Search: “routing”

applicationssonya-huang--x

We are going to see a lot of vertically focused AI native companies accelerate. Routers, open-source models and specialized post-training enabled by companies like @FireworksAI_HQ have all made dramat

1 Aug 2026

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Applications

CACHE-UK: A Stability-Aware Memory Editor for Sequentially Updated Quantized LLMs in Finance

DGX agent

arXiv:2607.28292v1 Announce Type: new Abstract: Large Language Models (LLMs) deployed in dynamic financial environments face a critical challenge: maintaining factual accuracy as market conditions, re

applicationsarxiv-cs-cl

31 Jul 2026

Critical attention scaling in long-context transformers

arXiv:2510.05554v2 Announce Type: replace Abstract: As large language models scale to longer contexts, attention layers suffer from a fundamental pathology: attention scores collapse toward uniformity

Divergence Decoding: Training-Free Capability Fusion

arXiv:2607.27248v1 Announce Type: cross Abstract: While large language models excel in reasoning, these generalists often lack knowledge for specialized scientific domains. Conversely, domain models~(

Efficient LLMs with AMP: Attention Heads and MLP Pruning

arXiv:2504.21174v2 Announce Type: replace Abstract: Deep learning drives a new wave in computing systems and triggers the automation of increasingly complex problems. In particular, Large Language Mod

Fidelity Is Not Safety: Gently-Compressed LLMs Pass Every Data-Free Quality Guard Yet Invent Procedure Steps in Agentic Execution

arXiv:2607.28196v1 Announce Type: new Abstract: Practitioners accept a compressed language model once it clears a stack of data-cheap quality guards: perplexity within a small factor of the original,

safetyarxiv-cs-cl

First-order Constrained Trilevel Optimization Over Distributed Networks for Robust Coreset Selection

arXiv:2607.27632v1 Announce Type: new Abstract: With the rapid advancement of the Internet of Things (IoT), massive amounts of data are generated across distributed edge networks. Training models on f

local-aiarxiv-cs-lg

Memory Decoder at Scale: A Pretrained, Parametric Long-Term Memory

arXiv:2607.27919v1 Announce Type: new Abstract: Decoder-only language models entangle long-term memory and reasoning in a single parameter set, making it difficult to scale memory capacity independent

Tutorials

MentorCollab: Selective Large-to-Small Inference-Time Guidance for Efficient Reasoning

arXiv:2602.05307v2 Announce Type: replace Abstract: Large reasoning models (LRMs) achieve strong performance by producing long chains of thought, but their inference costs are high and often generate

tutorialsarxiv-cs-cl

One Patch Is Enough: Reinforcement-Optimized Visual Token Grounding for MLLM-Based Scene Text Spotting

model-releasesarxiv-cs-cv

arXiv:2607.27902v1 Announce Type: new Abstract: Scene text spotting requires high-precision alignment between textual recognition and spatial localization. While visual-token grounding has emerged as

Policy Gradient Steering: Interventions from Behavioral Objectives

arXiv:2607.27574v1 Announce Type: new Abstract: Activation steering has emerged in large language models as a lightweight alternative for dynamically changing a model's behavior at inference time. How

safetyarxiv-cs-lg

Reasoning Consensus: Structural Ensembling of LLM Reasoning via Weighted DAG Aggregation

arXiv:2607.27783v1 Announce Type: new Abstract: Large Language Models (LLMs) explore problems through chain-of-thought, but this exploration is buried in unstructured prose. On high-stakes tasks, user

researcharxiv-cs-cl

SkillSight: Calibrating Generic Content Bias for Skill Retrieval

arXiv:2607.18785v2 Announce Type: replace Abstract: As large language model agents gain access to increasingly large skill libraries, retrieving the right skill becomes critical to reliable capability

safetyarxiv-cs-ai

VCP-DCN: Beyond Visual Concealed Property via Depth Collaborative Network for Camouflaged Object Detection

arXiv:2607.27843v1 Announce Type: new Abstract: Camouflaged Object Detection (COD) aims to identify and segment camouflaged objects in complex environments, which are often concealed because their col

local-aiarxiv-cs-cv

Why Are GUI Agents Correct but Late? Decode on the Decision-Time Critical Path, Tested with Pre-Compiled Policy Trees

arXiv:2607.28399v1 Announce Type: new Abstract: Computer-use agents often fail on transient GUI events because they produce the correct action only after the relevant window has already closed. We ide

FedWeave: Rethinking the Unit of Specialization in Heterogeneous Federated MoE-LoRA

arXiv:2607.26618v1 Announce Type: cross Abstract: Federated PEFT enables LLMs to collaboratively adapt to decentralized private data without sharing raw examples. However, task heterogeneity across cl

