Knowledge catalogue
Its chat template has this line: {%- set resolved_reasoning_effort = reasoning_effort|default('xhigh') %} xhigh is the most expensive of its three settings (low / medium / xhigh). If you never set one
arXiv:2609.04705v1 Announce Type: cross Abstract: Passive cooling eliminates the energy overhead and mechanical failure modes of fans, making it attractive for edge deployment, yet sustained Deep Neur
Github link: https://github.com/thatblend/LLMPSP I wanted to see what the PSP can theoretically handle and I got my answer - a 90M model is about the max it can do without atrocious inference speeds.
Moving Large Language Models (LLMs) from experimental prototypes into enterprise production exposes a critical truth: your infrastructure dictates both your performance ceilings and your unit economic
arXiv:2609.02186v1 Announce Type: cross Abstract: Quantum generative models offer a promising framework for exploring whether quantum computation can enhance generative machine learning. Flow matching
Years ago I built a server, shelved it after some time, but recently realized I have it sitting there. I'm looking to pop in a Tesla V100 32GB card (with proper cooling). But are the component too old
arXiv:2609.00058v1 Announce Type: cross Abstract: Developing high-performance CUDA kernels demands specialized knowledge in algorithm implementation, correctness validation, and hardware-aware paralle
I’m working on performance regression detection using machine learning/anomaly detection. My setup is basically: Healthy runs are used to learn normal behaviour Regression runs are used to see whether
arXiv:2608.11226v1 Announce Type: new Abstract: Reinforcement-learning post-training dominates modern language-model development, yet its power behavior on GPU hardware has not been characterized, and
arXiv:2608.08081v1 Announce Type: cross Abstract: Large mixture-of-experts (MoE) language models with 26--120 billion parameters exceed the memory capacity of consumer devices through three simultaneo
Securing infrastructure and services against a future cryptographically-relevant quantum computer has been a goal for Google for a decade, and we’ve dedicated ourselves to help developers by advancing
Hi everyone, I'm a systems analysis student researching a problem a lot of you probably know well: how much you actually trust the published VRAM/RAM requirements for open-source models before trying
arXiv:2608.03649v1 Announce Type: new Abstract: Fewer visual tokens do not guarantee lower end-to-end latency. We evaluate break-even with a reproducible protocol that accounts for decision overhead,
I let Gemma4-31b run on my laptop for like almost a day using a heavily altered pi to do a deep dive on our beloved Llama tangentially related Subreddit, and this was the conclusion. Feels pretty accu
arXiv:2504.03772v3 Announce Type: replace-cross Abstract: Respiratory diseases account for a significant portion of global mortality. Affordable and early detection is an effective way of addressing t
arXiv:2607.20162v1 Announce Type: new Abstract: The growing demand for intelligent processing at the edge of IoT networks is constrained by the severe computational and memory limitations of microcont
The math behind reinforcement learning (RL) post-training for large language models (LLMs) is notoriously unforgiving. As frontier AI labs push the boundaries of reasoning and coding models using RL p
arXiv:2607.14021v1 Announce Type: new Abstract: Dexterous manipulation remains a critical bottleneck in industrial automation; tasks such as cable routing, connector insertion, and precision assembly
arXiv:2607.13897v1 Announce Type: new Abstract: The broadcast nature of wireless channels exposes radio-frequency (RF) networks to anomalous and malicious transmissions, making anomaly detection a fun
arXiv:2607.06915v1 Announce Type: new Abstract: Scaling down the resolution of input images can greatly reduce the computational overhead of convolutional neural networks (CNNs), which is promising fo
arXiv:2607.02753v1 Announce Type: cross Abstract: Neuromorphic controllers for size, weight, and power-constrained systems require neural architectures that are both energy-efficient and interpretable
Hugging Face Jobs integration with LeRobot enables users to run robotics tasks and training jobs on Hugging Face's infrastructure, streamlining the deployment and execution of robot learning models. T
OpenAI is releasing some sort of device related to its AI-powered coding tool, Codex, on July 15th. In a video posted to X on Monday, OpenAI shows a square-shaped device with several buttons, alongsid
arXiv:2606.21868v1 Announce Type: new Abstract: Modern Mixture-of-Experts (MoE) models place most of their parameters in expert layers, yet only a small fraction of those experts are used for any toke
Protecting sensitive data used with AI is a critical part of our commitment to providing advanced and secure cloud infrastructure. Confidential Computing cryptographically protects data in use in hard
arXiv:2605.09848v1 Announce Type: new Abstract: Electrocardiogram (ECG) interpretation is essential for diagnosing a wide range of cardiac abnormalities. While deep learning has shown strong potential
arXiv:2603.07053v2 Announce Type: replace Abstract: Scientists face significant visualization challenges as time-varying datasets grow in speed and volume, often requiring specialized infrastructure a
arXiv:2604.02811v2 Announce Type: replace-cross Abstract: Functional verification consumes over 50% of the IC development lifecycle, where SystemVerilog Assertions (SVAs) are indispensable for formal
arXiv:2406.06543v2 Announce Type: replace-cross Abstract: Deep learning has driven significant technological advancements, but its high energy consumption limits its use on battery-operated edge devic
Less than a year ago, Apple made headlines for a lack of AI announcements at its annual WWDC event. Ten months later, the company has announced that hardware executive John Ternus will succeed longtim
