Anuva Sharma·2d agoSwin Transformer: How Shifted Windows Made Vision Transformers PracticalIn late 2020, the Vision Transformer (ViT) showed that a plain Transformer, the architecture behind modern language models, could classify…
Anuva Sharma·3d agoPredict Meaning, Not Pixels: A Visual Guide to the JEPA FamilyHere is a puzzle. Below are the training losses of two self-supervised runs with the same encoder, the same data and the same optimizer…
Anuva Sharma·5d agoHow DINO Teaches Vision Transformers to See Without LabelsLook at where the network pays attention in the clips above. It was never told what a ship, a bike or a dog is. It was never trained to…
Anuva Sharma·May 30The Journey to Multi-Head Latent AttentionWhy DeepSeek-V2 invented a strange-looking attention block — and how a tiny algebraic trick made the KV cache 57× smaller without dropping…A response icon5A response icon5
InPython in Plain EnglishbyAnuva Sharma·May 28From Integers to RotationsA visual journey through positional encodings — and why every step from “just add the number” to RoPE was inevitable
Anuva Sharma·Nov 4, 2025Beyond Output Quality: The Complete Guide to Evaluating AI AgentsSo you’ve built an AI agent. It’s answering questions, routing requests, maybe even processing refunds. Everything looks good… until you…
Anuva Sharma·Feb 24, 2025Exploring the Future of AI : Multi-Agent SystemsThe world of artificial intelligence (AI) is evolving at an unprecedented pace. One of the most exciting developments in AI is the concept…
Anuva Sharma·May 9, 2024Batching : Improving LLM ThroughputWelcome back to the second installment of our series on enhancing Large Language Model (LLM) performance. If you missed the previous…