kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-01-14 06:27:45 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity
79164be9b4
TensorRT-LLMs/advanced/lora.html
Kaiyu Xie 79164be9b4
Revert "Update gh-pages (#3393)" (#3394) 
This reverts commit 5e96c9b208.
2025-04-09 11:13:20 +08:00

758 lines
48 KiB
HTML
Raw Blame History

<!DOCTYPE html>
<html lang="en" data-content_root="../" >
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" /><meta name="viewport" content="width=device-width, initial-scale=1" />
<title>Run gpt-2b + LoRA using GptManager / cpp runtime &#8212; tensorrt_llm</title>
<script data-cfasync="false">
document.documentElement.dataset.mode = localStorage.getItem("mode") || "";
document.documentElement.dataset.theme = localStorage.getItem("theme") || "";
</script>
<!--
this give us a css class that will be invisible only if js is disabled
-->
<noscript>
<style>
.pst-js-only { display: none !important; }
</style>
</noscript>
<!-- Loaded before other Sphinx assets -->
<link href="../_static/styles/theme.css?digest=8878045cc6db502f8baf" rel="stylesheet" />
<link href="../_static/styles/pydata-sphinx-theme.css?digest=8878045cc6db502f8baf" rel="stylesheet" />
<link rel="stylesheet" type="text/css" href="../_static/pygments.css?v=a746c00c" />
<link rel="stylesheet" type="text/css" href="../_static/styles/nvidia-sphinx-theme.css?v=df3ac72c" />
<link rel="stylesheet" type="text/css" href="../_static/copybutton.css?v=76b2166b" />
<!-- So that users can add custom icons -->
<script src="../_static/scripts/fontawesome.js?digest=8878045cc6db502f8baf"></script>
<!-- Pre-loaded scripts that we'll load fully later -->
<link rel="preload" as="script" href="../_static/scripts/bootstrap.js?digest=8878045cc6db502f8baf" />
<link rel="preload" as="script" href="../_static/scripts/pydata-sphinx-theme.js?digest=8878045cc6db502f8baf" />
<script src="../_static/documentation_options.js?v=5929fcd5"></script>
<script src="../_static/doctools.js?v=888ff710"></script>
<script src="../_static/sphinx_highlight.js?v=dc90522c"></script>
<script src="../_static/clipboard.min.js?v=a7894cd8"></script>
<script src="../_static/copybutton.js?v=65e89d2a"></script>
<script>DOCUMENTATION_OPTIONS.pagename = 'advanced/lora';</script>
<link rel="icon" href="../_static/favicon.png"/>
<link rel="index" title="Index" href="../genindex.html" />
<link rel="search" title="Search" href="../search.html" />
<link rel="next" title="Expert Parallelism in TensorRT-LLM" href="expert-parallelism.html" />
<link rel="prev" title="Inference Request" href="inference-request.html" />
<meta name="viewport" content="width=device-width, initial-scale=1"/>
<meta name="docsearch:language" content="en"/>
<meta name="docsearch:version" content="" />
</head>
<body data-bs-spy="scroll" data-bs-target=".bd-toc-nav" data-offset="180" data-bs-root-margin="0px 0px -60%" data-default-mode="">
<div id="pst-skip-link" class="skip-link d-print-none"><a href="#main-content">Skip to main content</a></div>
<div id="pst-scroll-pixel-helper"></div>
<button type="button" class="btn rounded-pill" id="pst-back-to-top">
<i class="fa-solid fa-arrow-up"></i>Back to top</button>
<dialog id="pst-search-dialog">
<form class="bd-search d-flex align-items-center"
action="../search.html"
method="get">
<i class="fa-solid fa-magnifying-glass"></i>
<input type="search"
class="form-control"
name="q"
placeholder="Search the docs ..."
aria-label="Search the docs ..."
