next generation of agentic benchmarks should report.

cs.AI↗

arXiv2026

Synthetic Scenario Generation for Evaluation of Industry 4.0 Agents

Industrial agent benchmarks require realistic evaluation scenarios that integrate telemetry, failure modes, maintenance records, and domain standards. However, existing benchmarks such as AssetOpsBench rely on manually authored scenarios and cover a limited set of asset classes. We extend AssetOpsBench with a Smart Grid Transformer asset class and four IEC-grounded diagnostic tools for health-index prediction, dissolved-gas analysis, winding-temperature assessment, and load-profile assessment. We further introduce ScenarioGeneratorAgent, a pipeline for synthetic industrial-agent scenario generation. The pipeline constructs evidence-grounded asset profiles, allocates coverage-aware scenario budgets across operational domains, and generates candidates through a hybrid validation-and-repair loop that enforces schema validity, tool reachability, physical plausibility, standards alignment, and deduplication. To improve scalability, we apply two-level caching, parallel focus-group generation, thread-pool offloading, batched LLM calls, and early rejection filtering. On Smart Grid Transformer scenario generation, these optimizations reduce end-to-end runtime by $8\times$ for 50 scenarios while preserving quality, achieving a composite quality score of $74.2 \pm 1.9$ compared with $73.8 \pm 3.0$ for the unoptimized baseline. These results show that standards-grounded synthetic scenario generation can efficiently expand industrial-agent benchmarks without sacrificing scenario quality.

cs.AI↗

arXiv2026

Internalizing Tool Knowledge in Small Language Models via QLoRA Fine-Tuning

Large language models are increasingly used as planning components in agentic systems, but current tool-use pipelines often require full tool schemas to be included in every prompt, creating substantial token overhead and limiting the practicality of smaller models. This paper investigates whether tool-use knowledge can be internalized into small language models through parameter-efficient fine-tuning, enabling structured planning without explicit tool descriptions at inference time. Using AssetOpsBench as the primary benchmark, we fine-tune Gemma 4 E4B and Qwen3-4B with 8-bit QLoRA on approximately 1,700 tool-use examples spanning tool knowledge, question-to-plan mappings, and execution-style traces. We evaluate the resulting models under description-free inference, where the prompt omits the tool catalog entirely. The fine-tuned models outperform an informed unfine-tuned baseline that receives full tool descriptions, reducing input length by 82.6\% while improving structural and LLM-judge planning scores. In the best Gemma run, the model achieves an AT-F1 of 0.65 and an overall judge score of 3.88, compared with 0.47 and 2.88 for the informed baseline. Qwen3-4B achieves a strong overall judge score of 3.78 while using 62\% less memory and running 2.5$\times$ faster than Gemma, though it also exhibits greater catastrophic forgetting on general multiple-choice benchmarks. Additional ablations show that LoRA rank controls a quality--retention trade-off, with $r=32$ maximizing planning quality and smaller ranks preserving more general knowledge. These results suggest that, for fixed tool catalogs, QLoRA fine-tuning can shift tool knowledge from prompt context into model weights, substantially reducing inference overhead while maintaining or improving tool-planning quality.

cs.CL↗

arXiv2026

Evaluating Temporal Semantic Caching and Workflow Optimization in Agentic Plan-Execute Pipelines

Industrial asset operations workflows are latency-sensitive because a single user query may require coordination over sensor data, work orders, failure modes, forecasting tools, and domain-specific agents. We evaluate this problem on AssetOpsBench (AOB), an industrial agent benchmark whose plan-execute pipeline exposes repeated overhead from tool discovery, LLM planning, MCP tool execution, and final summarization. Existing LLM caching techniques such as KV-cache reuse and embedding-based semantic caching were designed for chatbot serving and break down when output validity depends on time, asset, or sensor parameters. We propose two complementary optimization layers for AOB plan-execute pipelines: a temporal semantic cache and a set of MCP workflow optimizations combining disk-backed tool-discovery caching and dependency-aware parallel step execution. MCP workflow optimizations corresponded to a 1.67x speedup and reduced median end-to-end latency by about 40.0% while the temporal-cache benchmark achieved a median of 30.6x speedup on cache hits. Beyond the speedup, our results expose a concrete failure mode of pure semantic caching for parameter-rich industrial queries, providing a critical analysis of how caching choices interact with evaluation correctness in MCP-backed agent benchmarks.

