AI Wiki — Knowledge Base

R9V Update: Created and adopted KVA projections based on Deepseek V4.1 Flash + HySparse2/MiMo-V3 for Qwen3.8 Flash Next. This is a game changer for models that don't natively implement it. 1.45-1.85x speedup in prefill to 3k+ at a small deficit to perplexity. [2x R9700, 128GB DDR5]

Here's my *first* implementation of KVA projectors on QFN (just the uncensored model for now) the highlights are basically as follows for using the projectors at each different layer: Starting at laye

model-releasesr-localllama

24 Sept 2026

Search: “benchmark”

Evaluating Feedback Focus and Pedagogical Adaptivity in LLM-Generated Feedback on Student Writing

FLEET: From Logits Entropy to Enhanced Trajectories in Text Generation

From LiDAR Maps to Visual Localization: Unified Visual Association for Robust Point-Line-Plane Pose Estimation

Geometry-Conditioned Visual Place Recognition in Natural Environments

Groundbench: Multi-Resolution Polygon Grounding Exposes the Geometry Gap in Vision-Language Models

Guides That Cause Actions: An Offline Study of Guide-Action Mutual Reinforcement in Multimodal Web Agents

How Much Were You Told? Measuring External Information in Peer Reviews

ICM: Intra-class Mixing for Domain Adaptation in Adverse Weather

Improving LLM-based Autonomous Web Agents with Filtering

InGuard: Towards Generalized Inner Guardrail for Safe Text-to-Image Generation

Interesting new paper from Google DeepMind. Studies agents that follow bad advice from users. Users often suggest a fix that sounds right an…

InternW0: A Foundational Physical World Model for Efficient Real-World Interactions

Inverse Problems Conditioned on Observation Ensembles: Applications and Methods

Just trained Tev1 0.8B, a tiny Jev-like classifier. Here it is running completely locally on my mac with @ollama & classifying some tasks. I…

KATOsuper: Surrogate-accelerated neural topology optimization with sensitivity-consistent Fourier neural operators

Learning Holographic Reduced Representations with Clifford Variational Autoencoders

Learning Local Heterogeneity and Cross-Region Context for Large-Scale Traffic Forecasting

Learning Risk Scores Robust to Unobserved Confounders

Learning Spectral Allocation: A Fractional Diffusion Framework for Adaptive Volumetric Segmentation

LEGO: Synergizing Expert GraphRAG and Expert Chain-of-Thought for Legal Reasoning

Less Language, More Latents: Annotation-Efficient VLAs for Driving

LiAM-SAM: Lifecycle-Aware Memory for Robust SAM2-Based MOT

Limiting-Kernel Q(lambda): Bridging Short and Long Horizons

LOCKR: A Hidden-State Trajectory-Guided Planner for Detecting and Repairing Stable-but-Wrong Lock-In in Diffusion Language Models

Meet, Compare, or Abstain: LatWeave for Deterministic Multi-Hop Question Answering on Knowledge Lattices

MENO: Memory-Efficient Neural Operator

MessyKitchens: Contact-rich object-level 3D scene reconstruction

MINER: Multi-crop INference-time Enhancement for Rare-Object Retrieval with Frozen Dual Encoders

MORSE: Multi-Context Ordering via Reverse Scoring for Evidence-Preserving Compression

NADI 2026: The Second Multidialectal Arabic Speech Processing Shared Task

NV-Reason-CT: 3D Visual Language Model for CT Analysis

Optimization without Future Compromises? Decentralized Coordination via Collective and Reinforcement Learning

PBLH Estimation from Satellite Radiances via a Dual-Encoder Transformer

PCQC: Privileged Counterfactual Question Credit for Multi-Turn Medical Dialogue

Pheno-GS: Phenoscape-scale Geodesic Sinkhorn

Planned Test-Time Scaling with Coordinated Reasoning Paths

Pose-Aware Multimodal Automatic Tagging for Greek Traditional Music

Power your agents: Gemini 3.8 Live with Live Avatar is now generally available

Proactive Defense: Hardening Code Pipelines and CI/CD Infrastructure

Quantifying the Occult: A Comparative Study of Hindu and Buddhist Deities Using Machine Learning Methods

Quantization-Robust Unlearning through the Lens of Retain-Forget Loss Landscapes Interaction

QUARTET: Quad-branch cross-Attention and Random-walk Traces for Enhancing Transformers on Relational Graphs

Qwen3.8-Flash-Next on 12GB VRAM - 65 tokens per second

R9V Update: Created and adopted KVA projections based on Deepseek V4.1 Flash + HySparse2/MiMo-V3 for Qwen3.8 Flash Next. This is a game changer for models that don't natively implement it. 1.45-1.85x speedup in prefill to 3k+ at a small deficit to perplexity. [2x R9700, 128GB DDR5]

Reinforcement Learning with Decomposed Subtasks

Robust Adversarial Reinforcement Learning with Risk Sensitivity and Critic Consistency Regularization

Ruby-ASR: Evidence-Preserving Supervision for Joint Orthographic and Lexical-Reading Recognition

S2A:Semantic-to-Spatial Alignment for Alignment-Free RGB-T Salient Object Detection