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: “benchmark”

Dynamic Alignment Compensation for Hallucination Mitigation in Large Vision-Language Models

FedEHR-Agents: Federated Agentic Optimization for Automated EHR Modeling

FENCE: A Financial and Multimodal Jailbreak Detection Dataset

Fidelity Is Not Enough: Dispatch-Level Instrumentation for Agentic Datasheet Extraction

GLM 5.3 and GLM 5.3 Flash ran locally on RTX PRO 6000 WS and built a penthouse using BlenderMCP

GraspHOI: Full-Body 3D Human-Object Reconstruction with Finger-Level Grasps from a Single In-the-Wild Image

I am the product lead for the training platform at @FireworksAI_HQ . Today the Training API and Fireworks Lab are GA. I won’t walk you throu…

INSPIRE: An Internalize-Then-Improve Approach for Example-Driven Mathematical Reasoning

Knowing Before Answering: Decoding Language Models for Reliable RAG

Learning Fast Monomial Orders for Grobner Basis Computations

Learning the Target Priors Before Image Translation: A Decoupled Training Paradigm for Cross-Modal Image Translation in Remote Sensing

Learning to Difference: Adaptive Reversible Differencing (AdaRDiff) for Time Series Forecasting

Localizing Global Discrepancies: Marginal Contributions and Contextual Anomaly Detection

Long Story Short: Story-level Video Understanding from 20K Short Films

Lossy Event Compression: From Event Stream Distortion to Task Performance

Manifold4D: Denoising on Point Cloud Rendered Manifolds for Video Re-shooting

Marginal Coverage Credit Reduces Redundant Exploration in Parallel State-Entropy Optimization

PACE: Publisher-Adaptive Content Extraction via Agentic Automation

PanelShield: Verifiable Closed-Loop Safe Planning for Robotic Industrial Panel Operation

PCBnet: A Dataset and Automatic Construction of SPICE Netlists from Schematic Images

PRISM: Agentic Retrieval with LLMs for Multi-Hop Question Answering

Probabilistic Multi-Robot Gas Source Localization with Uncalibrated Sensors: A Distributed Estimation Approach

QGPINNs: A Physics-Informed Neural Network Framework for Nonlocal Differential Equations on Quantum Graphs

Quantum SEDONet: Spectrally-Embedded Quantum Deep Operator Networks for Partial Differential Equations

Qwen3.8-Flash-Next in llama.cpp from CPU-only to 96GB VRAM: 8.5 to 109 tok/s, max context and parameters test. My findings on RTX 6000 PRO.

Qwen3.8-Flash-Next-NVFP4 vs Qwen3.8-27B-FP Test Results

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).

RASA: Disentangled Spatial-Motional Priors for Cross-Identity Character Animation

RecourseBench: A Modular Framework for Reproducible Algorithmic Recourse Evaluation

Residual-Guided Randomized Neural Networks

Retrieving Relations, Detecting Fallacies: A RAG Approach to Political Debate Analysis

Revisiting Local Context for Long-Horizon Streaming 3D Reconstruction

Run NVIDIA BioNeMo NIM Microservices for Protein Structure Prediction in Claude Science

SCALE: Self-uncertainty Conditioned Adaptive Looking and Execution for Vision-Language-Action Models

SegBench-GC: Testing Segmentation Invariance in Multi-Step Offline Goal-Conditioned Reinforcement Learning

Select, Don't Train: The Benefits of Modular Entity Disambiguation with LLM-Based Selection

Select, Label, Evaluate: Active Testing in NLP

Semantic Head Specialization Guides Hybrid ViT Attention for Multimodal LLMs

SinkSLOT: Sinkhorn via Sparse Lifted Optimal Transport

SOMTab: Set-Order Mamba for Efficient Tabular In-Context Learning

Steering Multimodal Large Language Models Decoding for Context-Aware Safety

SUFLECA: Scaling Up Feature Learning for CAD-to-image Alignment

SURE-Challenge: Evaluating Speech Evidence Before Speech-LLM Generation

Tensor-Accelerated Eager Multi-Resolution Grids for Evolving Large-Scale Substrates

The Autonomy Tax: Defense Training Breaks LLM Agents

Today, Transluce is releasing the most expansive independent evaluation to date of how AI systems respond to users experiencing mental healt…

Token-Budget Distillation: Transferring Full-Token Semantics to Compressed Video Vision-Language Models

Trajectory-Level Speculative Decoding for Diffusion Language Models

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).