Hey folks! I’m just a curious learner wandering into the GPU universe — and wow, it’s a ride. A few weeks ago, I cracked open Chapter 5 of Programming Massively Parallel Processors (yep, the Kirk &...

Hey everyone — welcome to my CUDA notes-turned-blog! I started confused about how GPU threads decide work in warps and ended up experimenting with matrix multiplication, reductions, branch divergen...

🔍 Attention Latency Comparison Version Matmul Kernel Time Overall End-to-End Time Speedup Naive (global memory) ~3.25 ms 38.2 ms ...

HyLoRA is a novel technique for compressing large language models using a hybrid of SVD (for static compression) and LoRA (for dynamic fine-tuning). This project demonstrates a complete pipeline to...

Before diving into the discovery process, here’s a quick look at the numbers that show just how effective tiling can be for matrix multiplication on GPUs: Test Configuration: GPU: Colab T4 M...

Project Overview Goal: Build a specialized AI model that reads meeting dialogues and extracts action items into a clean, structured JSON format. Value Proposition: Transforms unstructured meeting c...