deploy: 84b3c2016e

index.xml

@@ -1,5 +1,14 @@
-<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Eric X. Liu's Personal Page</title><link>/</link><description>Recent content on Eric X. Liu's Personal Page</description><generator>Hugo</generator><language>en</language><lastBuildDate>Sun, 03 Aug 2025 03:10:45 +0000</lastBuildDate><atom:link href="/index.xml" rel="self" type="application/rss+xml"/><item><title>A Deep Dive into PPO for Language Models</title><link>/posts/a-deep-dive-into-ppo-for-language-models/</link><pubDate>Sun, 03 Aug 2025 03:10:41 +0000</pubDate><guid>/posts/a-deep-dive-into-ppo-for-language-models/</guid><description>&lt;p>Large Language Models (LLMs) have demonstrated astonishing capabilities, but out-of-the-box, they are simply powerful text predictors. They don&amp;rsquo;t inherently understand what makes a response helpful, harmless, or aligned with human values. The technique that has proven most effective at bridging this gap is Reinforcement Learning from Human Feedback (RLHF), and at its heart lies a powerful algorithm: Proximal Policy Optimization (PPO).&lt;/p>
-&lt;p>You may have seen diagrams like the one below, which outlines the RLHF training process. It can look intimidating, with a web of interconnected models, losses, and data flows.&lt;/p></description></item><item><title>T5 - The Transformer That Zigged When Others Zagged - An Architectural Deep Dive</title><link>/posts/t5-the-transformer-that-zigged-when-others-zagged-an-architectural-deep-dive/</link><pubDate>Sun, 03 Aug 2025 03:10:41 +0000</pubDate><guid>/posts/t5-the-transformer-that-zigged-when-others-zagged-an-architectural-deep-dive/</guid><description>&lt;p>In the rapidly evolving landscape of Large Language Models, a few key architectures define the dominant paradigms. Today, the &amp;ldquo;decoder-only&amp;rdquo; model, popularized by the GPT series and its successors like LLaMA and Mistral, reigns supreme. These models are scaled to incredible sizes and excel at in-context learning.&lt;/p>
+<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Eric X. Liu's Personal Page</title><link>/</link><description>Recent content on Eric X. Liu's Personal Page</description><generator>Hugo</generator><language>en</language><lastBuildDate>Sun, 03 Aug 2025 03:15:05 +0000</lastBuildDate><atom:link href="/index.xml" rel="self" type="application/rss+xml"/><item><title>A Deep Dive into PPO for Language Models</title><link>/posts/a-deep-dive-into-ppo-for-language-models/</link><pubDate>Sun, 03 Aug 2025 03:14:20 +0000</pubDate><guid>/posts/a-deep-dive-into-ppo-for-language-models/</guid><description>&lt;p>Large Language Models (LLMs) have demonstrated astonishing capabilities, but out-of-the-box, they are simply powerful text predictors. They don&amp;rsquo;t inherently understand what makes a response helpful, harmless, or aligned with human values. The technique that has proven most effective at bridging this gap is Reinforcement Learning from Human Feedback (RLHF), and at its heart lies a powerful algorithm: Proximal Policy Optimization (PPO).&lt;/p>
+&lt;p>You may have seen diagrams like the one below, which outlines the RLHF training process. It can look intimidating, with a web of interconnected models, losses, and data flows.&lt;/p></description></item><item><title>Mixture-of-Experts (MoE) Models Challenges &amp; Solutions in Practice</title><link>/posts/mixture-of-experts-moe-models-challenges-solutions-in-practice/</link><pubDate>Sun, 03 Aug 2025 03:14:20 +0000</pubDate><guid>/posts/mixture-of-experts-moe-models-challenges-solutions-in-practice/</guid><description>&lt;p>Mixture-of-Experts (MoEs) are neural network architectures that allow different parts of the model (called &amp;ldquo;experts&amp;rdquo;) to specialize in different types of inputs. A &amp;ldquo;gating network&amp;rdquo; or &amp;ldquo;router&amp;rdquo; learns to dispatch each input (or &amp;ldquo;token&amp;rdquo;) to a subset of these experts. While powerful for scaling models, MoEs introduce several practical challenges.&lt;/p>
+&lt;h3 id="1-challenge-non-differentiability-of-routing-functions">
+ 1. Challenge: Non-Differentiability of Routing Functions
+ &lt;a class="heading-link" href="#1-challenge-non-differentiability-of-routing-functions">
+ &lt;i class="fa-solid fa-link" aria-hidden="true" title="Link to heading">&lt;/i>
+ &lt;span class="sr-only">Link to heading&lt;/span>
+ &lt;/a>
+&lt;/h3>
+&lt;p>&lt;strong>The Problem:&lt;/strong>
+Many routing mechanisms, especially &amp;ldquo;Top-K routing,&amp;rdquo; involve a discrete, hard selection process. A common function is &lt;code>KeepTopK(v, k)&lt;/code>, which selects the top &lt;code>k&lt;/code> scoring elements from a vector &lt;code>v&lt;/code> and sets others to $-\infty$ or $0$.&lt;/p></description></item><item><title>T5 - The Transformer That Zigged When Others Zagged - An Architectural Deep Dive</title><link>/posts/t5-the-transformer-that-zigged-when-others-zagged-an-architectural-deep-dive/</link><pubDate>Sun, 03 Aug 2025 03:14:20 +0000</pubDate><guid>/posts/t5-the-transformer-that-zigged-when-others-zagged-an-architectural-deep-dive/</guid><description>&lt;p>In the rapidly evolving landscape of Large Language Models, a few key architectures define the dominant paradigms. Today, the &amp;ldquo;decoder-only&amp;rdquo; model, popularized by the GPT series and its successors like LLaMA and Mistral, reigns supreme. These models are scaled to incredible sizes and excel at in-context learning.&lt;/p>
 &lt;p>But to truly understand the field, we must look at the pivotal models that explored different paths. Google&amp;rsquo;s T5, or &lt;strong>Text-to-Text Transfer Transformer&lt;/strong>, stands out as one of the most influential. It didn&amp;rsquo;t just introduce a new model; it proposed a new philosophy. This article dives deep into the architecture of T5, how it fundamentally differs from modern LLMs, and the lasting legacy of its unique design choices.&lt;/p></description></item><item><title>Some useful files</title><link>/posts/useful/</link><pubDate>Mon, 26 Oct 2020 04:14:43 +0000</pubDate><guid>/posts/useful/</guid><description>&lt;ul>
 &lt;li>&lt;a href="https://ericxliu.me/rootCA.pem" class="external-link" target="_blank" rel="noopener">rootCA.pem&lt;/a>&lt;/li>
 &lt;li>&lt;a href="https://ericxliu.me/vpnclient.ovpn" class="external-link" target="_blank" rel="noopener">vpnclient.ovpn&lt;/a>&lt;/li>