autocomplete="off"
autocorrect="off"
autocapitalize="off"
spellcheck="false"/>
<span class="search-button__kbd-shortcut"><kbd class="kbd-shortcut__modifier">Ctrl</kbd>+<kbd>K</kbd></span>
</form>
</dialog>
<div class="pst-async-banner-revealer d-none">
<aside id="bd-header-version-warning" class="d-none d-print-none" aria-label="Version warning"></aside>
</div>
<header class="bd-header navbar navbar-expand-lg bd-navbar d-print-none">
<div class="bd-header__inner bd-page-width">
<button class="pst-navbar-icon sidebar-toggle primary-toggle" aria-label="Site navigation">
<span class="fa-solid fa-bars"></span>
</button>
<div class="col-lg-3 navbar-header-items__start">
<div class="navbar-item">
<a class="navbar-brand logo" href="../index.html">
<img src="../_static/nvidia-logo-horiz-rgb-blk-for-screen.svg" class="logo__image only-light" alt="tensorrt_llm - Home"/>
<img src="../_static/nvidia-logo-horiz-rgb-wht-for-screen.svg" class="logo__image only-dark pst-js-only" alt="tensorrt_llm - Home"/>
<p class="title logo__title">tensorrt_llm</p>
</a></div>
</div>
<div class="col-lg-9 navbar-header-items">
<div class="navbar-header-items__end">
<div class="navbar-item navbar-persistent--container">
<button class="btn search-button-field search-button__button pst-js-only" title="Search" aria-label="Search" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="fa-solid fa-magnifying-glass"></i>
<span class="search-button__default-text">Search</span>
<span class="search-button__kbd-shortcut"><kbd class="kbd-shortcut__modifier">Ctrl</kbd>+<kbd class="kbd-shortcut__modifier">K</kbd></span>
</button>
</div>
<div class="navbar-item">
<button class="btn btn-sm nav-link pst-navbar-icon theme-switch-button pst-js-only" aria-label="Color mode" data-bs-title="Color mode" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="theme-switch fa-solid fa-sun fa-lg" data-mode="light" title="Light"></i>
<i class="theme-switch fa-solid fa-moon fa-lg" data-mode="dark" title="Dark"></i>
<i class="theme-switch fa-solid fa-circle-half-stroke fa-lg" data-mode="auto" title="System Settings"></i>
</button></div>
</div>
</div>
<div class="navbar-persistent--mobile">
<button class="btn search-button-field search-button__button pst-js-only" title="Search" aria-label="Search" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="fa-solid fa-magnifying-glass"></i>
<span class="search-button__default-text">Search</span>
<span class="search-button__kbd-shortcut"><kbd class="kbd-shortcut__modifier">Ctrl</kbd>+<kbd class="kbd-shortcut__modifier">K</kbd></span>
</button>
</div>
<button class="pst-navbar-icon sidebar-toggle secondary-toggle" aria-label="On this page">
<span class="fa-solid fa-outdent"></span>
</button>
</div>
</header>
<div class="bd-container">
<div class="bd-container__inner bd-page-width">
<dialog id="pst-primary-sidebar-modal"></dialog>
<div id="pst-primary-sidebar" class="bd-sidebar-primary bd-sidebar">
<a class="navbar-brand logo" href="../index.html">
<img src="../_static/nvidia-logo-horiz-rgb-blk-for-screen.svg" class="logo__image only-light" alt="tensorrt_llm - Home"/>
<img src="../_static/nvidia-logo-horiz-rgb-wht-for-screen.svg" class="logo__image only-dark pst-js-only" alt="tensorrt_llm - Home"/>
<p class="title logo__title">tensorrt_llm</p>
</a>
<div class="sidebar-header-items sidebar-primary__section">
<div class="sidebar-header-items__end">
<div class="navbar-item">
<button class="btn btn-sm nav-link pst-navbar-icon theme-switch-button pst-js-only" aria-label="Color mode" data-bs-title="Color mode" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="theme-switch fa-solid fa-sun fa-lg" data-mode="light" title="Light"></i>
<i class="theme-switch fa-solid fa-moon fa-lg" data-mode="dark" title="Dark"></i>
<i class="theme-switch fa-solid fa-circle-half-stroke fa-lg" data-mode="auto" title="System Settings"></i>
</button></div>
</div>
</div>
<div class="sidebar-primary-items__start sidebar-primary__section">
<div class="sidebar-primary-item">
<nav class="bd-docs-nav bd-links"
aria-label="Table of Contents">
<p class="bd-links__title" role="heading" aria-level="1">Table of Contents</p>
<div class="bd-toc-item navbar-nav"><p aria-level="2" class="caption" role="heading"><span class="caption-text">Getting Started</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../overview.html">Overview</a></li>
<li class="toctree-l1"><a class="reference internal" href="../quick-start-guide.html">Quick Start Guide</a></li>
<li class="toctree-l1"><a class="reference internal" href="../key-features.html">Key Features</a></li>
<li class="toctree-l1"><a class="reference internal" href="../torch.html">PyTorch Backend</a></li>
