Searcharxiv⌕ Search

arXiv subjects

Amr Mohamed

Publications and source records attributed to Amr Mohamed.

At least 19 recordsLinked to original sources

K2-Horizon: A Frontier Fleet of Open Source Language Models

We present K2-Horizon, a fleet of six fully open language models, from a 0.9B dense model to a 375B-A23B MoE, built for general-purpose reasoning and agentic work across math and science, software engineering, deep research, office automation, and beyond. The models lead or are competitive with strong open models of their size on agentic and coding benchmarks. On the modeling side, to strengthen reasoning, we introduce Mixture-of-Value Attention (MoVA), applying MoE to attention value projections, and softplus attention gating, a more flexible, theoretically motivated gate. Our 36B-A4B model uses both and nearly matches a 32B dense model trained on the same data, with one-seventh the active parameters. We also introduce Uno, lossless diffusion adapters that speed up decoding by up to 2.2x without a separate draft model. Our release follows the LLM360 definition of open source for transparency and reproducibility: the full development history, including intermediate checkpoints from every training stage, pre-training data, training logs, and all training, evaluation, and infrastructure code, notably our production-scale xLLM framework for dense and MoE training and the RL360 reinforcement-learning stack. Beyond the data itself, we release and document in detail every pipeline that built it, the part of model development that open releases most rarely disclose. These pipelines produced roughly 10T synthetic pre-training tokens, showing that synthetic data can be a major ingredient of pre-training at scale. Four of the models share one 22T-token pre-training corpus and recipe, revealing consistent scaling behavior and enabling controlled studies of scale and sparsity. All artifacts are available at https://ifm.ai/k2/

cs.AI↗

Jais 3: Introducing Arabic-centric Agentic AI

We present Jais 3, a fleet of six open, Arabic-centric agentic language models ranging from 0.9B to 375B parameters. Jais 3 is built to complete tasks for Arabic speakers, not only to answer questions in plain text: it is trained to plan multi-step tasks, call tools and Model Context Protocol (MCP) servers, operate a terminal, search the web, and read, edit, and produce PDF, spreadsheet, and presentation files, while communicating with the user in Arabic. Instead of pretraining from scratch, we adapt each model of the fully open K2 Horizon fleet to Arabic through continual pretraining, followed by supervised fine-tuning (SFT), final token preference optimization (FTPO), and reinforcement learning with verifiable rewards (RLVR). We combine natural Arabic text with translated, transliterated, and synthetic Arabic, and replay the base model's own training data to keep its reasoning, agentic, and English abilities. Jais 3 covers Modern Standard Arabic and the major regional dialects, follows instructions with Arabic-specific constraints, and inherits the 512K-token context window of K2 Horizon (all models except the 0.9B). To our knowledge, its 375B model is the largest open Arabic-centric language model to date. For evaluation, we revisit how Arabic models are assessed, moving to fully generative evaluation on audited Arabic benchmarks. Across model sizes, Jais 3 matches or outperforms open-weight models of comparable size on Arabic and agentic benchmarks, with particularly strong agentic capabilities and Arabic instruction following, and far fewer hallucinations. We release the weights and the checkpoints from every training stage of all six models, together with samples of the training data, the data recipes and training configurations, the training logs, the audited benchmarks, and the evaluation code.

cs.CL↗

Markovian Generation Chains in Large Language Models

The widespread use of large language models (LLMs) raises an important question: how do texts evolve when they are repeatedly processed by LLMs? In this paper, we define this iterative inference process as Markovian generation chains, where each step takes a specific prompt template and the previous output as input, without including any prior data. In iterative rephrasing and round-trip translation experiments, the output either converges to a small recurrent set or continues to produce novel sentences within a finite horizon. Through sentence-level Markov chain modeling and analysis of simulated data, we show that iterative process can either increase or reduce the diversity depending on factors such as the temperature parameter and the initial input sentence. These results offer valuable insights into the dynamics of iterative LLM inference and their implications for multi-agent LLM systems.

