Knowledge catalogue
arXiv:2609.34314v1 Announce Type: new Abstract: Video-language models are increasingly used as judges of video understanding, both for evaluating model outputs and for training reward models. Whether
arXiv:2609.34605v1 Announce Type: new Abstract: Multi-teacher on-policy distillation (MOPD) has emerged as a popular post-training paradigm for integrating specialized capabilities in frontier languag
arXiv:2609.32856v1 Announce Type: new Abstract: Vector polygon generation converts visual inputs, e.g., remote sensing (RS) images, into vectorized polygonal geometries, supporting applications such a
arXiv:2609.33746v1 Announce Type: new Abstract: At each decoding step a language model attends over the key-value (KV) cache of every earlier token, so at long context the attention call is bounded by
arXiv:2609.33489v1 Announce Type: new Abstract: Simultaneous and sequential block updates are two basic optimization strategies used across machine learning, such as neural-network training, federated
arXiv:2609.34528v1 Announce Type: new Abstract: Open-vocabulary semantic segmentation (OVSS) enables pixel-level prediction over arbitrary text-specified vocabularies and has shown strong generalizati
arXiv:2609.32259v1 Announce Type: cross Abstract: Recent multi-agent LLM systems increasingly combine heterogeneous models for specialized agent roles. However, text-based communication requires each
arXiv:2609.32188v1 Announce Type: new Abstract: Text-to-image (TTI) models increasingly generate high-quality images from natural-language prompts, yet figurative language exposes a failure: a vehicle
arXiv:2609.33765v1 Announce Type: new Abstract: Generative robot policies provide expressive behavior priors, but updating a large diffusion or flow-matching model through online interaction is costly
arXiv:2609.32547v1 Announce Type: new Abstract: Large language models are commonly evaluated by generating a small number of stochastic responses for each prompt in a benchmark. Because inference budg
arXiv:2609.32429v1 Announce Type: cross Abstract: Smaller activation outliers do not necessarily imply better low-bit quantization: their alignment with the quantizer matters. We introduce PrismQuant,
arXiv:2609.34768v1 Announce Type: cross Abstract: Millimeter-wave (mmWave) radar enables privacy-preserving human perception, but the extreme sparsity of point clouds from commercial single-chip senso
arXiv:2609.34415v1 Announce Type: new Abstract: The transition from Large Language Models (LLMs) to agents shifts safety stakes from toxic text to irreversible environmental harm. While current defens
arXiv:2609.33646v1 Announce Type: cross Abstract: Browser-use agents require seamless alignment between structured web metadata and visual information, while preserving relevant context across long in
arXiv:2609.33855v1 Announce Type: new Abstract: Self-evolving vision-language models train on questions they generate from unlabeled images. Since these questions have no gold answers, prior methods l
arXiv:2609.34299v1 Announce Type: new Abstract: Dataset distillation (DD) condenses a large original dataset into a small distilled dataset with high training utility. Decoupled statistical matching m
arXiv:2609.34426v1 Announce Type: new Abstract: This paper addresses the problem of safe offline reinforcement learning, which involves training a policy to satisfy safety constraints using an offline
arXiv:2609.33269v1 Announce Type: cross Abstract: World Action Models (WAMs) jointly generate video and robot actions through iterative diffusion and perform strongly in robotic manipulation. However,
arXiv:2609.35581v1 Announce Type: cross Abstract: We introduce QC-Stark, a benchmark for evaluating large language models (LLMs) on 11 quantum computing (QC) tasks, spanning circuit construction, debu
arXiv:2609.33638v1 Announce Type: new Abstract: We introduce RareTrap, a framework for estimating the probability of severe behaviors in black box large language models (LLMs). A key challenge for pro
arXiv:2601.04301v3 Announce Type: replace Abstract: As frontier AI systems are pretrained on web-scale data, test set contamination has become a critical concern for accurately assessing their capabil
arXiv:2609.32659v1 Announce Type: new Abstract: Quantization-Aware Pre-Training (QAPT) can increase the inference efficiency of DNNs, but a problematic behaviour known as rounding boundary weight osci
arXiv:2609.35349v1 Announce Type: new Abstract: The evaluation cost of transformers with softmax attention scales quadratically with sequence length. Kernel attention addresses this by replacing softm
arXiv:2609.33097v1 Announce Type: new Abstract: Recent large-scale Vision-and-Language Navigation (VLN) models deliver strong accuracy but remain costly to deploy due to their large parameter counts a
We evaluate several models on 100 tasks from the [internal Binary Exploitation benchmark] (selected at random), and find that GLM-5.3 develops full control flow hijacks in 4% of the trials; Claude Myt
arXiv:2609.33351v1 Announce Type: cross Abstract: Fidelity-based quantum retrieval ranks candidates by the fidelity between query and archive states. Applying a shared input-independent unitary after
I'm running Qwen 3.8 27B Q4 XS with ~30 t/s decode (no MTP) at 100k context with q8/q5 KV cache on a 16GB AMD GPU. I wanted to share my setup since many believe Qwen 3.8 27B to be infeasible on 16GB V
I started mapping the building blocks shared across all the models in audio.cpp. The result ended up being more interesting than I expected. Qwen has become by far the most common language backbone in
Qwen3.8-Flash is 40% off through the rest of the month, making now a perfect time to run your evals. Designed for high-volume applications like coding and coworking assistants, @Alibaba_Qwen's model o
arXiv:2609.33848v1 Announce Type: new Abstract: Large language model (LLM) agents increasingly undertake extreme-long (xlong) horizon tasks, where a single execution can span hours, hundreds of model-
