model releases

Pass the Baton: Trajectory-Relayed On-Policy Distillation

arXiv:2607.26057v1 Announce Type: cross Abstract: On-policy distillation (OPD) grounds token-level supervision in the student's own trajectory, yet suffers from prefix failure: once the student commit

PATHFinder Agent for Tailored Prenatal Care

DGX agent

arXiv:2607.24768v1 Announce Type: new Abstract: Prenatal care is an important preventive service designed to improve outcomes for pregnant individuals. The American College of Obstetricians and Gyneco

model-releasesarxiv-cs-ai

29 Jul 2026

PatientAgentBench: A Benchmark Framework for Evaluating Patient-Facing Health AI Agents

arXiv:2607.25485v1 Announce Type: new Abstract: Health AI is evolving from answering questions to agentic systems that converse with patients, reason about health records, and act on their behalf. Pri

PEANUT: Perturbations by Eigenvector Alignment for Attacking Graph Neural Networks Under Topology-Driven Message Passing

arXiv:2603.26136v3 Announce Type: replace Abstract: Message Passing Neural Networks (MPNNs) have achieved strong performance on tasks involving relational data. However, small perturbations to graph s

Penelope: Localized Latent Recurrence for Efficient Structured Reasoning

arXiv:2607.25915v1 Announce Type: new Abstract: Complex structured reasoning tasks often require additional computation, yet current language models obtain it mainly by increasing parameter scale or b

PerceptionBench: Evaluating Atomic Visual Perception in Multimodal Large Language Models

arXiv:2607.24957v1 Announce Type: new Abstract: We introduce PerceptionBench, a benchmark specifically designed to evaluate the atomic visual perception capabilities of Multimodal Large Language Model

Personalization, Personas, and Forecasting in Value Alignment

arXiv:2607.24782v1 Announce Type: new Abstract: LLM behavior may be conditioned by human identity in several ways: they may be asked to adapt to users, role-play populations, or forecast how people wo

Physics-Aware End-to-End Deep Reinforcement Learning for Quadcopter Control with Actuator Dynamics

arXiv:2607.25985v1 Announce Type: cross Abstract: Unmanned aerial vehicles (UAVs), particularly quadcopters, present unique challenges for autonomous control due to their underactuated dynamics: only

Physics-Grounded Fluid Video Generation with a Simulation Dataset and Dual-Stream Optical-Flow Supervision

arXiv:2607.25321v1 Announce Type: new Abstract: Video diffusion models generate visually compelling content but routinely violate elementary physics when the subject involves fluids: liquid columns br

PIcsC: Partitioning-Induced Covariate Shift Correction

arXiv:2607.25441v1 Announce Type: new Abstract: Covariate shift across training-data partitions biases model selection and parameter estimation in cross-validation, lifelong learning, and federated le

PilotRL: Training Language Model Agents via Global Planning-Guided Progressive Reinforcement Learning

arXiv:2508.00344v5 Announce Type: replace Abstract: Large Language Models (LLMs) have shown remarkable advancements in tackling agent-oriented tasks. Despite their potential, existing work faces chall

Polistemics: Evaluating LLMs as Information Mediators in Politics & Elections

arXiv:2607.25953v1 Announce Type: new Abstract: As LLMs increasingly mediate the political information citizens rely on, there is still no standardized way to assess whether they do so responsibly. We

Probabilistic Symbolic Regression for Equation Discovery via Operator-induced and Regularized Symbolic Forests

arXiv:2509.19710v2 Announce Type: cross Abstract: Symbolic regression has emerged as a powerful tool for artificial intelligence-driven scientific discovery by learning interpretable analytical expres

PSA: llama.cpp now loads MTP tensors by default for any draft-mtp arch, even with MTP disabled

model-releasesr-localllama

If your GGUF has MTP/NextN tensors baked in (GLM-5.2, hy_v3, qwen35moe, step35, etc.), recent llama.cpp builds load them by default — even if you never pass --spec-type draft-mtp. Before, they were sk

Quantizing Kimi K3 (2.8T A50B) to GGUF ourselves - Q3_K_S works, 1.1 TB on disk

model-releasesr-localllama

we're experimenting with our own dynamic GGUF quants of kimi k3, made from the original weights with our llama.cpp fork. Q3_K_S is done and works 1114.76 GiB on disk. Q1 and Q2 are in progress, result

Quoting Matthew Green

model-releasessimon-willison

Right now we’re in the midst of a historic transition from traditional public-key algorithms based on EC-based cryptography and RSA, moving over to new post-quantum algorithms based on novel problems.

