Tag
29 articles
This article explains tile-based GPU programming concepts, focusing on NVIDIA's cuTile and Triton frameworks, and how they enable efficient Flash Attention in large language models.
NVIDIA introduces Nemotron-Labs-3-Puzzle-75B-A9B, a compressed hybrid MoE LLM delivering 2.03x server throughput, leveraging hardware-aware compression and knowledge distillation.
This article explains how artificial intelligence and machine learning optimize retail pricing and promotional strategies, using laptop sales as an example of sophisticated data-driven decision making.
Meta's new non-invasive brain-to-text AI system, Brain2Qwerty v2, translates brain activity into typed sentences without requiring surgery. The technology is advancing rapidly, with AI optimization playing a key role.
OpenAI has reportedly cut inference costs for its AI models by more than half, significantly reducing the number of GPUs needed to process ChatGPT responses.
Researchers at UC San Diego introduce DFlash, a new speculative decoding technique that drafts whole token blocks in parallel, achieving up to 15x throughput improvement on NVIDIA Blackwell.
Learn how FAPO, a new AI tool from Cisco, automatically improves AI prompts by analyzing each step of a task to make AI systems more accurate and reliable.
As KV cache memory outpaces model weights in large language models, three compression techniques—TurboQuant, OSCAR, and EpiCache—are emerging as key contenders. While each offers distinct methods for optimization, they are seen as complementary rather than competitive.
Learn how to create and apply SkillOpt Markdown files to dramatically improve AI agent performance on procedural tasks, boosting models like GPT-5.5 by 23 points.
Researchers introduce GEPA, a reflective prompt-evolution framework that enhances small language models' performance on multi-step arithmetic problems through structured feedback and multi-component prompt design.
This explainer explores how AI-driven smartphone optimization works and why specific settings significantly impact system performance and user experience. It covers machine learning architectures, data collection, and privacy implications.
This article explains how Google DeepMind's Gemma 4 QAT checkpoints, particularly the Q4_0 and mobile formats, optimize large language models for edge deployment by reducing memory usage and computational requirements through advanced quantization techniques.