arXiv:2609.33796v1 Announce Type: cross Abstract: Translating natural-language access-control requirements into policies requires careful reasoning about permissions, constraints, and exceptions, and
Thanks to tcclaviger, vLLM now has expert RAM offloading support (link). This makes frontier models much more accessible on a local setup! I was able to run the original DeepSeek-V4-Flash-Vision-Exp o
arXiv:2609.31795v1 Announce Type: new Abstract: We describe our entry to the LoViF 2026 AIGC Image Compression Challenge, a benchmark for ultra-low-bitrate coding of AI-generated images under a strict
arXiv:2609.34367v1 Announce Type: new Abstract: Learned image codecs (LICs) achieve high reconstruction quality, but their decoding speed is often insufficient for immersive virtual reality (VR). Gaus
arXiv:2609.32681v1 Announce Type: new Abstract: 4D radar provides geometric and motion cues that complement visual semantics, but integrating it into vision-language-action (VLA) models requires both
arXiv:2609.32802v1 Announce Type: cross Abstract: One variable sets what an upstream fault costs a staged pipeline of language-model agents: re-derivability, how much of what a stage needs it can rebu
arXiv:2609.32757v1 Announce Type: cross Abstract: VLM bounding-box localization is both language generation and spatial commitment. Parseable fields such as bbox_2d make localization easy to score, bu
arXiv:2609.32734v1 Announce Type: new Abstract: AI-generated image detectors are often evaluated on benchmarks where real and synthetic images differ in content, quality, or generation artifacts, allo
arXiv:2609.31927v1 Announce Type: cross Abstract: MathNet-Retrieve asks a retriever to find, for a math problem, a document stating the same problem. An LLM under one fixed prompt writes each gold doc
arXiv:2609.32409v1 Announce Type: new Abstract: Amortized neural likelihoods enable computationally expensive inference for models with analytically intractable or unspecified likelihoods, but their b
arXiv:2606.18022v2 Announce Type: replace Abstract: Masked diffusion models (MDMs) generate sequences by iteratively refining a partially masked state and committing tokens in parallel. We introduce r
arXiv:2609.32415v1 Announce Type: new Abstract: Diffusion Transformers (DiTs) have become the standard backbone for high-quality generative modeling, yet deploying them in conditional generation tasks
You may think that jev, OpenClaw or TurboQuant are super cool, but actually the coolest LLM invention happened two years ago As we all know, the best source of reliable information about LLMs is YouTu
arXiv:2609.32298v1 Announce Type: new Abstract: Drifting Models train a one-step generator by recomputing a finite-sample drift field at every iteration and taking an optimizer step toward the drifted
arXiv:2609.34231v1 Announce Type: new Abstract: Metamaterials are artificially engineered structures whose mechanical and physical behaviors are strongly shaped by geometry rather than composition. Vo
arXiv:2604.04410v2 Announce Type: replace Abstract: Aligning language models with human preferences is essential for ensuring their safety and reliability. Although most existing approaches assume spe
We have released Qwen3.8-Flash-Next quantized with GSQ and RCO, together with a second, capability-targeted build in which half of the model's experts have been removed. Flash-Next is a sparse mixture
arXiv:2609.35764v1 Announce Type: new Abstract: Sparse inertial pose estimation promises camera-free motion capture from consumer devices, but consumer sensors are unreliable: firmware-fused orientati
arXiv:2609.32460v1 Announce Type: new Abstract: Recent video generation models produce realistic videos and show potential as a foundation for world models. These advances create opportunities for gen
arXiv:2606.16575v3 Announce Type: replace Abstract: Deep neural networks (DNNs) have achieved remarkable success in scientific computing, yet they often suffer from spectral bias in capturing oscillat
arXiv:2609.32241v1 Announce Type: cross Abstract: Neural image watermarks can be forged by extracting watermark-bearing residuals from released images and transferring them to unrelated content. While
arXiv:2609.34905v1 Announce Type: new Abstract: Power sampling has emerged as a training-free approach to LLM reasoning, eliciting capabilities comparable to reinforcement learning by sharpening the m
arXiv:2609.33687v1 Announce Type: new Abstract: Onboard satellite models often require frequent updates, but the weights adapted to earlier data distributions can quickly become outdated. However, upd
arXiv:2609.35433v1 Announce Type: new Abstract: Reinforcement learning from verifiable rewards (RLVR) frequently reuses rollouts across multiple policy updates, increasing the mismatch between the cur
arXiv:2609.33944v1 Announce Type: new Abstract: Jointly generating future video and actions has become a standard recipe for world-action models, and the strongest systems denoise the two streams on s
arXiv:2609.35686v1 Announce Type: new Abstract: Mechanistic interpretability (MI) aims to explain a model's behaviour through analyzing its internal computations; circuit-based explanations aim to iso
arXiv:2609.35695v1 Announce Type: new Abstract: Aligning large language models (LLMs) to diverse user preferences is fundamentally hindered by standard alignment paradigms that optimize for monolithic
arXiv:2609.32444v1 Announce Type: new Abstract: We study training-inference mismatch in reinforcement learning with verifiable rewards (RLVR) for large language models, where rollouts are sampled by a
arXiv:2609.32749v1 Announce Type: cross Abstract: Model distillation transfers capabilities through supervised fine-tuning (SFT) on teacher responses, often collected from commercial APIs, raising que
arXiv:2609.32756v1 Announce Type: new Abstract: Foundation models are rarely used as generic, frozen feature extractors; instead, they are fine-tuned for the target downstream application. This practi