Search: “local-ai”

What does FFN compression change downstream? Same-state causal restoration in diffusion language models

arXiv:2609.31685v1 Announce Type: new Abstract: Diffusion language models (DLMs) enable flexible, parallel generation, but their iterative denoising remains computationally expensive, motivating incre

Human

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Local Ai

What Should Data Teach? Moving Bottlenecks Across Circuit, Store, and Use

DGX agent

arXiv:2609.32991v1 Announce Type: new Abstract: What should data teach a language model at a particular point in training? A circuit view reveals three distinct bottlenecks: forming a computation, mak

local-aiarxiv-cs-lg

29 Sept 2026

What Should We Freeze? Guarded Freezing: Connectivity Shapes the Fine-Tuning of Pretrained Models

arXiv:2609.32124v1 Announce Type: new Abstract: When adapting pre-trained models through fine-tuning, freezing weights alone might not preserve performance, as updates elsewhere can change the inputs

When Privacy Moves ML-Mediated Decisions On Device: Information and Incentive Misalignment in Auctions

arXiv:2609.33312v1 Announce Type: cross Abstract: Moving ML-mediated decision making onto privacy-preserving clients decentralises the economic decision along with the inference. Shared budget constra

'Where Can I Trust You?': Boundary-Aware Evaluation of Surrogate Fidelity

arXiv:2609.32230v1 Announce Type: new Abstract: Surrogate models are commonly evaluated by how often they agree with their teacher model over an evaluation set. Local variation in this agreement is we

XFlow: A Workflow Model for Instruction-Guided Lesion Segmentation in Chest X-rays

arXiv:2609.34513v1 Announce Type: new Abstract: Existing text-guided segmentation models in the medical domain cover only a narrow set of anatomical structures and lesions in chest X-rays (CXRs), and

You Only Edit Once: Incentivizing In-Context Capability of LLMs via Local Demonstration Refinement

arXiv:2609.33609v1 Announce Type: new Abstract: In-context learning (ICL) is crucial for boosting the inference performance of large language models (LLMs). However, the effectiveness of ICL in LLMs i

Zero-Storage Procedural Neural Synthesis via Boundary Dynamics: Formal Verification in Lean 4 and Bare-Metal Gauntlet Validation

arXiv:2609.33066v1 Announce Type: new Abstract: Contemporary neural inference architectures rely on dense floating-point weight matrices stored in high-bandwidth memory (VRAM), incurring severe memory

Adaptive Interaction Graphs for Particle Simulation

arXiv:2609.30822v1 Announce Type: new Abstract: Learned particle simulators based on graph neural networks achieve strong one-step accuracy, but errors compound over long horizons. An underexplored va

Atelier: Learning Local Self-Supervised Features for CryoEM Volumes via Hypernetworks

arXiv:2609.30569v1 Announce Type: new Abstract: CryoEM map interpretation requires features that are spatially localized, consistent across samples, and informative across spatial scales. Most deep le

AtomWorld-Mem: Memory-Restored World States for Long-Horizon Atomistic Evolution

arXiv:2609.31133v1 Announce Type: new Abstract: High-fidelity atomistic evolution over long timescales requires more than observing the current crystal configuration. Instantaneous atomistic snapshots

Budgeted Quotient-Residual Guidance for Frozen Pocket-Conditioned Molecular Diffusion

arXiv:2609.31222v1 Announce Type: new Abstract: Pocket-conditioned molecular diffusion updates ambient atom coordinates, but many lead-optimization objectives are expressed on quotient features such a

Can You Check That? The Checkability Boundary for Local LLM Network Automation

arXiv:2609.31540v1 Announce Type: cross Abstract: Sending every network-automation input to a third-party frontier LLM exports sensitive artifacts such as production configurations, topologies, and lo

Cheap, open agents make LLM pollution harder to mitigate

arXiv:2609.31054v1 Announce Type: new Abstract: Large Language Model (LLM) pollution occurs when synthetic responses contaminate data intended to capture human behavior. High deployment costs have so

DeepEdu-v1: Efficient and Scalable Agentic LLMs for Vietnamese Education

arXiv:2609.31568v1 Announce Type: new Abstract: AI tutoring could markedly improve learning outcomes for students in developing regions such as Vietnam, yet the two obvious paths both fall short. Clou

