https://chejinying.com/tech/tags/llm/ Recent content in Llm on Jinying Tech Blog Hugo en-us Sat, 04 Apr 2026 00:08:50 +0800 https://chejinying.com/tech/posts/llm/overview/ Sat, 04 Apr 2026 00:08:50 +0800 https://chejinying.com/tech/posts/llm/overview/ <blockquote> <p>This post uses many abbreviations (BPE, FFN, RoPE, etc.). See the <a href="https://chejinying.com/tech/posts/llm/abbreviations/">LLM Abbreviations Glossary</a> for a quick reference.</p> </blockquote> <h1 id="what-is-llm">What is LLM?</h1> <p>At its core, LLM is a <strong>next-token predictor</strong>.</p> <p>Given a sequence of tokens, it predicts the most probable next token. By repeating this (autoregressive generation), it produces coherent text.</p> <pre tabindex="0"><code>Input: &#34;The cat sat on the&#34; Model: P(&#34;mat&#34;)=0.23, P(&#34;floor&#34;)=0.18, P(&#34;roof&#34;)=0.07, ... Pick: &#34;mat&#34; </code></pre><p>Here&rsquo;s the high-level pipeline — every concept in this post maps to one of these stages:</p>