<li class="toctree-l1"><a class="reference internal" href="../release-notes.html">Release Notes</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Installation</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../installation/linux.html">Installing on Linux</a></li>
<li class="toctree-l1"><a class="reference internal" href="../installation/build-from-source-linux.html">Building from Source Code on Linux</a></li>
<li class="toctree-l1"><a class="reference internal" href="../installation/grace-hopper.html">Installing on Grace Hopper</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">LLM API</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../llm-api/index.html">API Introduction</a></li>
<li class="toctree-l1"><a class="reference internal" href="../llm-api/reference.html">API Reference</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">LLM API Examples</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1 has-children"><a class="reference internal" href="../llm-api-examples/index.html">LLM Examples Introduction</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_multilora.html">Generate text with multiple LoRA adapters</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_quantization.html">Generation with Quantization</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_inference_async.html">Generate Text Asynchronously</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_guided_decoding.html">Generate text with guided decoding</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_inference_customize.html">Generate text with customization</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_logits_processor.html">Control generated text using logits post processor</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_inference_async_streaming.html">Generate Text in Streaming</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_inference_distributed.html">Distributed LLM Generation</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_medusa_decoding.html">Generate Text Using Medusa Decoding</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_lookahead_decoding.html">Generate Text Using Lookahead Decoding</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_inference.html">Generate text</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_auto_parallel.html">Automatic Parallelism with LLM</a></li>
</ul>
</details></li>
<li class="toctree-l1"><a class="reference internal" href="../llm-api-examples/customization.html">Common Customizations</a></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="../llm-api-examples/llm_api_examples.html">Examples</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_multilora.html">Generate text with multiple LoRA adapters</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_quantization.html">Generation with Quantization</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_inference_async.html">Generate Text Asynchronously</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_guided_decoding.html">Generate text with guided decoding</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_inference_customize.html">Generate text with customization</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_logits_processor.html">Control generated text using logits post processor</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_inference_async_streaming.html">Generate Text in Streaming</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_inference_distributed.html">Distributed LLM Generation</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_medusa_decoding.html">Generate Text Using Medusa Decoding</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_lookahead_decoding.html">Generate Text Using Lookahead Decoding</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_inference.html">Generate text</a></li>
<li class="toctree-l2"><a class="reference internal" href="../llm-api-examples/llm_auto_parallel.html">Automatic Parallelism with LLM</a></li>
</ul>
</details></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Model Definition API</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../python-api/tensorrt_llm.layers.html">Layers</a></li>
<li class="toctree-l1"><a class="reference internal" href="../python-api/tensorrt_llm.functional.html">Functionals</a></li>
<li class="toctree-l1"><a class="reference internal" href="../python-api/tensorrt_llm.models.html">Models</a></li>
<li class="toctree-l1"><a class="reference internal" href="../python-api/tensorrt_llm.plugin.html">Plugin</a></li>
<li class="toctree-l1"><a class="reference internal" href="../python-api/tensorrt_llm.quantization.html">Quantization</a></li>
<li class="toctree-l1"><a class="reference internal" href="../python-api/tensorrt_llm.runtime.html">Runtime</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">C++ API</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../_cpp_gen/executor.html">Executor</a></li>