cs.CL↗

Distill Locally, Schedule Globally: Flow Maps for Few-Step Text-to-Speech

Flow-matching text-to-speech (TTS) models achieve high synthesis quality but require many neural function evaluations (NFEs) to integrate their generative trajectories. Recent few-step flow-map distillation approaches for TTS construct targets from numerically integrated teacher trajectories, creating a trade-off between target accuracy and training cost. We propose Local Flow-Map Distillation (LFMD), which adapts Eulerian Map Distillation to conditional TTS and avoids teacher trajectory integration during target construction. For inference, we derive a sampling schedule (TD-DP) from teacher dynamics and consistency of the learned maps, with a single cost graph supporting multiple NFE budgets without external audio-metric evaluation. Because scheduling offers no flexibility at one NFE, we refine this regime with alignment-aware temporal self-distillation using soft-DTW. Across Seed-TTS and LibriSpeech-PC, LFMD improves low-NFE synthesis over a matched integral-distillation baseline. On Seed-TTS, the refined student reaches 1.80% WER with 1-NFE, compared with 1.76% for its 32-NFE teacher.

cs.SD↗

The Analyst in the Prompt: Role, Retrieval, and Memory Biases in LLM Financial Analysis

Large Language Models (LLMs) increasingly use user context such as memory, profiles, and role prompts to personalize their responses. This personalization can affect evidence-based judgment: the same evidence may lead to different conclusions under different user contexts. Finance provides a high-stakes setting to study this problem because decisions often depend on interpreting long and complex documents. We test this using 3,575 SEC filings across twelve LLMs. We compare persona-conditioned retrieval, neutral retrieval, and memory-framed context to separate the effect of evidence selection from the effect of interpretation. We find that most user-context spillover comes from how models interpret the same evidence under different roles, rather than from retrieving different evidence. We then test two simple mitigation strategies: expressing the same investor mindset as a user profile instead of an assistant role, and separating evidence-based and personalized outputs. Both reduce spillover, but neither removes it completely, and their effectiveness varies substantially across models.

cs.CL↗

CARVE: Verified Expansion for Variable-Length Generation in Diffusion Language Models

Masked diffusion language models predict tokens from a partially observed response canvas, enabling bidirectional conditioning and parallel token refinement. Yet standard masked-diffusion decoders use a rigid inference interface: the number of masked positions allocated to the answer is fixed before generation begins. Choosing this length is difficult. A short canvas can truncate reasoning or code, while a long canvas wastes computation and can perturb denoising. We introduce CARVE (Counterfactual-Aware Reveal with Verified Expansion), a training-free variable-length algorithm for masked diffusion LMs. Starting from a shorter canvas, CARVE can grow the response during decoding by inserting additional [MASK] positions. Rather than keeping every insertion, CARVE tests a candidate expanded canvas and asks a counterfactual question: would the model make similar predictions for the unresolved positions in the original canvas if the extra masked space were present? The inserted masks are kept only when they induce low Jensen-Shannon (JS) divergence on aligned unresolved positions. This makes length growth a verified stability decision rather than a pure confidence heuristic. CARVE applies without retraining to both full-canvas and blockwise diffusion decoders. Across code generation and mathematical reasoning benchmarks, CARVE consistently improves average performance over fixed-length baselines across all evaluated model families. Crucially, CARVE achieves these accuracy gains while reducing inference cost, reaching half the FLOPs of fixed-length decoding in some settings.

cs.AI↗

The Impact of LLM-Assistants on Software Developer Productivity: A Systematic Review and Mapping Study