cs.AI↗

arXiv2026

On the Convergence Theory of Pipeline Gradient-based Analog In-memory Training

Aiming to accelerate the training of large deep neural networks (DNN) in an energy-efficient way, analog in-memory computing (AIMC) emerges as a solution with immense potential. AIMC accelerator keeps model weights in memory without moving them from memory to processors during training, reducing overhead dramatically. Despite its efficiency, scaling up AIMC systems presents significant challenges. Since weight copying is expensive and inaccurate, data parallelism is less efficient on AIMC accelerators. It necessitates the exploration of pipeline parallelism, particularly asynchronous pipeline parallelism, which utilizes all available accelerators during the training process. This paper examines the convergence theory of stochastic gradient descent on AIMC hardware with an asynchronous pipeline (Analog-SGD-AP). Although there is empirical exploration of AIMC accelerators, the theoretical understanding of how analog hardware imperfections in weight updates affect the training of multi-layer DNN models remains underexplored. Furthermore, the asynchronous pipeline parallelism results in stale weights issues, which render the update signals no longer valid gradients. To close the gap, this paper investigates the convergence properties of Analog-SGD-AP on multi-layer DNN training. We show that the Analog-SGD-AP converges with iteration complexity $O(\varepsilon^{-2}+\varepsilon^{-1})$ despite the aforementioned issues, which matches the complexities of digital SGD and Analog SGD with synchronous pipeline, except the non-dominant term $O(\varepsilon^{-1})$. It implies that AIMC training benefits from asynchronous pipelining almost for free compared with the synchronous pipeline by overlapping computation.

cs.LG↗

arXiv2026

Efficient Quantization of Mixture-of-Experts with Theoretical Generalization Guarantees

Sparse Mixture-of-Experts (MoE) allows scaling of language and vision models efficiently by activating only a small subset of experts per input. While this reduces computation, the large number of parameters still incurs substantial memory overhead during inference. Post-training quantization has been explored to address this issue. Because uniform quantization suffers from significant accuracy loss at low bit-widths, mixed-precision methods have been recently explored; however, they often require substantial computation for bit-width allocation and overlook the varying sensitivity of model performance to the quantization of different experts. We propose a theoretically grounded expert-wise mixed precision strategy that assigns bit-width to each expert primarily based on their change in routers l2 norm during training. Experts with smaller changes are shown to capture less frequent but critical features, and model performance is more sensitive to the quantization of these experts, thus requiring higher precision. Furthermore, to avoid allocating experts to lower precision that inject high quantization noise, experts with large maximum intra-neuron variance are also allocated higher precision. Experiments on large-scale MoE models, including Switch Transformer and Mixtral, show that our method achieves higher accuracy than existing approaches, while also reducing inference cost and incurring only negligible overhead for bit-width assignment.

cs.LG↗

arXiv2026

Robust Heterogeneous Analog-Digital Computing for Mixture-of-Experts Models with Theoretical Generalization Guarantees

Sparse Mixture-of-Experts (MoE) models enable efficient scalability by activating only a small sub-set of experts per input, yet their massive parameter counts lead to substantial memory and energy inefficiency during inference. Analog in-memory computing (AIMC) offers a promising solution by eliminating frequent data movement between memory and compute units. However, mitigating hardware nonidealities of AIMC typically requires noise-aware retraining, which is infeasible for large MoE models. In this paper, we propose a retraining-free heterogeneous computation framework in which noise-sensitive experts, which are provably identifiable by their maximum neuron norm, are computed digitally while the majority of the experts are executed on AIMC hardware. We further assign densely activated modules, such as attention layers, to digital computation due to their high noise sensitivity despite comprising a small fraction of parameters. Extensive experiments on large MoE language models, including DeepSeekMoE and OLMoE, across multiple benchmark tasks validate the robustness of our approach in maintaining accuracy under analog nonidealities.