<li class="toctree-l1"><a class="reference internal" href="../_cpp_gen/runtime.html">Runtime</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Command-Line Reference</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../commands/trtllm-build.html">trtllm-build</a></li>
<li class="toctree-l1"><a class="reference internal" href="../commands/trtllm-serve.html">trtllm-serve</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Architecture</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../architecture/overview.html">TensorRT-LLM Architecture</a></li>
<li class="toctree-l1"><a class="reference internal" href="../architecture/core-concepts.html">Model Definition</a></li>
<li class="toctree-l1"><a class="reference internal" href="../architecture/checkpoint.html">TensorRT-LLM Checkpoint</a></li>
<li class="toctree-l1"><a class="reference internal" href="../architecture/workflow.html">TensorRT-LLM Build Workflow</a></li>
<li class="toctree-l1"><a class="reference internal" href="../architecture/add-model.html">Adding a Model</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Advanced</span></p>
<ul class="current nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="gpt-attention.html">Multi-Head, Multi-Query, and Group-Query Attention</a></li>
<li class="toctree-l1"><a class="reference internal" href="gpt-runtime.html">C++ GPT Runtime</a></li>
<li class="toctree-l1"><a class="reference internal" href="executor.html">Executor API</a></li>
<li class="toctree-l1"><a class="reference internal" href="graph-rewriting.html">Graph Rewriting Module</a></li>
<li class="toctree-l1"><a class="reference internal" href="inference-request.html">Inference Request</a></li>
<li class="toctree-l1 current active"><a class="current reference internal" href="#">Run gpt-2b + LoRA using GptManager / cpp runtime</a></li>
<li class="toctree-l1"><a class="reference internal" href="expert-parallelism.html">Expert Parallelism in TensorRT-LLM</a></li>
<li class="toctree-l1"><a class="reference internal" href="kv-cache-reuse.html">KV cache reuse</a></li>
<li class="toctree-l1"><a class="reference internal" href="speculative-decoding.html">Speculative Sampling</a></li>
<li class="toctree-l1"><a class="reference internal" href="disaggregated-service.html">Disaggregated-Service (experimental)</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Performance</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../performance/perf-overview.html">Overview</a></li>
<li class="toctree-l1"><a class="reference internal" href="../performance/perf-benchmarking.html">Benchmarking</a></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="../performance/performance-tuning-guide/index.html">Performance Tuning Guide</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="../performance/performance-tuning-guide/benchmarking-default-performance.html">Benchmarking Default Performance</a></li>
<li class="toctree-l2"><a class="reference internal" href="../performance/performance-tuning-guide/useful-build-time-flags.html">Useful Build-Time Flags</a></li>
<li class="toctree-l2"><a class="reference internal" href="../performance/performance-tuning-guide/tuning-max-batch-size-and-max-num-tokens.html">Tuning Max Batch Size and Max Num Tokens</a></li>
<li class="toctree-l2"><a class="reference internal" href="../performance/performance-tuning-guide/deciding-model-sharding-strategy.html">Deciding Model Sharding Strategy</a></li>
<li class="toctree-l2"><a class="reference internal" href="../performance/performance-tuning-guide/fp8-quantization.html">FP8 Quantization</a></li>
<li class="toctree-l2"><a class="reference internal" href="../performance/performance-tuning-guide/useful-runtime-flags.html">Useful Runtime Options</a></li>
</ul>
</details></li>
<li class="toctree-l1"><a class="reference internal" href="../performance/perf-analysis.html">Performance Analysis</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Reference</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../reference/troubleshooting.html">Troubleshooting</a></li>
<li class="toctree-l1"><a class="reference internal" href="../reference/support-matrix.html">Support Matrix</a></li>
<li class="toctree-l1"><a class="reference internal" href="../reference/precision.html">Numerical Precision</a></li>
<li class="toctree-l1"><a class="reference internal" href="../reference/memory.html">Memory Usage of TensorRT-LLM</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Blogs</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../blogs/H100vsA100.html">H100 has 4.6x A100 Performance in TensorRT-LLM, achieving 10,000 tok/s at 100ms to first token</a></li>
<li class="toctree-l1"><a class="reference internal" href="../blogs/H200launch.html">H200 achieves nearly 12,000 tokens/sec on Llama2-13B with TensorRT-LLM</a></li>