Lottery Tickets Are Not Deployment Tickets

arXiv:2607.27031v1 Announce Type: cross Abstract: Reports on how sparsification, compression, and lottery tickets change model behavior have been mixed in the prior literature, with beneficial effects

safetyarxiv-cs-cv

MedARC: Training-Free Adaptive Redundancy Compression of Visual Tokens for 3D Medical Vision-Language Models

arXiv:2607.26554v1 Announce Type: new Abstract: Integrating 3D medical images with vision-language models (VLMs) holds substantial promise for computer-aided diagnosis. However, volumetric images gene

local-aiarxiv-cs-cv

Misalignment Has a Personality: A Big Five Account of Emergent Misalignment

arXiv:2607.26389v1 Announce Type: new Abstract: Fine-tuning a language model on data containing a narrow flaw, such as insecure code or incorrect mathematical answers, can cause broad misalignment thr

safetyarxiv-cs-cl

Not every page in your PDF needs the same treatment. A scanned cover, a dense table, a clean text page, a figure-heavy diagram.... most pars…

model-releasesjerry-liu--x

Not every page in your PDF needs the same treatment. A scanned cover, a dense table, a clean text page, a figure-heavy diagram.... most parsing pipelines throw all of them at the same parser, forcing

Representation Trajectories Matters: Complementary Evidence for OOD Detection and Image Classification

arXiv:2607.26565v1 Announce Type: new Abstract: Vision models do not form a representation at once; each block revises it. We ask whether the resulting computation path contains evidence that the fina

researcharxiv-cs-cv

Situational Awareness, the $20bn hedge fund founded by former OpenAI employee Leopold Aschenbrenner, has sought to raise fresh capital from …

Situational Awareness, the $20bn hedge fund founded by former OpenAI employee Leopold Aschenbrenner, has sought to raise fresh capital from investors after suffering heavy losses during the recent rou

safetygary-marcus--x

Uncertainty-Guided LLM Semantic Augmentation for Heterogeneous Treatment Effect Estimation

arXiv:2607.26599v1 Announce Type: new Abstract: Estimating heterogeneous treatment effects is central to targeted interventions, such as personalized promotions and precision medicine. We focus on the

local-aiarxiv-cs-lg

Under the Hood: Serving Kimi K3

model-releasesdigitalocean

DigitalOcean launched Kimi K3 on day 0. It’s already one of the most popular models on the platform and across the market: second most likes on Hugging Face, sixth most traffic on OpenCode. Getting a

Beyond Self-Knowledge: Propagating Uncertainty Across Reasoning and Retrieval in LLMs

arXiv:2607.25600v1 Announce Type: cross Abstract: Retrieval-augmented generation improves knowledge-intensive question answering, but indiscriminate retrieval can introduce irrelevant evidence and unn

safetyarxiv-cs-ai

From CUDA to MLX: How K-Search Brings Decades of Kernel Expertise to Apple Silicon

model-releasesberkeley-ai-research

Figure 1: CUDA-to-MLX optimization translation map. CUDA optimization knowledge can be translated into architecture-native MLX strategies rather than copied instruction-for-instruction. We face a new

RoboHarness: Memory-Driven Orchestration of Heterogeneous Robot Policies for Long-Horizon Planning

arXiv:2607.18060v2 Announce Type: replace Abstract: Long-horizon robotic tasks require diverse capabilities that no single policy can reliably provide. Heterogeneous policies offer complementary stren

safetyarxiv-cs-ro

Toward an Organizational Science of Multi-Agent LLM Systems: Decoupling Who, How, and Which Algorithm

arXiv:2607.25446v1 Announce Type: new Abstract: Multi-agent frameworks built on large language models (LLMs) routinely entangle three logically distinct concerns: who is on the team (organization), ho

CausalGate: Causal Importance Distillation for Transformer Module Pruning

arXiv:2607.22720v1 Announce Type: cross Abstract: Existing adaptive inference methods for Large Language Models rely on observational heuristics, such as hidden-state similarity or activation magnitud

Cheap Probes Predict Expensive Training in 3D-CT Vision--Language Models

arXiv:2607.22771v1 Announce Type: cross Abstract: Picking the frozen image encoder for a 3D~CT vision--language model (VLM), together with the token-compression scheme on top of it, is a search over m

Applications

Co-Learning for Missing Arbitrary Modalities in Multi-modal Classification

arXiv:2607.24683v1 Announce Type: cross Abstract: Multi-modal classification leverages complementary information across diverse data sources to enhance predictive performance. However, real-world scen

applicationsarxiv-cs-ai

DeepSeek V4 Flash, up to 32 tok/s on AMD Ryzen AI MAX+ 395

model-releasesr-localllama

Hey fellow llamas. we have something new for Strix Halo owners we thought would be useful to share. i'll keep it short: We were able to fit DeepSeek V4 Flash plus its speculative draft on a single Ryz