Large language model assistants (LLM-assistants) present new opportunities to transform software development. Developers are increasingly adopting these tools across tasks, including coding, testing, debugging, documentation, and design. Yet, despite growing interest, there is no synthesis of how LLM-assistants affect software developer productivity. In this paper, we present a systematic review and mapping of 39 peer-reviewed studies published between January 2014 and December 2024 that examine this impact. Our analysis reveals that the majority of studies report considerable benefits from LLM-assistants, though a notable subset identifies critical risks. Commonly reported gains include accelerated development, minimized code search, and the automation of trivial and repetitive tasks. However, studies also highlight concerns around cognitive offloading and reduced team collaboration. Our study reveals that whether LLM-based assistants improve or degrade code quality remains unresolved, as existing studies report contradictory outcomes contingent on context and evaluation criteria. While the majority of studies (90%) adopt a multi-dimensional perspective by examining at least two SPACE dimensions, reflecting increased awareness of the complexity of developer productivity, only 15% extend beyond three dimensions, indicating substantial room for more integrated evaluations. Satisfaction, Performance, and Efficiency are the most frequently investigated dimensions, whereas Communication and Activity remain underexplored. Most studies are exploratory (59%) and methodologically diverse, but lack longitudinal and team-based evaluations. This review surfaces key research gaps and provides recommendations for future research and practice. All artifacts associated with this study are publicly available at https://zenodo.org/records/18489222

cs.SE↗

Communication Security via Temporal Dependency

Communication security has traditionally been built upon one of two external resources: shared secret keys or a communication advantage over the eavesdropper. However, many practical wireless scenarios, including infrastructure-less, emergency, and highly dynamic networks, cannot guarantee either resource, motivating the need for a new communication security principle. This paper introduces a new communication security paradigm that exploits temporal dependency as a security resource. Unlike conventional secrecy techniques that prevent an eavesdropper from recovering transmitted bits, the proposed paradigm allows packet decoding but prevents correct interpretation by making original and dummy packets computationally indistinguishable. Rather than protecting individual transmissions, successive transmissions are intentionally coupled so that future communication depends on correctly interpreting previous ones. As one realization, we develop a state-chained random linear network coding (RLNC) framework in which the synchronization state required to interpret each transmission block is embedded in the previous block. Therefore, synchronization failures propagate across future transmissions, resulting in persistent eavesdropper asynchronization. We analytically characterize the probability and persistence of eavesdropper asynchronization, together with the computational complexity of resynchronization, and develop transmission strategies based on transmit power and intentional-interference optimization. Numerical results demonstrate sub-second eavesdropper asynchronization under a worst- case adversarial model with no channel advantage, no secret assumptions, complete protocol knowledge, and an arbitrarily stronger eavesdropper.

math.OC↗

Latency Decoupling in Low-Feedback Multi-User Networks via Overhearing-Driven NOMA

Conventional wireless protocols such as Hybrid Automatic Repeat Request (HARQ) rely on frequent and timely feedback, which becomes impractical in low-feedback regimes including non-terrestrial networks and massive IoT. This limitation is particularly critical in heterogeneous multi-user systems with unknown and asymmetric channels, where a single weak user can dominate the overall completion latency. We propose Overhearing-driven Non-Orthogonal Multiple Access (ONOMA), a novel cross-layer transmission scheme that minimizes latency without requiring instantaneous or statistical CSI at the transmitter. ONOMA integrates Random Linear Network Coding (RLNC) with symbol-aware NOMA and explicitly exploits overhearing and acknowledgment timing. In the first phase, users overhear RLNC transmissions, and the relative timing of acknowledgments is used to implicitly infer channel strength ordering. In the second phase, symbol reconstruction enables interference-free decoding for strong users, effectively decoupling user latencies. An adaptive power allocation policy is derived from acknowledgment timing-based channel estimates. Analytical and simulation results show that ONOMA outperforms TDMA, multicast, FDMA, inter-session, and classical NOMA, reducing completion time by up to 34% in two-user and 50% in larger asymmetric networks.