<li class="toctree-l1"><a class="reference internal" href="../blogs/Falcon180B-H200.html">Falcon-180B on a single H200 GPU with INT4 AWQ, and 6.7x faster Llama-70B over A100</a></li>
<li class="toctree-l1"><a class="reference internal" href="../blogs/quantization-in-TRT-LLM.html">Speed up inference with SOTA quantization techniques in TRT-LLM</a></li>
<li class="toctree-l1"><a class="reference internal" href="../blogs/XQA-kernel.html">New XQA-kernel provides 2.4x more Llama-70B throughput within the same latency budget</a></li>
</ul>
</div>
</nav></div>
</div>
<div class="sidebar-primary-items__end sidebar-primary__section">
</div>
</div>
<main id="main-content" class="bd-main" role="main">
<div class="bd-content">
<div class="bd-article-container">
<div class="bd-header-article d-print-none">
<div class="header-article-items header-article__inner">
<div class="header-article-items__start">
<div class="header-article-item">
<nav aria-label="Breadcrumb" class="d-print-none">
<ul class="bd-breadcrumbs">
<li class="breadcrumb-item breadcrumb-home">
<a href="../index.html" class="nav-link" aria-label="Home">
<i class="fa-solid fa-home"></i>
</a>
</li>
<li class="breadcrumb-item active" aria-current="page"><span class="ellipsis">Run gpt-2b + LoRA using GptManager / cpp runtime</span></li>
</ul>
</nav>
</div>
</div>
</div>
</div>
<div id="searchbox"></div>
<article class="bd-article">
<section id="run-gpt-2b-lora-using-gptmanager-cpp-runtime">
<span id="lora"></span><h1>Run gpt-2b + LoRA using GptManager / cpp runtime<a class="headerlink" href="#run-gpt-2b-lora-using-gptmanager-cpp-runtime" title="Link to this heading">#</a></h1>
<p>First build a model with LoRA and inflight-batching enabled.</p>
<div class="highlight-bash notranslate"><div class="highlight"><pre><span></span>git-lfs<span class="w"> </span>clone<span class="w"> </span>https://huggingface.co/qychen/luotuo-lora-7b-0.1
git-lfs<span class="w"> </span>clone<span class="w"> </span>https://huggingface.co/kunishou/Japanese-Alpaca-LoRA-7b-v0
<span class="nv">BASE_MODEL</span><span class="o">=</span>llama-7b-hf
python<span class="w"> </span>examples/llama/convert_checkpoint.py<span class="w"> </span>--model_dir<span class="w"> </span><span class="si">${</span><span class="nv">BASE_MODEL</span><span class="si">}</span><span class="w"> </span><span class="se">\</span>
<span class="w"> </span>--output_dir<span class="w"> </span>/tmp/llama_7b/trt_ckpt/fp16/1-gpu/<span class="w"> </span><span class="se">\</span>
<span class="w"> </span>--dtype<span class="w"> </span>float16
trtllm-build<span class="w"> </span>--checkpoint_dir<span class="w"> </span>/tmp/llama_7b/trt_ckpt/fp16/1-gpu/<span class="w"> </span><span class="se">\</span>
<span class="w"> </span>--output_dir<span class="w"> </span>/tmp/llama_7b_with_lora_qkv/trt_engines/fp16/1-gpu/<span class="w"> </span><span class="se">\</span>
<span class="w"> </span>--remove_input_padding<span class="w"> </span><span class="nb">enable</span><span class="w"> </span><span class="se">\</span>
<span class="w"> </span>--gpt_attention_plugin<span class="w"> </span>float16<span class="w"> </span><span class="se">\</span>
<span class="w"> </span>--context_fmha<span class="w"> </span><span class="nb">enable</span><span class="w"> </span><span class="se">\</span>
<span class="w"> </span>--paged_kv_cache<span class="w"> </span><span class="nb">enable</span><span class="w"> </span><span class="se">\</span>
<span class="w"> </span>--gemm_plugin<span class="w"> </span>float16<span class="w"> </span><span class="se">\</span>
<span class="w"> </span>--lora_plugin<span class="w"> </span>float16<span class="w"> </span><span class="se">\</span>
<span class="w"> </span>--max_batch_size<span class="w"> </span><span class="m">128</span><span class="w"> </span><span class="se">\</span>
<span class="w"> </span>--max_input_len<span class="w"> </span><span class="m">512</span><span class="w"> </span><span class="se">\</span>
<span class="w"> </span>--max_seq_len<span class="w"> </span><span class="m">562</span><span class="w"> </span><span class="se">\</span>
<span class="w"> </span>--lora_dir<span class="w"> </span>Japanese-Alpaca-LoRA-7b-v0<span class="w"> </span><span class="se">\</span>
<span class="w"> </span>--max_lora_rank<span class="w"> </span><span class="m">8</span><span class="w"> </span><span class="se">\</span>
<span class="w"> </span>--lora_target_modules<span class="w"> </span><span class="s2">&quot;attn_q&quot;</span><span class="w"> </span><span class="s2">&quot;attn_k&quot;</span><span class="w"> </span><span class="s2">&quot;attn_v&quot;</span>
</pre></div>
</div>
<p>To pass LoRAs into the cpp runtime they must be converted to the format below.
