xai / study.html
lyffseba's picture
xai: research portal v2 (sources, polish, sync)
5f56d31 verified
Raw
History Blame
2.03 kB
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Study path · xai</title>
<link rel="stylesheet" href="style.css">
</head>
<body>
<div class="wrap">
<h1>Study path</h1>
<p class="meta">~1–2 days · offline · no invented numbers</p>
<p class="nav">
<a href="index.html">home</a>
<a href="inkling.html">Inkling</a>
<a href="kimi-k3.html">Kimi K3</a>
<a href="compare.html">compare</a>
<a href="study.html" class="active">study path</a>
<a href="sources.html">sources</a>
<a href="msc.html">MSc</a>
</p>
<hr>
<ol>
<li>
<strong>Map (20 min)</strong>
this portal ·
<a href="sources.html">sources</a> ·
<a href="https://huggingface.co/datasets/lyffseba/xai-studies/blob/main/studies/derived/DASHBOARD.md">dashboard</a>
</li>
<li>
<strong>Inkling deep (half day)</strong>
map → SPEC → architecture → modular walkthrough with vendored <code>modular_inkling.py</code>
</li>
<li>
<strong>Kimi K3 deep (half day)</strong>
map → SPEC + lineage → KDA notes → API protocol → limitations / open questions
</li>
<li>
<strong>Compare again (30 min)</strong>
<a href="compare.html">comparison</a>; answer: which long-context bet? which MoE story for production? what would you fine-tune first?
</li>
<li>
<strong>MSc (optional)</strong>
<a href="msc.html">dissertation</a> · notebook needs local data + TF/Keras to run
</li>
</ol>
<h2>Offline verify</h2>
<pre>python3 studies/run.py test
# expect: validate OK inkling · validate OK kimi-k3 · package check OK</pre>
<h2>One-liner facts</h2>
<ul>
<li><strong>Inkling:</strong> ~975B / 41B · 6/256+2 · relative PE · SWA/global · image+audio · 1M</li>
<li><strong>Kimi K3:</strong> 2.8T · 16/896 · KDA+AttnRes · vision/video · 1M · active undisclosed · weights ETA 2026-07-27</li>
</ul>
<div class="foot"><a href="index.html">← home</a></div>
</div>
</body>
</html>