PinnedPriyanshu maurya·Jul 7, 2024Understanding Shannon Entropy and KL-Divergence through Information Theory“ Information theory gives us precise language for describing a lot of things. How uncertain am I? How much does knowing the answer to…
Priyanshu maurya·Jul 1Speculative Decoding- Basics, DFlash and DeepSeek’s DSparkRecently, DeepSeek introduced a new framework called DSpark, which significantly speeds up how fast their Large Language Models (LLMs) can…A response icon1A response icon1
Priyanshu maurya·Jun 9Training Models on Noisy LabelsWe already know that human-annotated datasets usually contain noise, such as missing or incorrect labels. [USE THIS FRIENDS LINK TO…A response icon1A response icon1
Priyanshu maurya·Jul 31, 2025The Surprisingly Simple Trick to Remove Normalization from TransformersA breakdown of the “Transformers without Normalization” paper, which challenges a fundamental building block of modern deep learning.
Priyanshu maurya·Jul 29, 2025From GRPO to GSPO: Fixing a Subtle Flaw in GRPO for More Stable LLM TrainingA deep dive into the Qwen team’s new algorithm, why it’s a significant leap forward from GRPO, and what it means for the future of…
Priyanshu maurya·Jul 11, 2025Is Summed Group ReLU a Thing !This document chronicles an exploration that began with a simple, curiosity-driven question: Can we design a new neural network activation…
Priyanshu maurya·Mar 5, 2025All You Need to Know About ‘DeepSeek AI’s Open Source Week’ — And Why It’s a Game ChangerDeepSeek AI’s recent “Open Source Week” wasn’t just another series of code releases; it was a strategic unveiling of powerful technologies…A response icon1A response icon1
Priyanshu maurya·Mar 1, 2025Chain-of-Draft: The Simple Yet Powerful Alternative to Chain-of-ThoughtREAD FOR FREE HEREA response icon1A response icon1
Priyanshu maurya·Feb 27, 2025Diffusion LLMs: The Next Big Leap in Language AI?READ FOR FREE HEREA response icon1A response icon1
Priyanshu maurya·Feb 20, 2025“Curse of Depth” in Large Language Models: How to Make Every Layer CountIn the rapidly evolving world of Large Language Models (LLMs), bigger isn’t always better — at least not in the way we traditionally think…