cs.IT↗

Jais 2: A Family of Arabic-Centric Open Large Language Models

Jais 2 is a family of Arabic-centric large language models developed jointly by MBZUAI, Cerebras, and Inception, designed to advance Arabic-centric language modeling, with strong performance across the Arabic and culturally grounded benchmarks evaluated in this report. The family includes, to our knowledge, the largest open Arabic-centric LLM trained from scratch at 70B parameters, and a competitive 8B-parameter variant among the evaluated open models. A custom Arabic-centric vocabulary enables efficient training and inference. In addition, an optimized architecture and training recipe yield highly compute-efficient training. With a substantially smaller token budget than comparable models, Jais 2 achieves strong Arabic performance on the benchmarks considered in this report and competitive English results. The models obtain leading results among the evaluated open models on OALL2 and AraGen. They also perform strongly on several culturally grounded Arabic benchmarks, including poetry, religion, cuisine, and dream interpretation, as well as in general tasks such as translation and summarization. We release the models in HuggingFace under a commercially permissive license. Jais 2 70B is also released as a chat app on the Web, iOS, and Android; it runs on Cerebras hardware, delivering up to 2,000 tokens per second, and enabling high-throughput Arabic-centric chat serving in our deployment setting. By uniting scale, linguistic diversity, cultural fidelity, openness, and speed, Jais 2 provides an open-weight foundation intended to support further research and development in Arabic-centric LLMs.

cs.CL↗

LESS Is More: Mutual-Stability Sampling for Diffusion Language Models

Diffusion large language models (dLLMs) offer a promising alternative to autoregressive decoding by iteratively refining masked sequences, enabling parallel token updates and bidirectional conditioning. Their practical efficiency, however, is limited by sampling procedures that execute a fixed number of reverse denoising steps selected before decoding, spending computation on already-stable positions and sometimes committing unstable ones too early. We present \textsc{LESS}, a training-free, model-agnostic adaptive sampler that treats token commitment as an online stopping problem. \textsc{LESS} implements mutual-stability sampling through a joint stability rule that makes a masked position eligible for unmasking only when its top-1 prediction has high confidence, its top-1 token persists across recent reverse steps, and its predictive distribution is stable under top-$K$ inter-step Jensen--Shannon divergence. We evaluate \textsc{LESS} on Dream-7B, LLaDA-8B, and LLaDA-1.5-8B, covering full-sequence diffusion and semi-autoregressive blockwise sampling regimes, across seven benchmarks spanning general knowledge, math, and code. \textsc{LESS} improves average accuracy over strong training-free adaptive samplers while using $72.1\%$ fewer reverse steps than fixed-budget decoding. Since each reverse step requires a Transformer forward pass, these step-count reductions translate into fewer forward evaluations, lower measured wall-clock latency, and lower estimated inference compute.

cs.CL↗

The Masked Advantage: Uncovering Local-Language Access to Cultural Knowledge in LLMs

Large language models are increasingly used to answer culturally grounded questions across languages, yet it remains unclear whether local cultural knowledge is better accessed through English or the local language. Existing evaluations face two key limitations: many rely on parallel template-based questions that may not reflect how cultural knowledge naturally appears, and raw accuracy conflates general language proficiency with language-conditioned knowledge access. We address these issues with a controlled framework built on real-world cultural questions collected from regional benchmarks and local sources. By crossing question type (culture-agnostic vs. culture-specific) with query language (English vs. local language), and estimating ability with a shared 1PL item response theory model, we separate proficiency from localized knowledge access. Across 13 locales and roughly 80 models, we find a consistent English advantage on culture-agnostic questions, indicating stronger English proficiency. However, after accounting for this proficiency gap, local languages show a positive knowledge-access advantage in nearly all locale-model settings. This advantage is often masked in raw accuracy but becomes more visible for frontier, regionally aligned, or language-adapted models. Our results suggest that weaker local-language performance does not necessarily imply weaker cultural knowledge; rather, local cultural knowledge may be more accessible through the local language but hidden by limited language proficiency.

cs.CL↗

SEArch: Optimistic Policy Selection Between Scene Noise and Drift for UAV Radar Search