As the automotive landscape accelerates toward software-defined vehicles, Cockpit Domain Controllers (CDCs) are becoming the core of next-generation in-cabin experiences. The ability to rapidly develo
Figure 1: CUDA-to-MLX optimization translation map. CUDA optimization knowledge can be translated into architecture-native MLX strategies rather than copied instruction-for-instruction. We face a new
Great report from Google and MIT. Co-designing ML models and the accelerators is an exciting research direction. This report introduces Coco, an agent platform TPU architects use for hardware and mode
arXiv:2610.03972v1 Announce Type: new Abstract: Ising machines have emerged as promising hardware accelerators for intractable optimization and sampling problems, yet their practical impact increasing
Note: This writeup was put together with the help of AI. So I've got dual AMD MI50 32GB cards. If you know these cards, you know the story - AMD dropped official ROCm support for gfx906 after ROCm 5.7
I am planning to buy hardware to run local models. I'm not sure what kind of hardware is best and what will I be able to run. I'm torn between getting a used M1 Max 64gb Macbook for €1500, M3 Max 128g
Silicon chip startup xMEMS Labs Inc. today said it has started mass production of its XMC-2400 and XMC-4800 chips for cooling artificial intelligence hardware. No one has commercialized solid-state ac
arXiv:2609.36954v1 Announce Type: cross Abstract: Distributed inference depends on GPU collective communication that must keep pace with evolving hardware and specialized workloads. However, existing
NVIDIA Confidential Computing (CC) secures large‑language‑model inference by providing encrypted virtual machines, GPUs, and NVLink, enabling processing of sensitive data in trusted hardware. Running
arXiv:2607.26661v1 Announce Type: new Abstract: Ascend C operator optimization is critical for NPU (Neural Processing Unit) inference performance but requires deep hardware expertise.While large langu
arXiv:2605.26118v1 Announce Type: cross Abstract: Porting deep learning algorithms to new hardware accelerators requires developers to repeatedly apply the same low-level optimizations -- quantization
arXiv:2505.13741v2 Announce Type: replace Abstract: Direct training of Spiking Neural Networks (SNNs) on neuromorphic hardware can greatly reduce energy costs compared to GPU-based training. However,
arXiv:2605.07782v1 Announce Type: new Abstract: LLMs can generate hardware descriptions from natural language specifications, but the resulting Verilog often contains width mismatches, combinational l
arXiv:2610.10311v1 Announce Type: new Abstract: Emerging computer hardware often trades reliability for energy efficiency; here we show that large-language models (LLMs) can be trained to tolerate thi
arXiv:2606.10294v1 Announce Type: cross Abstract: Deploying neural networks on unconventional hardware demands architectures that co-optimize task accuracy and platform-specific constraints such as en
arXiv:2510.16028v4 Announce Type: replace-cross Abstract: Neural networks increasingly run on hardware outside the user's control (cloud GPUs, inference marketplaces). Yet ML-as-a-Service reveals litt
This article from SemiAnalysis tracks the performance evolution of DeepSeek V4, a 1.6 trillion parameter model, over its first 43 days of training on Huawei and NVIDIA hardware including GB300 NVL72,
arXiv:2602.07518v2 Announce Type: replace-cross Abstract: Kolmogorov-Arnold Networks (KANs) shift neural computation from linear layers to learnable nonlinear edge functions, but implementing these no
arXiv:2604.24492v1 Announce Type: cross Abstract: Designing deep networks that meet strict latency and accuracy constraints on edge accelerators increasingly relies on hardware-aware optimization, inc
Apple announced that Tim Cook will step down as CEO on September 1, 2026, ending his 15-year tenure, with John Ternus, the senior vice president of hardware engineering, succeeding him as chief execut
In capital markets, operational success is built on predictable performance and low latency, where even tiny variations in packet timing alter queue priority and trade execution. At the same time, bot
NVIDIA DOCA AI agent skills supply verified API signatures, hardware capability checks, and build constraints that let agents fully understand the DOCA library (Flow, GPUNetIO, PCC, RDMA) for BlueFiel
arXiv:2609.34683v1 Announce Type: new Abstract: The optimization of LLM serving engines, such as vLLM and SGLang, is largely benchmark-driven: optimizations, scheduling policies, hardware and system d
In real-time streaming, every millisecond counts. For example, at 30 frames per second (fps), developers have a strict frame budget of just 33.3 ms (and only 16.6 ms at 60 fps) to ingest camera frames
arXiv:2609.17730v1 Announce Type: cross Abstract: Approximate multipliers can reduce hardware area and energy consumption in Deep Neural Network (DNN) inference; however, they introduce computational
arXiv:2608.26418v1 Announce Type: cross Abstract: Modern AI workloads and the hardware that runs them evolve on different timescales: architectural definition precedes volume silicon by years, while t
The internet connected billions of people and mobile devices, putting computers in every hand. Now, we’re in the middle of the next big technology shift, deploying millions of autonomous AI agents to
Yet another vLLM fork thread here, but this time its for older INT8-centric hardware. This is a complete INT8 serving stack for Qwen3.8 27B based on vLLM, AITER, and a 27B GPTQ INT8 quant w/ DFlash2.
arXiv:2608.20725v1 Announce Type: cross Abstract: Convolution is a principal computational bottleneck in deep neural networks, and its efficiency depends on tight integration between algorithms and GP
arXiv:2606.01339v1 Announce Type: cross Abstract: Long-term time-series forecasting needs models that are accurate yet efficient enough for commodity hardware. Lightweight linear forecasters are remar