PinnedInThe Synaptic StackbyDeepak kumar sahoo·Sep 9, 2025FlashAttention-3: The Engine Powering Next-Gen LLMsHow a smarter attention algorithm, built on cutting-edge GPU tech, is making AI faster, cheaper, and more efficient than ever.
PinnedInThe Ancient CompassbyDeepak kumar sahoo·Sep 15, 2025Whispers of the Hills: A Pilgrimage to Vallimalai Murugan TempleTucked away in the lush green embrace of the Eastern Ghats, just a short drive from the bustling town of Vellore, lies a hidden spiritual…
PinnedInThe Synaptic StackbyDeepak kumar sahoo·Sep 3, 2025How Fused Kernels Are Powering the LLM Revolution (And Why You Should Care)How a simple idea doing more work in less time became the secret sauce behind fast AI inference.A response icon2A response icon2
InThe Ancient CompassbyDeepak kumar sahoo·Sep 15, 2025Vellore Fort: A Sentinel of South Indian HistoryNestled in the heart of Tamil Nadu, the city of Vellore is more than just a bustling urban center, it is a custodian of centuries-old…A response icon1A response icon1
InThe Synaptic StackbyDeepak kumar sahoo·Sep 10, 2025What Is Plan-and-Solve Prompting?The Plan-and-Solve prompting style is a two-phase framework designed to guide LLMs through complex tasks by first creating a clear roadmap…A response icon1A response icon1
InThe Synaptic StackbyDeepak kumar sahoo·Sep 7, 2025TMA: How Tensor Memory Accelerator is Supercharging AI Kernels ?If WGMMA is the engine, TMA is the fuel pipeline, invisible, essential, and quietly transforming how data flows in modern AI.
InThe Synaptic StackbyDeepak kumar sahoo·Sep 7, 2025Direct Preference Optimization (DPO): An In-depth AnalysisDPO is a highly influential and widely recognized fine-tuning method that played a pivotal role in advancing language model alignment…
InThe Synaptic StackbyDeepak kumar sahoo·Sep 6, 2025WGMMA: How Warpgroup Matrix Multiply-Accumulate is Revolutionizing AI Compute -From Transformers to…If deep learning runs on matrix multiplication, then WGMMA is its new turbo engine.
InThe Synaptic StackbyDeepak kumar sahoo·Sep 2, 2025Radix Attention: How a Data Structure Inspired a Revolution in Efficient Transformer InferenceIn the rapidly evolving world of artificial intelligence, especially in the domain of large language models (LLMs), efficiency is king. As…A response icon2A response icon2
Deepak kumar sahoo·Aug 31, 2025Gen AI Evaluation: Foundational Principles (Part -1)This is a follow up to its previous article, if you’ve not read it already, then would recommend to read it first. Here’s the link