Unmanned Aerial Vehicles (UAVs) equipped with radar sensors are deployed for target search missions in diverse environments, where targets exhibit characteristic signatures (e.g., respiration micro-motion in human search) detectable through occlusions. A fundamental challenge arises from shifts in radar statistics as the UAV moves through a dynamic and potentially non-stationary environment, rendering any fixed signal-processing strategy suboptimal; yet perception and adaptation must run onboard a resource-constrained aerial node in real time. Since no single detector performs well across all conditions, we adopt a multi-policy paradigm and formulate UAV target search as an online policy selection problem over a library of specialized detectors, with performance measured by regret, the cumulative loss gap relative to the best policy in each scene. The setting couples in-scene stochastic noise with inter-scene shifts. Whereas prior methods capture only one regime, we account for both through the Stochastically Extended Adversary (SEA) framework, without requiring oracle knowledge of scene dynamics. Because adaptation must run at the UAV, we instantiate SEA through \textsc{SEArch}, a lightweight optimistic Follow the Regularized Leader (OFTRL) selector with an adaptive learning rate, achieving regret $O(\barσ_T \sqrt{T} + \sqrt{J})$, where $\barσ_T$ captures radar measurement noise and $J$ is the number of scene transitions over the mission horizon $T$. To enable rapid adaptation under frequent scene changes, we further introduce \textsc{W-SEArch}, a windowed variant that restarts every $w$ rounds and achieves regret $O(\barσ_I \sqrt{w})$ under at most one transition per window. Experiments show up to 30\% regret reduction compared to non-adaptive baselines across a range of non-stationary settings.

cs.NI↗

Do Good, Stay Longer? Temporal Patterns and Predictors of Newcomer-to-Core Transitions in Conventional OSS and OSS4SG

Open Source Software (OSS) sustainability relies on newcomers transitioning to core contributors, but this pipeline is broken, with most newcomers becoming inactive after initial contributions. Open Source Software for Social Good (OSS4SG) projects, which prioritize societal impact as their primary mission, may be associated with different newcomer-to-core transition outcomes than conventional OSS projects. We compared 375 projects (190 OSS4SG, 185 OSS), analyzing 92,721 contributors and 3.5 million commits. OSS4SG projects retain contributors at 2.2X higher rates and contributors have 19.6% higher probability of achieving core status. Early broad project exploration predicts core achievement (22.2% importance); conventional OSS concentrates on one dominant pathway (61.62% of transitions) while OSS4SG provides multiple pathways. Contrary to intuition, contributors who invest time learning the project before intensifying their contributions (Late Spike pattern) achieve core status 2.4-2.9X faster (21 weeks) than those who contribute intensively from day one (Early Spike pattern, 51-60 weeks). OSS4SG supports two effective temporal patterns while only Late Spike achieves fastest time-to-core in conventional OSS. Our findings suggest that finding a project aligned with personal values and taking time to understand the codebase before major contributions are key strategies for achieving core status. Our findings show that project mission is associated with measurably different environments for newcomer-to-core transitions and provide evidence-based guidance for newcomers and maintainers.

cs.SE↗

Beyond Random Sampling: Efficient Language Model Pretraining via Curriculum Learning

Curriculum learning-organizing training data from easy to hard-has improved efficiency across machine learning domains, yet remains underexplored for language model pretraining. We present the first systematic investigation of curriculum learning in LLM pretraining, with over 200 models trained on up to 100B tokens across three strategies: vanilla curriculum learning, pacing-based sampling, and interleaved curricula, guided by six difficulty metrics spanning linguistic and information-theoretic properties. We evaluate performance on eight benchmarks under three realistic scenarios: limited data, unlimited data, and continual training. Our experiments show that curriculum learning consistently accelerates convergence in early and mid-training phases,reducing training steps by $18-45\%$ to reach baseline performance. When applied as a warmup strategy before standard random sampling, curriculum learning yields sustained improvements up to $3.5\%$. We identify compression ratio, lexical diversity (MTLD), and readability (Flesch Reading Ease) as the most effective difficulty signals. Our findings demonstrate that data ordering-orthogonal to existing data selection methods-provides a practical mechanism for more efficient LLM pretraining.

