<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>ASR on 🌲Treetopia🌲</title>
    <link>https://tree2601.github.io/en/tags/asr/</link>
    <description>Recent content in ASR on 🌲Treetopia🌲</description>
    <generator>Hugo -- 0.154.2</generator>
    <language>en</language>
    <lastBuildDate>Mon, 05 Jan 2026 16:56:40 +0800</lastBuildDate>
    <atom:link href="https://tree2601.github.io/en/tags/asr/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>L20 8-GPU Server Deep Dive: Integrated Deployment Guide for Multimodal AI Systems (LLM &#43; VLM &#43; RAG &#43; ASR &#43; Dify &#43; MinerU)</title>
      <link>https://tree2601.github.io/en/posts/2026/l20/</link>
      <pubDate>Mon, 05 Jan 2026 16:56:40 +0800</pubDate>
      <guid>https://tree2601.github.io/en/posts/2026/l20/</guid>
      <description>&lt;h3 id=&#34;overview&#34;&gt;Overview&lt;/h3&gt;
&lt;p&gt;This guide provides a step-by-step walkthrough for deploying a full-stack multimodal AI system on a single server equipped with 8x NVIDIA L20 GPUs. The stack includes LLM, VLM, Embedding/Reranker (RAG), ASR, Dify (LLM Orchestration Agent Platform), and MinerU (PDF Extraction).&lt;/p&gt;
&lt;h3 id=&#34;vram-estimation-for-llms&#34;&gt;VRAM Estimation for LLMs&lt;/h3&gt;
&lt;p&gt;&lt;img alt=&#34;formula&#34; loading=&#34;lazy&#34; src=&#34;https://tree2601.github.io/images/l20/formula.png&#34;&gt;&lt;/p&gt;
&lt;p&gt;&lt;img alt=&#34;parameter&#34; loading=&#34;lazy&#34; src=&#34;https://tree2601.github.io/images/l20/parameter.png&#34;&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Key Strategy:&lt;/strong&gt; Since Large Language Model (LLM) performance correlates more strongly with parameter scale (B) than with quantization levels, we prioritize models with higher parameter counts. For this deployment, we selected the &lt;strong&gt;int4 AWQ versions&lt;/strong&gt; of &lt;strong&gt;Qwen3-235B&lt;/strong&gt; and &lt;strong&gt;GLM-4.5V-106B&lt;/strong&gt; to maximize overall intelligence and performance within the available VRAM.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