Distributed Learning as a Service: The Developer's Perspective

arXiv:2609.31061v1 Announce Type: new Abstract: Application developers of distributed learning services face challenges that a typical federated learning loop does not address. Specifically, the model

DyMD: Preserving Interaction Dynamics through Distribution Matching Distillation in Few-Step Video World Models

arXiv:2609.31349v1 Announce Type: cross Abstract: Large video diffusion models offer expressive priors for embodied prediction and learning, yet their many-step sampling remains costly for interactive

G^2PTQ: Improving LLM Post-Training Quantization with Generalized Gradient Compensation

arXiv:2609.31009v1 Announce Type: cross Abstract: Post-training quantization (PTQ) is a practical approach to reducing the memory and computational footprint of large language models (LLMs) without re

Holo4

Holo4-27B-GGUF is a vision-language model (VLM) for Computer Use, built on the Qwen3.8 dense architecture and developed by H Company. Used with the hai-agents harness, it can send screenshots and tool

Jagarin: A Three-Layer Architecture for Hibernating Personal Duty Agents on Mobile

arXiv:2603.05069v3 Announce Type: replace Abstract: Personal AI agents face a deployment paradox on mobile: persistent background execution drains the battery and conflicts with platform background li

Learning from Next-Frame Prediction: Autoregressive Video Modeling Encodes Effective Representations

arXiv:2512.21004v2 Announce Type: replace Abstract: Recent advances in pretraining general foundation models have significantly improved performance across diverse downstream tasks. While autoregressi

Liked Muse, so I cut the 30B model in half by width, distilled it back, and it does 57 of 60 tool tasks its parent does 60 of

I've liked how Muse-Glimmer worked, so I wanted to see if I could produce a smaller 'kid' out of it. Ornith's sharp decisions on when to think and which tool to call were the other thing I liked, so O

LocUS: Head Selection and Subspace Projection for Targeted Activation Steering

arXiv:2609.31122v1 Announce Type: new Abstract: Activation steering is a powerful training-free paradigm for controlling large language models at inference time. However, standard approaches estimate

local-aiarxiv-cs-cl

LogicTree-RAG: Logic Tree-guided Retrieval-Augmented Generation for Long-form Patent Drafting

arXiv:2609.30943v1 Announce Type: new Abstract: Long-form technical text generation underpins knowledge-intensive workflows, yet remains challenging for large language models (LLMs) due to the need fo

LUMO (Lightweight Unified Multilingual Orchestrator): A Privacy Preserving Offline Voice Assistant

arXiv:2609.30692v1 Announce Type: new Abstract: Reliable voice interaction is essential in environments with limited internet connectivity and strong privacy. However, most existing voice assistants d

macOS 27 ships a free local LLM on Apple Silicon Macs. I made it easy to use from Node and Python

Apple Silicon Macs on macOS 26+ come with a small LLM built in. No download, no API key, and nothing leaves your Mac. Why I built it I was making a tool that writes API docs from code, and I didn't wa

Mica v0.1 4B got diamonds in survival Minecraft on its first run. 26 decisions from an empty inventory.

I've been working on Mica, a 4B decision model, and wanted to see how far it could get in actual Minecraft, not a sim. New world, empty inventory, and the goal was a diamond pickaxe. Last time I poste

Pocket-STVG: lightweight architecture for Spatio-Temporal Video Grounding

arXiv:2609.31135v1 Announce Type: cross Abstract: Spatio-Temporal Video Grounding (STVG) aims to localize the spatio-temporal tube in a video corresponding to a natural language query. While recent me

Region-Level Black-Box Defense Against Stealthy Embedding-Space Backdoors in CLIP

arXiv:2609.31558v1 Announce Type: new Abstract: Contrastive Language--Image Pretraining (CLIP) has emerged as a dominant vision backbone due to its strong transferability and zero-shot capabilities. H

Speculative reward hacking in coding agents

I audited thousands of agent rollouts in DeepSWE-1.1. Over 80% contained reasoning about an imagined grader. Yet no grader/verifier is mentioned in prompts nor accessible to the agents. Agents reasone

Staged Depth Training: A Representation Curriculum for PINNs

arXiv:2609.30299v1 Announce Type: new Abstract: Representation quality is a central determinant of PINNs' performance, yet standard training leaves representations to emerge implicitly while fitting t

