Rohan Singh's Weblog

Bookmarks

• Olmo 3 and the Open LLM Renaissance

  As the capabilities of large language models (LLMs) have continued to progress, AI research has generally become less accessible to those outside of frontier labs. Although a variety of open-source LLMs are publicly available, there are two key issues that have consistently impeded progress in open research: The performance gap between closed and open models. The prevalence of open-weight models, and the scarcity of fully-open models.
  Saved: January 12, 2026
  ai llm open source

• Inside NVIDIA GPUs: Anatomy of high performance matmul kernels

  This post introduces all of the core hardware concepts and programming techniques that underpin state-of-the-art (SOTA) NVIDIA GPU matrix-multiplication (matmul) kernels.
  Saved: January 12, 2026
  ai gpu