Engineering
Realtime-Venus: A full-duplex interaction system with asynchronous delegation
Realtime-Venus is a full-duplex conversational AI system built from two separately trained 9B models — one for audio-visual interaction and…
Engineering
Flash-dLLM: IO-Aware KV Caching and Parallel Decoding for Fast, Memory-Efficient Diffusion LLMs
Flash-dLLM is a training-free inference acceleration framework for diffusion large language models that identifies GPU memory I/O, rather than compute,…
Engineering
SiliconBench: Speed, Memory, and Fidelity for LLM Serving on Unified-Memory Desktops
SiliconBench benchmarks nine LLM serving engines on Apple Silicon unified-memory desktops across speed, memory headroom, and output fidelity rather than…
Engineering
Agensh: Scaling Organizational Intelligence to 1,024 Agents
Agensh is a multi-agent harness that removes the central-orchestrator bottleneck limiting existing multi-agent frameworks by having concurrent workers self-organize —…
Engineering
Recursive self-improvement of AI research agents
The paper presents AIDE^2, a system in which an AI research agent recursively rewrites and benchmarks its own code, keeping…
Engineering
Bellman Policy Optimization
Bellman Policy Optimization (BPO) is a critic-free reinforcement learning method for LLM reasoning that reformulates Policy Mirror Descent as a…
Engineering
StableVQ: Practical Guidelines for Stable Vector-Quantized Tokenizer Training
StableVQ addresses training instability in vector-quantized visual tokenizers by arguing the root cause is entanglement between encoder-decoder and codebook training,…
Engineering
JEV-as-a-Judge: Accept When Confident, Escalate When Unsure
JEV-as-a-Judge is a decision-only LLM evaluation model designed as a cheap first-pass filter, tested against sixteen generative and reward-model judges…
Engineering
Measuring the Checker: Mutation Analysis for GPU-Kernel Benchmark Oracles
This paper introduces mutation analysis to measure whether GPU-kernel benchmark correctness checkers — which now feed leaderboards and RL reward…
Engineering
JEPA-Anything: Learning Predictive Models across Different Worlds
Researchers introduce JEPA-Anything, a domain-agnostic world-modeling framework built on orthogonal predictive factorization (OPF), which extends joint-embedding predictive architectures by decomposing…
Engineering
Video DeltaNet: A Video-Native Hybrid Attention for Livestream Video Generation
Researchers present Video DeltaNet (VDN), a hybrid attention architecture for video diffusion models that combines local softmax attention with a…
Engineering
Register Tokens for Bounded-State Reasoning in Diffusion Language Models
Researchers propose register tokens — dedicated fixed-position tokens whose hidden states are trained to carry reasoning progress — to let…