<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>Cuda on Suyog Garg</title>
    <link>https://suyoggarg.com/tags/cuda/</link>
    <description>Recent content in Cuda on Suyog Garg</description>
    <image>
      <title>Suyog Garg</title>
      <url>https://suyoggarg.com/</url>
      <link>https://suyoggarg.com/</link>
    </image>
    <generator>Hugo -- 0.128.0</generator>
    <language>en</language>
    <copyright>Suyog Garg</copyright>
    <lastBuildDate>Fri, 19 Jun 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://suyoggarg.com/tags/cuda/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Making PyTorch model training super-fast on cuda</title>
      <link>https://suyoggarg.com/en/posts/2026-06-19-fast-cuda-training/</link>
      <pubDate>Fri, 19 Jun 2026 00:00:00 +0000</pubDate>
      <guid>https://suyoggarg.com/en/posts/2026-06-19-fast-cuda-training/</guid>
      <description>DISCLAIMER: Made with Google Gemini-Pro prompts.
I restructured all my Python installation in the system and have now committed to use only Anaconda conda environments to manage my Pythons. Below are some steps I followed to clean my directories!
A deep dive into fixing OpenMP runtime conflicts, eliminating TorchDynamo graph breaks, and breaking the silent HDF5 global process lock for Gravitational-Wave VAE models.
The Starting Point: An Absolute Stall When training heavy gravitational-wave parameter estimation and population models (utilizing libraries like bilby and gwpopulation), we hit a wall.</description>
    </item>
  </channel>
</rss>
