<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Subword on Aayush Bajaj's Augmenting Infrastructure</title><link>https://abaj.ai/tags/subword/</link><description>Recent content in Subword on Aayush Bajaj's Augmenting Infrastructure</description><generator>Hugo</generator><language>en</language><copyright>© 2026 Aayush Bajaj</copyright><lastBuildDate>Fri, 10 Jul 2026 08:20:25 +1000</lastBuildDate><atom:link href="https://abaj.ai/tags/subword/index.xml" rel="self" type="application/rss+xml"/><item><title>Tokenisers</title><link>https://abaj.ai/wiki/ml/dl/natural-language-processing/tokenisers/</link><pubDate>Fri, 10 Jul 2026 01:42:07 +1000</pubDate><guid>https://abaj.ai/wiki/ml/dl/natural-language-processing/tokenisers/</guid><description>&lt;p>before a transformer sees a single number, text must be cut into pieces and each piece mapped to an integer. the tokeniser is that cut — the least glamorous and most consequential preprocessing step in the whole pipeline, since it fixes the vocabulary, the sequence length, and what the model &lt;em>can&lt;/em> represent at all.&lt;span class="margin-note" data-note="most &amp;#39;the model can&amp;#39;t spell / can&amp;#39;t do arithmetic&amp;#39; failures trace back to the tokeniser, not the weights">
 &lt;span class="margin-note-indicator">𐃏&lt;/span>
&lt;/span>
&lt;/p></description></item></channel></rss>