Virtual-LPU / index.html
iparaco's picture
Update index.html
a465cc1 verified
Raw History Blame Contribute Delete
3.06 kB
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Virtual LPU V7 | IPARA</title>
<style>
* { box-sizing: border-box; }
body {
margin: 0;
background: #f4f7fa;
color: #182b3a;
font: 17px/1.6 system-ui, sans-serif;
}
main {
max-width: 760px;
margin: 6vh auto;
padding: clamp(24px, 5vw, 56px);
background: white;
border-top: 5px solid #08786d;
}
h1 { margin: 0; font-size: clamp(36px, 7vw, 60px); line-height: 1.1; }
h2 { margin: 32px 0 6px; font-size: 22px; }
p { margin: 8px 0 16px; }
.brand { color: #08675f; font-weight: 700; }
nav { display: flex; flex-wrap: wrap; gap: 12px; margin-top: 28px; }
a {
padding: 10px 16px;
border: 1px solid #08786d;
border-radius: 6px;
color: #08675f;
font-weight: 600;
text-decoration: none;
}
a:first-child, a:hover { background: #08675f; color: white; }
a:focus-visible { outline: 3px solid #e49a15; outline-offset: 3px; }
</style>
</head>
<body>
<main>
<p class="brand">IPARA · Training Control Center</p>
<h1>Virtual LPU V7</h1>
<p>
When a GPU waits for its next batch, a utilization graph shows the
pause. Virtual LPU brings training progress and hardware measurements
into one view so you can investigate what caused it.
</p>
<h2>What works today</h2>
<p>
V7 can detect a running training process and display live measurements
in a local dashboard. A lightweight Python adapter adds measurements
from inside the training loop. This Space shares a development snapshot.
Validation on our dual Tesla P100 system is still in progress.
</p>
<h2>Research at IPARA</h2>
<p>
Our team contributed to the LoViF 2026 challenge report on
real-world all-in-one image restoration. The paper compares how
participating teams approached the same restoration task.
</p>
<p>
<a href="https://arxiv.org/abs/2607.21118"
target="_blank" rel="noopener">Read the LoViF 2026 paper</a>
<h2>Help us test it</h2>
<p>
If you work with PyTorch or GPU performance, try Virtual LPU on one
workload. Change one setting, such as the number of data-loading
workers, and share your GPU model, batch size, and samples per second
before and after. A result showing no speedup is useful too.
</p>
<p>
To build on the project, select <strong>Duplicate this Space</strong>
from the Space menu. If you find it useful, leave a
<strong>like ♥</strong> on Hugging Face.
</p>
<nav aria-label="Project links">
<a href="https://huggingface.co/spaces/iparaco/Virtual-LPU/tree/main"
target="_blank" rel="noopener">Browse source files</a>
<a href="https://huggingface.co/spaces/iparaco/Virtual-LPU/discussions"
target="_blank" rel="noopener">Share a test result</a>
</nav>
</main>
</body>
</html>