Statistical Foundations for a Google Play User-Review Sentiment Index: Signal Fusion, Shrinkage, Distributional Validation, and Dynamic Smoothing

arXiv:2609.31513v1 Announce Type: cross Abstract: We develop a statistically explicit sentiment index for Google Play user reviews and establish the mathematical results supporting its construction. N

local-aiarxiv-cs-cl

The Shape of Events: Edge-Based Inductive Biases via Cross-Domain Distillation

arXiv:2609.30478v1 Announce Type: new Abstract: Convolutional neural networks trained on ImageNet are known to exhibit a strong preference for local high-frequency texture, an inductive bias that tran

To Solve Bilevel Optimization with Nonconvex Lower Levels, We Need Second-Order Stationarity

arXiv:2609.30501v1 Announce Type: new Abstract: Although bilevel optimization (BLO) has emerged as a powerful framework for addressing many complex and nested machine learning problems in recent years

Towards Universal Representation-Based Process Control

arXiv:2609.30790v1 Announce Type: new Abstract: Many temporal process learning and monitoring pipelines operate in local windows, making window-level decisions unavoidable in practice. In such setting

Transformer-based Monte Carlo Localization in Construction Meshes

arXiv:2609.31357v1 Announce Type: new Abstract: To be able to perform inspection or digitization tasks, mobile robots on construction sites must be able to localize themselves reliably with respect to

local-aiarxiv-cs-ro

Up and Down the Abstraction Ladder: Code-Based Skills for Language Agents

arXiv:2609.31076v1 Announce Type: new Abstract: Language agents struggle to act and learn in environments that require long sequences of low-level actions. Code-based abstractions can make these agent

Artist here, Minimax H3 is amazing for creating images for my OCs in ref2v using Fizgig H3 Still custom node.

I am an artist working on a worldbuilding, character design and art toy project, handdrawn using Procreate on iPad with no AI use at all. However, I use AI a lot as a hobby and messing around locally

BFS Best Face Swap & Body Swap Loras By Alissonerdx

https://huggingface.co/Alissonerdx/BFS-Best-Face-Swap/tree/main https://huggingface.co/Alissonerdx/BFS-Best-Face-Swap/blob/main/workflows/Body%20Swap%20Qwen%20Image%202.1%20-%20V1.json https://hugging

Hardware Roofline Inference Calculator

Hello everyone, I made a calculator for the theoretical HW roofline for decoding / prefill based on several parameters (LLM model architecture, quants, GPU, Memory, ..). It still a theoretical bound,

I built Slopus, a free, open-source desktop app for generating and editing AI videos locally (Minimax H3)

Slopus is a desktop app for making AI videos on your own GPU, from the first prompt to the exported file. What it does: Generate scenes shot by shot. You lay out scenes on a board, write each shot, an

Navigating Cost Efficient Hardware in these Volatile Times

Where to even begin on this one...I guess I should start by acknowledging the risk vs reward for vendors other than nvidia, so: Yes I understand that nvidia are dominant currently on speed (llm and im

One .char model, Consistent face, body & cloths: Minimax H3

Hey Guys, Based on my last post on stills, I thought to experiment the same .char model with Minimax H3. I created a character model sia.char which encodes references, prompts around face, body & clot

Prevent CUDA OOM in PyTorch with dynamic lane switching

I built MEM v3 to solve a frustrating problem in PyTorch: CUDA Out-of-Memory crashes during long training and fine-tuning runs. Instead of restarting when memory spikes or keeping batch sizes overly s

Qwen3.8-27B FP8 dual GPUs

Hardware RTX 5090 (32 GB) + RTX 4070 Ti Super (16 GB) = 48 GB VRAM 32 GB DDR5 6200 RAM Arch Linux, KDE on the 5090 (takes ~1-1.5 GB VRAM) daily driver OrcaRouter Qwen3.8-27B Uncensored, block-FP8 (28.

... so, yeah.

Finally got 3.8-Flash-Next running on my M4Pro 48GB Mac with https://huggingface.co/ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-GGUF Dense 3.8-27B is just faster... and maybe better due to quantization lev

SupersonicLabs/Julia-1 · Hugging Face

New open source Jev like model for running on local devices from a group called Supersonic Labs. It's a 144M param local non-generative local classifier. From their site: Julia 1 opens our research in

What do people do with their upscaled images?

I see nearly everyone talks about upscaling their image and video gens to 4K, 8K and perhaps beyond. Now nearly every WF comes with an upscale portion tied to it either after decode or even from the l