cs.LG↗

arXiv2025

Context-Aware Mixture-of-Experts Inference on CXL-Enabled GPU-NDP Systems

Mixture-of-Experts (MoE) models scale large language models through conditional computation, but inference becomes memory-bound once expert weights exceed the capacity of GPU memory. In this case, weights must be offloaded to external memory, and fetching them incurs costly and repeated transfers. We address this by adopting CXL-attached near-data processing (CXL-NDP) as the offloading tier to execute cold experts in place, converting expensive parameter movement into cheaper activation movement. Unlike prior GPU-NDP systems that are largely context-agnostic and reactive, we develop a context-aware MoE system that uses prefill-stage activation statistics to guide decoding-stage expert placement, dynamically pins hot experts in GPU-side HBM, and maps the remainder to CXL-NDP. To meet NDP's limited compute throughput, we introduce context-aware mixed-precision quantization that allocates per-expert bitwidths (1-4 bit) based on prefill stage. The resulting MoE inference system overlaps GPU and NDP execution while minimizing cross-device movement. The evaluation on the GPU-NDP system shows that our approach achieves up to an 8.7-fold decoding throughput improvement over the state-of-the-art method, while incurring only a 0.13% average accuracy drop.

cs.LG↗

arXiv2025

SparseST: Exploiting Data Sparsity in Spatiotemporal Modeling and Prediction

Spatiotemporal data mining (STDM) has a wide range of applications in various complex physical systems (CPS), i.e., transportation, manufacturing, healthcare, etc. Among all the proposed methods, the Convolutional Long Short-Term Memory (ConvLSTM) has proved to be generalizable and extendable in different applications and has multiple variants achieving state-of-the-art performance in various STDM applications. However, ConvLSTM and its variants are computationally expensive, which makes them inapplicable in edge devices with limited computational resources. With the emerging need for edge computing in CPS, efficient AI is essential to reduce the computational cost while preserving the model performance. Common methods of efficient AI are developed to reduce redundancy in model capacity (i.e., model pruning, compression, etc.). However, spatiotemporal data mining naturally requires extensive model capacity, as the embedded dependencies in spatiotemporal data are complex and hard to capture, which limits the model redundancy. Instead, there is a fairly high level of data and feature redundancy that introduces an unnecessary computational burden, which has been largely overlooked in existing research. Therefore, we developed a novel framework SparseST, that pioneered in exploiting data sparsity to develop an efficient spatiotemporal model. In addition, we explore and approximate the Pareto front between model performance and computational efficiency by designing a multi-objective composite loss function, which provides a practical guide for practitioners to adjust the model according to computational resource constraints and the performance requirements of downstream tasks.

cs.LG↗

arXiv2025

Analog Foundation Models

Analog in-memory computing (AIMC) is a promising compute paradigm to improve speed and power efficiency of neural network inference beyond the limits of conventional von Neumann-based architectures. However, AIMC introduces fundamental challenges such as noisy computations and strict constraints on input and output quantization. Because of these constraints and imprecisions, off-the-shelf LLMs are not able to achieve 4-bit-level performance when deployed on AIMC-based hardware. While researchers previously investigated recovering this accuracy gap on small, mostly vision-based models, a generic method applicable to LLMs pre-trained on trillions of tokens does not yet exist. In this work, we introduce a general and scalable method to robustly adapt LLMs for execution on noisy, low-precision analog hardware. Our approach enables state-of-the-art models $\unicode{x2013}$ including Phi-3-mini-4k-instruct and Llama-3.2-1B-Instruct $\unicode{x2013}$ to retain performance comparable to 4-bit weight, 8-bit activation baselines, despite the presence of analog noise and quantization constraints. Additionally, we show that as a byproduct of our training methodology, analog foundation models can be quantized for inference on low-precision digital hardware. Finally, we show that our models also benefit from test-time compute scaling, showing better scaling behavior than models trained with 4-bit weight and 8-bit static input quantization. Our work bridges the gap between high-capacity LLMs and efficient analog hardware, offering a path toward energy-efficient foundation models. Code is available at https://github.com/IBM/analog-foundation-models.

cs.LG↗

arXiv2025

Accelerating LLM Inference via Dynamic KV Cache Placement in Heterogeneous Memory System

