AI Wiki — Knowledge Base

R9V: A designer set of kernels I've been working on for R9700s/RDNA4. Qwen3.8-Flash-Next Unsloth IQ4_XS (w/ TP on 2 R9700s, MTP, SSD n-gram, 128k ctx, vision): TG256 of 78 tok/s (~3x increase), PP8192 of 1510 tok/s (~30x increase).

TL;DR: Ninfer/DS4 but for RDNA4 Highly custom kernels built for RDNA4, applied to vLLM-Radiance to greatly improve Qwen3.8 Flash Next speeds. This is mostly for dual R9700s with preferably 48GB of RAM

model-releasesr-localllama

31 Aug 2026

Search: “routing”

CompareBench: A Benchmark for Visual Comparison Reasoning in Vision-Language Models

DensityKV: Density-Guided KV Cache Compression for Long Video Generation

Embedding Models for Stance-Aware Argument Retrieval

Emergent aggregation from collective foraging

Exact Risk Ratios for Weighted Data Selection in Linear Regression

HARTS: Efficient Agentic Reinforcement Learning for Hybrid-Attention Models over Arbitrary Rollout Trees

Manage agents, tools and skills at scale with AWS Agent Registry

Plan Along the Way: Event-Triggered Foundation-Model Planning for TAMP Execution in Partially Observable Manipulation

R9V: A designer set of kernels I've been working on for R9700s/RDNA4. Qwen3.8-Flash-Next Unsloth IQ4_XS (w/ TP on 2 R9700s, MTP, SSD n-gram, 128k ctx, vision): TG256 of *78 tok/s* (~3x increase), PP8192 of *1510 tok/s* (~30x increase).

Revisiting Local Context for Long-Horizon Streaming 3D Reconstruction

Task-State Adaptation with Prototype Memory for Multi-Task Dense Prediction

ZipMVS: Multi-View Stereo with Compressed Cost Volumes

llama.cpp Open PRs list - CPU/RAM/Disk/Hybrid Related - Better for CPU-only & Hybrid inference

You do not need frontier models for everything. For example, open models are great for automations. If you are not sure where to adopt open …

4DSynth: Controllable Procedural World Synthesis for Dynamic Embodied Simulation

Accelerating Scientific Research with Gemini in the Real-World

Agentic AI for operating scientific instruments for nanoscale characterization

Andreessen Horowitz raises $1.1B AI infrastructure fund for chips, robots and more

AraMS-28k: The Largest Publicly Released Line-Level Dataset of Historical Arabic Manuscripts with Margin and Insertion-Anchor Annotations

CIFQA: A Deterministic Tool-Grounded Multi-Agent LLM Framework for Financial Query Answering

Compositional Online Learning for Semantic Data Processing Systems

Cross-simulator transfer with foundation model summaries: Towards robust SKA-era reionization inference

Explainable Artificial Intelligence for Customer Churn Prediction in Telecommunications: A Framework for CRM Integration

G2D: Generative-to-Discriminative Collaborative Inference for Zero-Shot Image Classification

Magnon-induced phononic Chern insulator

MM-Spectrum: Multimodal Multi-spectral Molecular Structural Elucidation with a Stable MoE Framework

PICasso: An AI-Enabled Design Framework for Autonomous Optimization of Silicon Photonic Devices

SCIT: Testing Causal Cache Carriers in Latent Chain-of-Thought Models

SOLO: Stable Omni-terrain Long-Horizon Perceptive Humanoid Locomotion

Successive Capacity Growth: Task-Complexity-Driven Width and Depth Expansion for Vision Transformer Encoders in JEPA World Models

SymbolLKG: Towards Verifiable Logical Reasoning via Logical Knowledge Graph and Symbolic Solvers

The DeepSeek harness is engineering at its finest. 202K stars on GitHub. Damn! Clearly, they built this harness with future AI needs and cap…

The Latent Diagnostic Taxonomy: A Framework for Constructing Classifiers and Diagnosing Their Decisions, Applied to Prompt Injection Detection

The Reasoning Tax: Token Economics of LLM Reasoning Across Task Types and Deployment Contexts

The Thousand-Graph Hypothesis: A Testable Hypothesis of Task-Conditioned Relation Materialization in Repository-Level Code Reasoning

Today I hit 181 toks/s (aggregate) on Qwen3.8-Flash-Next on 2x DGX Sparks

Towards a universal meta-optics solver via large language models

Towards Expert Financial QA via Self-Improving RAG

TTPO: Test-Time Policy Optimization

UrbanGround: From Local Perception to Spatial Agency in a Real-Scale City

Vagdhenu: A Vrutta (Meter) Aware Shloka-to-Chant (TTS) System for Sanskrit

Who Remains, What Changes: Identity Anchored Composed Gait Retrieval

A Constitutive Markov Physics-Informed Neural Operator (MPNO) for Autoregressive Stability in Transient Dynamics

Addressing the Reasoning Gap: Mechanistic Circuit-Based Knowledge Editing in Large Language Models

AGRO-Nav: Autonomous Graph-based Orchard Navigation

AI/ML Life Cycle Management for Interoperable AI Native RAN

b10660

b10664

R9V: A designer set of kernels I've been working on for R9700s/RDNA4. Qwen3.8-Flash-Next Unsloth IQ4_XS (w/ TP on 2 R9700s, MTP, SSD n-gram, 128k ctx, vision): TG256 of 78 tok/s (~3x increase), PP8192 of 1510 tok/s (~30x increase).