RDQ: Residual Distribution Quantization for Large Language Models

arXiv:2607.10137v2 Announce Type: replace Abstract: Post-training quantization (PTQ) of large language models degrades sharply below 4-bit precision. We identify the root cause as residual stream dist

RDVSv2: A Large-scale Benchmark for RGB-D Video Salient Object Detection

arXiv:2607.25392v1 Announce Type: new Abstract: We introduce RDVSv2, a large-scale benchmark for RGB-D video salient object detection (RGB-D VSOD) with dense frame-level annotations. Existing datasets

ReDesign: Recovering Editable Design Structures from Images via Agentic Decomposition

arXiv:2607.25565v1 Announce Type: new Abstract: Recovering an editable design file from a raster image is a common and costly bottleneck in modern design workflows, yet remains challenging since edita

RefBench-PRO: Perceptual and Reasoning Oriented Benchmark for Referring Expression Comprehension

arXiv:2512.06276v3 Announce Type: replace-cross Abstract: Referring Expression Comprehension (REC) is a vision-language task that localizes a specific image region based on a textual description. Exis

Reference Traces for Auditing Invisible Weight Updates and Guiding Exact-Budget Protection

arXiv:2607.09800v3 Announce Type: replace Abstract: Direct low-precision write-back can erase optimizer proposals, while aggregate update visibility need not identify parameters worth protecting. We s

Regime-Adaptive Bayesian Optimization via Dirichlet Process Mixtures of Gaussian Processes

arXiv:2601.20043v2 Announce Type: replace Abstract: Standard Bayesian Optimization (BO) assumes uniform smoothness across the search space an assumption violated in multi-regime problems such as molec

Reinforcement Learning for Code Optimization

arXiv:2607.25970v1 Announce Type: cross Abstract: RL for code correctness is now established: have the model generate a program, run it against hidden test cases, and reward solutions that pass. Exten

ReLATE: Reliability-Guided Evidence Fusion for Robust UAV--Satellite cross-view Geo-Localization

arXiv:2607.25524v1 Announce Type: cross Abstract: Unmanned aerial vehicle (UAV)-satellite cross-view geo-localization matches UAV images against satellite imagery and has achieved impressive accuracy

REPREC: Representation Driven Parameter-Efficient Recommendation System

arXiv:2607.24845v1 Announce Type: cross Abstract: Large language models (LLMs) have been applied to sequential recommendation by formulating it as a natural language task. Previous work has improved p

Research Report on Noise-Shaped One-Bit Coefficients in Discrete Polynomial Fourier Extension

arXiv:2607.24868v1 Announce Type: new Abstract: This report studies noise-shaped one-bit coefficients in normalized discrete polynomial Fourier extension. For first-order Sigma-Delta quantization, the

Rethinking CD: A Reproducibility Study and Extension on the Ineffectiveness of Contrastive Decoding at Mitigating Object Hallucinations in MLLMs

arXiv:2607.25196v1 Announce Type: new Abstract: Contrastive decoding (CD) has been proposed as a training-free strategy for mitigating object hallucinations in multimodal large language models (MLLMs)

Retraction-Free Optimization over the Stiefel Manifold for the LoRA Fine-Tuning

arXiv:2607.25299v1 Announce Type: cross Abstract: Optimization over the Stiefel manifold plays a significant role in various machine learning tasks. Existing methods either use the retraction operator

RIDGE: An Autonomous Framework for Validation and Method Discovery in LLM-Generated Option Pricing

arXiv:2607.25199v1 Announce Type: cross Abstract: Automated code generation is becoming an important tool in quantitative finance, where large language models can generate option pricing implementatio

RoboProcessBench: Benchmarking Process-Aware Understanding in Vision-Language Robotic Manipulation

model-releasesarxiv-cs-ro

arXiv:2606.13040v2 Announce Type: replace Abstract: Vision-language models (VLMs) are increasingly explored as visual critics, reward generators, and failure detectors in robotic manipulation. These r

RRS-10K: A Multitask Vision-Language Model Benchmark for Rare Remote Sensing Image Interpretation

arXiv:2607.24810v1 Announce Type: new Abstract: Vision-language models (VLMs) have achieved strong performance on general remote sensing tasks. However, their capability for rare scenes remains insuff

RSIBench-Data: Benchmarking Data-Centric Research for Recursive Self-Improvement

arXiv:2607.25886v1 Announce Type: cross Abstract: Recursive self-improvement requires turning evidence of model failures into better models. Data-centric post-training research entails diagnosing capa