Large Language Model (LLM) inference is increasingly constrained by memory bandwidth, with frequent access to the key-value (KV) cache dominating data movement. While attention sparsity reduces some memory traffic, the relevance of past tokens varies over time, requiring the full KV cache to remain accessible and sustaining pressure on both bandwidth and capacity. With advances in interconnects such as NVLink and LPDDR5X, modern AI hardware now integrates high-bandwidth memory (HBM) with high-speed off-package DRAM, making heterogeneous memory systems a practical solution. This work investigates dynamic KV cache placement across such systems to maximize aggregated bandwidth utilization under capacity constraints. Rather than proposing a specific scheduling policy, we formulate the placement problem mathematically and derive a theoretical upper bound, revealing substantial headroom for runtime optimization. To our knowledge, this is the first formal treatment of dynamic KV cache scheduling in heterogeneous memory systems for LLM inference.

cs.AR↗

arXiv2025

Paged Attention Meets FlexAttention: Unlocking Long-Context Efficiency in Deployed Inference

Large Language Models (LLMs) encounter severe memory inefficiencies during long-context inference due to conventional handling of key-value (KV) caches. In this work, we introduce a novel integration of PagedAttention with PyTorch's FlexAttention, addressing internal fragmentation and inefficiencies associated with monolithic KV cache allocations. Implemented within IBM's Foundation Model Stack (FMS), our fused attention kernel efficiently gathers scattered KV data. Our benchmarks on an NVIDIA L4 GPU (24GB) demonstrate significantly reduced inference latency, growing only linearly (~2x) with sequence length from 128 to 2048 tokens when utilizing a global KV cache, compared to exponential latency increases without caching. While peak memory usage remains largely unchanged for single-step evaluations (dominated by model weights and activations), paged attention causes minimal incremental memory usage, observable only at sequence lengths exceeding 2048 tokens due to its power-of-two cache allocations. We open-source the full implementation and discuss its implications for future long-context model deployment.

cs.LG↗

arXiv2025

EvidenceMoE: A Physics-Guided Mixture-of-Experts with Evidential Critics for Advancing Fluorescence Light Detection and Ranging in Scattering Media

Fluorescence LiDAR (FLiDAR), a Light Detection and Ranging (LiDAR) technology employed for distance and depth estimation across medical, automotive, and other fields, encounters significant computational challenges in scattering media. The complex nature of the acquired FLiDAR signal, particularly in such environments, makes isolating photon time-of-flight (related to target depth) and intrinsic fluorescence lifetime exceptionally difficult, thus limiting the effectiveness of current analytical and computational methodologies. To overcome this limitation, we present a Physics-Guided Mixture-of-Experts (MoE) framework tailored for specialized modeling of diverse temporal components. In contrast to the conventional MoE approaches our expert models are informed by underlying physics, such as the radiative transport equation governing photon propagation in scattering media. Central to our approach is EvidenceMoE, which integrates Evidence-Based Dirichlet Critics (EDCs). These critic models assess the reliability of each expert's output by providing per-expert quality scores and corrective feedback. A Decider Network then leverages this information to fuse expert predictions into a robust final estimate adaptively. We validate our method using realistically simulated Fluorescence LiDAR (FLiDAR) data for non-invasive cancer cell depth detection generated from photon transport models in tissue. Our framework demonstrates strong performance, achieving a normalized root mean squared error (NRMSE) of 0.030 for depth estimation and 0.074 for fluorescence lifetime.

cs.CV↗

arXiv2024

A Provably Effective Method for Pruning Experts in Fine-tuned Sparse Mixture-of-Experts

The sparsely gated mixture of experts (MoE) architecture sends different inputs to different subnetworks, i.e., experts, through trainable routers. MoE reduces the training computation significantly for large models, but its deployment can be still memory or computation expensive for some downstream tasks. Model pruning is a popular approach to reduce inference computation, but its application in MoE architecture is largely unexplored. To the best of our knowledge, this paper provides the first provably efficient technique for pruning experts in finetuned MoE models. We theoretically prove that prioritizing the pruning of the experts with a smaller change of the routers l2 norm from the pretrained model guarantees the preservation of test accuracy, while significantly reducing the model size and the computational requirements. Although our theoretical analysis is centered on binary classification tasks on simplified MoE architecture, our expert pruning method is verified on large vision MoE models such as VMoE and E3MoE finetuned on benchmark datasets such as CIFAR10, CIFAR100, and ImageNet.

cs.LG↗