第三届中国 Rust 开发者大会 使用 Rust 构建 LLM 应用 夏歌 SECTION TITLE SECTION TITLE 我们能不能直接用 Rust • 训练 • 推理 • AI 应用相关的工具 • WASI-NN spec • WasmEdge 已经支持 Pytorch 、 TensorFlow Lite • WASI-NN 2.0 比如 Langchain Rewrite 培养更广泛的 Rust 开发 围绕 LLM 生态封装相应的 Rust 框 架，让开发者能够使用简单的 Rust 写 应用 如何用 Rust 实现的 构建和部署 AI 相关工作流的 serverless 平台 • 上传 Rust function ，平台负责将 Rust 编译成 Wasm ，并运行在 WasmEdge 安全容 器中 • 平台封装了一些常用 LLM 和 SaaS 的 API ，并发布成了

Performance 5 3 License and Copyright 7 4 Installation 9 5 The Basics 15 6 Tutorial 35 7 PyMuPDF, LLM & RAG 45 8 Resources 49 9 Opening Files 51 10 Text 53 11 Images 73 12 Annotations 89 13 Drawing versions up to 1.24.2. 44 Chapter 6. Tutorial CHAPTER SEVEN PYMUPDF, LLM & RAG Integrating into your Large Language Model (LLM) framework and overall RAG (Retrieval-Augmented Generation) solution provides provides the fastest and most reliable way to deliver document data. There are a few well known LLM solutions which have their own interfaces with - it is a fast growing area, so please let us know if

design? •jippity.pro: 350 lines of working code written by a 9 yo in 3h •Entire project passed to an LLM alongside a change request •Will we really need in the future: • Boilerplate generation? AI could something quickly • Tomorrow: entropy that is difficult for humans+AI to reduceHowever • Current LLM technology akin to an extra engineer, not a language feature • 100 lines/s way faster than human,

a Message Handling Library, with an Excursion into the Roots of Programming" 2Frontmatter No AI/LLM was used in the creation of this talk. Code is simplified for slides; may have some errors in translation

i=1 biki ki = f(tn + cih, yn + (ai1k1 + ai2k2 + · · · + ai,i−1ki−1)h) Linear Multistep Methods (LLM) yn+s + as−1 · yn+s−1 + as−2 · yn+s−2 + · · · + a0 · yn = h · � bs · f(tn+s, yn+s) + · · · + b0 ·