Do Visual Features Improve Other-Initiated Repair Detection? A Dyadic Multimodal Approach

arXiv:2607.23845v1 Announce Type: new Abstract: Other-initiated Self-repair, or in short Other-initiated Repair (OIR), is an essential mechanism in conversational interaction, whereby a recipient sign

researcharxiv-cs-ai

DraftExpert: Expansion-Aware Self-Speculative Decoding for End-Device MoE Inference

arXiv:2607.24434v1 Announce Type: cross Abstract: Large Mixture-of-Experts (MoE) language models are attractive for end-device deployment because only a small subset of experts is active per token, bu

DuoAD: Leveraging [CLS] Dual Characteristics for Training-Free Few-Shot Anomaly Detection

arXiv:2607.23924v1 Announce Type: cross Abstract: Vision foundation models have enabled strong training-free anomaly detection (AD). However, most existing approaches rely primarily on independent loc

GOTS: Greedy Orthogonal Token Selection for High-Resolution Vision-Language Models

arXiv:2607.23913v1 Announce Type: new Abstract: Modern vision-language models (VLMs) increasingly rely on dynamic or high-resolution visual encoding, producing thousands of visual tokens that substant

HeraSys: Collaborative Serving of Multiple LLM Workflows via Fine-Grained End-to-End Optimization

arXiv:2607.22578v1 Announce Type: new Abstract: The proliferation of Large Language Models (LLMs) has shifted serving systems from processing isolated requests to orchestrating high-concurrency, multi

safetyarxiv-cs-ai

Joint Optimization for Greedy Longest-match Tokenization

arXiv:2607.23362v1 Announce Type: new Abstract: Recent work has shown that subword vocabularies can be trained to optimize compression for a specific inference rule rather than relying on greedy heuri

researcharxiv-cs-cl

Kimi K3: Open Frontier Intelligence

arXiv:2607.24653v1 Announce Type: new Abstract: We introduce Kimi K3, a 2.8T parameter Mixture-of-Experts model with 104 billion activated parameters, native vision capabilities, and a 1-million-token

Agents

Learning Traversability-Aware Global Planners for Long Horizon Off-Road Navigation

arXiv:2607.23743v1 Announce Type: new Abstract: Autonomous navigation across large off-road environments remains a challenging problem. Onboard sensors perceive only the immediate surroundings, yet sa

agentsarxiv-cs-ro

MoLGE: Mixture of Language Group Experts for Efficient Scaling of Massively Multilingual Speech Recognition

arXiv:2607.24030v1 Announce Type: new Abstract: Massively multilingual automatic speech recognition (ASR) models covering hundreds of languages must maintain robust performance across diverse linguist

MSVS-VAE: Multi-Scale Anchored VecSet for High-Fidelity 3D Reconstruction

arXiv:2607.24436v1 Announce Type: new Abstract: High-fidelity 3D generative modeling increasingly relies on the latent diffusion paradigm, where the reconstruction quality of the underlying 3D VAE bec

researcharxiv-cs-cv

Hardware

Omni-Prune: Query-Aware Unified Token Pruning for Efficient Omnimodal Large Language Models

arXiv:2607.23445v1 Announce Type: cross Abstract: Omnimodal large language models (OmniLLMs) are rapidly extending multimodal reasoning to cover synchronized audio and video. However, the resulting au

hardwarearxiv-cs-cl

Operational Proto-Introspection in Looped Language Models: Process-Quality Taps, Executable Branching, and the Readout-Control Boundary

arXiv:2607.18553v2 Announce Type: replace-cross Abstract: Can a language model read the quality of its ongoing computation, and can an external intervention turn that readout into better outcomes? We

researcharxiv-cs-ai

Tutorials

PathSelect: Sequential Token Selection for Whole Slide Pathology

arXiv:2607.23631v1 Announce Type: new Abstract: Gigapixel Whole-Slide Images (WSIs) present a fundamental computational bottleneck for vision-language models (VLMs) due to extreme sequence lengths. Ex

tutorialsarxiv-cs-cv

PIVOT: Efficient Query-Group Indexing for Token-Level Sparse Attention

arXiv:2607.24593v1 Announce Type: new Abstract: Token-level sparse attention, as implemented by DeepSeek Sparse Attention (DSA) in production systems, makes the downstream attention efficient but shif

ProvenanceGuard: Source-Aware Factuality Verification for MCP-Based LLM Agents

arXiv:2606.18037v2 Announce Type: replace Abstract: Tool-using LLM agents increasingly use the Model Context Protocol (MCP) to answer from heterogeneous evidence sources, including search, APIs, datab

PTStore (Prefix Tensor Store): Distributed Prefix Caching and Replication for High Throughput Inference Serving

arXiv:2607.22648v1 Announce Type: new Abstract: Inspired by the design of client caching in Content Delivery Networks (CDNs), PTStore distributes and replicates popular tensors that form reusable KV c

researcharxiv-cs-ai