@Apurvak3

FlashAttention: Making Transformers Faster and More Memory-Efficient

Dec 26, 2025 · 5 min read · Large Language Models (LLMs) like GPT, BERT, and modern Transformers rely heavily on the self-attention mechanism. While powerful, self-attention is also the biggest performance bottleneck when working with long sequences. In 2022, Tri Dao and collab...

🧠 Transformers Explained Simply: From Word2Vec to Multi-Head Attention

Nov 2, 2025 · 5 min read · # A deep dive into the paper “Attention Is All You Need” and how modern NLP models like BERT and GPT evolved from it. 🌟 Introduction If you’ve ever wondered how models like BERT, GPT, or T5 understand language, the answer lies in one architectur...

# 🧠 Neural Machine Translation of Rare Words with Subword Units —

Oct 29, 2025 · 4 min read · By Apurva KanthPublished on Hashnode 💭 The Problem: When Neural Networks Don’t Know a Word Traditional Neural Machine Translation (NMT) systems work with a fixed vocabulary — often limited to around 30,000–50,000 words. But languages are full of ra...

🧬 C2S-Scale 27B: When AI Learns the Language of Cells

Oct 17, 2025 · 4 min read · 🧬 C2S-Scale 27B: When AI Learns the Language of Cells Working in the AI and automation space, I’ve seen how scaling models transforms their reasoning capabilities. But applying those same scaling principles to biological systems was something I neve...

🎬 Building a Context-Based Movie Recommendation System using Cosine Similarity

Oct 12, 2025 · 3 min read · 🧠 Introduction Have you ever wondered how Netflix or Prime Video knows exactly what you might like next?In this blog, I’ll walk you through how I built a content-based movie recommendation system using Python, Pandas, and Cosine Similarity — from da...