cs.CL↗

IntAgent: NWDAF-Based Intent LLM Agent Towards Advanced Next Generation Networks

Intent-based networks (IBNs) are gaining prominence as an innovative technology that automates network operations through high-level request statements, defining what the network should achieve. In this work, we introduce IntAgent, an intelligent intent LLM agent that integrates NWDAF analytics and tools to fulfill the network operator's intents. Unlike previous approaches, we develop an intent tools engine directly within the NWDAF analytics engine, allowing our agent to utilize live network analytics to inform its reasoning and tool selection. We offer an enriched, 3GPP-compliant data source that enhances the dynamic, context-aware fulfillment of network operator goals, along with an MCP tools server for scheduling, monitoring, and analytics tools. We demonstrate the efficacy of our framework through two practical use cases: ML-based traffic prediction and scheduled policy enforcement, which validate IntAgent's ability to autonomously fulfill complex network intents.

cs.NI↗

Shorter but not Worse: Frugal Reasoning via Easy Samples as Length Regularizers in Math RLVR

Large language models (LLMs) trained for step-by-step reasoning often become excessively verbose, raising inference cost. Standard Reinforcement Learning with Verifiable Rewards (RLVR) pipelines filter out ``easy'' problems for training efficiency, leaving the model to train primarily on harder problems that require longer reasoning chains. This skews the output length distribution upward, resulting in a \textbf{model that conflates ``thinking longer'' with ``thinking better''}. In this work, we show that retaining and modestly up-weighting moderately easy problems acts as an implicit length regularizer. Exposing the model to solvable short-chain tasks constrains its output distribution and prevents runaway verbosity. The result is \textbf{\emph{emergent brevity for free}}: the model learns to solve harder problems without inflating the output length, \textbf{ despite the absence of any explicit length penalization}. RLVR experiments using this approach on \textit{Qwen3-4B-Thinking-2507} (with a 16k token limit) achieve baseline pass@1 AIME25 accuracy while generating solutions that are, on average, nearly twice as short. The code is available at \href{https://github.com/MBZUAI-Paris/Frugal-AI}{GitHub}, with datasets and models on \href{https://huggingface.co/collections/MBZUAI-Paris/k2-think-mini-68dcfa8b114686a4bd3dc2bc}{Hugging Face}.

cs.LG↗

Fast-Decoding Diffusion Language Models via Progress-Aware Confidence Schedules

Diffusion large language models (dLLMs) offer a promising alternative to autoregressive models, but their practical utility is severely hampered by slow, iterative sampling. We present SchED, a training-free, model-agnostic early-exit algorithm that aggregates full-span logit margins and halts decoding once a smooth, progress-dependent confidence threshold is met. We evaluated SchED on two dLLM families (Dream and LLaDA), in base and instruction-tuned variants across ten benchmarks spanning downstream tasks including multiple-choice question answering (MCQ), math, long-form QA/summarization, and translation. SchED delivers large, stable accelerations: on instruction-tuned models, it achieves $3.8$-$4.0\times$ speedups while retaining $99.8$-$100\%$ of the baseline score on average. On base models, SchED yields consistent speedup gains with $99.1$-$100\%$ performance retention, with up to $2.34\times$ under more aggressive settings. Using a conservative speed metric that heavily penalizes quality loss (QPS, $γ{=}4$), we show that SchED is robust and clearly outperforms prior confidence-based early-exit methods, which break down on long-form generation. An entropy analysis of the model's token predictions reveals that instruction tuning speeds up the decay of predictive entropy. By turning genuine confidence stabilization into computational savings, SchED makes dLLM decoding substantially more efficient.

cs.CL↗