RSMeM: Knowledge-Enhanced Memory Evolution for Remote Sensing Agents with Systematic Evaluation

arXiv:2607.24772v1 Announce Type: new Abstract: Geoscience research requires complex analysis and domain expertise, with remote sensing (RS) observations as a key foundation. However, existing RS agen

SAGE: Stochastic Prompt Optimization via Agent-Guided Exploration

arXiv:2606.18902v2 Announce Type: replace Abstract: Context engineering has emerged as a primary lever for improving AI systems without parameter updates. Recent work showing that textual gradients do

SAM-MI: A Mask-Injected Framework for Enhancing Open-Vocabulary Semantic Segmentation with SAM

arXiv:2511.20027v2 Announce Type: replace Abstract: Open-vocabulary semantic segmentation (OVSS) aims to segment and recognize objects universally. Trained on extensive high-quality segmentation data,

ScaleResfusion: Residual Rectified Flow based on Residual Vector Field

arXiv:2607.25275v1 Announce Type: cross Abstract: Real-world Image Restoration (Real-IR) aims to recover high-quality (HQ) images from complex and unknown degradations. Although recent diffusion-based

ScoreShield: Differentially Private Release of Similarity Scores

arXiv:2607.25041v1 Announce Type: cross Abstract: A growing number of applications, such as biometrics and retrieval-augmented generation (RAG), rely on cosine similarity scores computed between vecto

SecDrift: Measuring Sector-Conditioned Security Drift in AI-Generated Code

arXiv:2607.25225v1 Announce Type: cross Abstract: LLMs are increasingly used for code generation in critical infrastructure, yet the security effect of domain-specific prompting is understudied. We pr

Sense it with your eyes: Sensation Generation and Understanding for Advertisements

arXiv:2607.25314v1 Announce Type: new Abstract: Sensory advertising evokes human senses through visual cues, enabling audiences to mentally simulate experiences and increasing persuasive impact. Despi

SGTP: Sampling-based Game-Theoretic Planning for Real-Time Multi-Vehicle Autonomous Racing

model-releasesarxiv-cs-ro

arXiv:2607.25388v1 Announce Type: new Abstract: Autonomous multi-vehicle racing requires real-time planning of diverse competitive behaviors in intense interactions. Existing planners often struggle t

Sharpness-Aware Minimization and Muon: Robustness under the Spectral Norm

arXiv:2607.26001v1 Announce Type: new Abstract: Sharpness-Aware Minimization (SAM) aims to improve generalization by encouraging insensitivity to small, worst-case parameter perturbations. However, th

Sharpness-aware Model Merging with Salience Recovery for LLM-based Cross-Domain Sequential Recommendation

arXiv:2607.25366v1 Announce Type: cross Abstract: LLM-based Cross-Domain Sequential Recommendation (CDSR) leverages LLMs to enhance target performance via deep semantic reasoning, alleviating the depe

Shieldstral

arXiv:2607.25857v1 Announce Type: new Abstract: We introduce Shieldstral, a 3B-parameter policy-adaptive multimodal safety classifier that matches or outperforms models nearly 7imes its size on text s

Similar Models Learn Differently: Final-Window Pretraining Shapes Post-Training Beyond SFT

arXiv:2607.25063v1 Announce Type: new Abstract: Developers judge a model checkpoint by how it behaves. After supervised fine-tuning (SFT), two checkpoints that perform about the same across relevant b

Simulation-based parameter estimation via a combination of embedded normalizing flows and implied empirical probabilities under moment restrictions

arXiv:2607.25026v1 Announce Type: cross Abstract: In this work, we present a simulation-based parameter estimation framework for a model defined by a computational simulation of a physical system. We

SpecPrefetch: Parameter-Efficient Expert Prefetching for Sparse MoE Foundation Models

arXiv:2607.24787v1 Announce Type: new Abstract: Sparse Mixture-of-Experts (MoE) models expand foundation model capacity through conditional expert activation, but their full expert pools remain diffic

SpeechLLM Meets Federated Learning for End-to-End ASR: English and Italian Case Studies

arXiv:2607.25716v1 Announce Type: new Abstract: Federated learning (FL) enables privacy-preserving training of automatic speech recognition (ASR) systems across distributed data sources, yet its appli

Spend Experts Where You Are Unsure: Confidence-Adaptive Routing for Mixture-of-Experts LoRA

arXiv:2607.26052v1 Announce Type: new Abstract: Mixture-of-Experts (MoE) variants of Low-Rank Adaptation (LoRA) route every token to a fixed number of experts k. Tokens differ in how uncertain the mod