The script below will convert a Hugging Face LoRA model to the correct NumPy tensor.</p>
<div class="highlight-bash notranslate"><div class="highlight"><pre><span></span>python3<span class="w"> </span>tensorrt_llm/examples/hf_lora_convert.py<span class="w"> </span>-i<span class="w"> </span>Japanese-Alpaca-LoRA-7b-v0<span class="w"> </span>-o<span class="w"> </span>Japanese-Alpaca-LoRA-7b-v0-weights<span class="w"> </span>--storage-type<span class="w"> </span>float16
python3<span class="w"> </span>tensorrt_llm/examples/hf_lora_convert.py<span class="w"> </span>-i<span class="w"> </span>luotuo-lora-7b-0.1<span class="w"> </span>-o<span class="w"> </span>luotuo-lora-7b-0.1-weights<span class="w"> </span>--storage-type<span class="w"> </span>float16
</pre></div>
</div>
<p>Refer to the <a class="reference external" href="https://github.com/triton-inference-server/tensorrtllm_backend/blob/main/docs/lora.md">tensorrtllm_backend documentation</a> for a Multi-LoRA example using Triton.</p>
<section id="lora-tensor-format-details">
<h2>LoRA tensor format details<a class="headerlink" href="#lora-tensor-format-details" title="Link to this heading">#</a></h2>
<p>To run inference with <code class="docutils literal notranslate"><span class="pre">LoraWeights</span></code> using <code class="docutils literal notranslate"><span class="pre">GptManager</span></code>, <code class="docutils literal notranslate"><span class="pre">InferenceRequests</span></code> must have <code class="docutils literal notranslate"><span class="pre">LoraWeights</span></code> (<code class="docutils literal notranslate"><span class="pre">lora_weights</span></code>) and <code class="docutils literal notranslate"><span class="pre">LoraConfig</span></code> (<code class="docutils literal notranslate"><span class="pre">lora_config</span></code>) parameters.</p>
<p><code class="docutils literal notranslate"><span class="pre">LoraTaskId</span></code> the unique task ID for the given LoRA.</p>
<p>To perform inference with a specific LoRA for the first time, <code class="docutils literal notranslate"><span class="pre">lora_task_id</span></code>, <code class="docutils literal notranslate"><span class="pre">lora_weights</span></code>, and <code class="docutils literal notranslate"><span class="pre">lora_config</span></code> must all be given. The LoRA will be cached, so that subsequent requests for the same task only require <code class="docutils literal notranslate"><span class="pre">lora_task_id</span></code>.
If the cache is full, the oldest LoRA will be evicted to make space for new ones. An error is returned if <code class="docutils literal notranslate"><span class="pre">lora_task_id</span></code> is not cached.</p>
<p><code class="docutils literal notranslate"><span class="pre">LoraWeights</span></code> contains the weights for all the LoRAs. Currently, this should include weights for all TP and PP ranks.
The weights tensor has the shape <code class="docutils literal notranslate"><span class="pre">[num_lora_modules_layers,</span> <span class="pre">D</span> <span class="pre">x</span> <span class="pre">Hi</span> <span class="pre">+</span> <span class="pre">Ho</span> <span class="pre">x</span> <span class="pre">D</span> <span class="pre">]</span></code>. The last dimension holds the in / out adapter weights for the associated module (for example, <code class="docutils literal notranslate"><span class="pre">attn_qkv</span></code>) and model layer.</p>
<p>Each of the in / out tensors are first flattened and then concatenated together in the format above.
The first dimension (of size <code class="docutils literal notranslate"><span class="pre">num_lora_module_layers</span></code>) has an entry for each module-layer (that is, there is an entry for <code class="docutils literal notranslate"><span class="pre">attn_q</span> <span class="pre">layer1</span></code> and another for <code class="docutils literal notranslate"><span class="pre">attn_k</span> <span class="pre">layer1</span></code>).</p>
<p><code class="docutils literal notranslate"><span class="pre">D=adapter_size</span> <span class="pre">(i.e.</span> <span class="pre">R</span> <span class="pre">value),</span> <span class="pre">Hi=hidden_size_in,</span> <span class="pre">Ho=hidden_size_out.</span></code></p>
<p><code class="docutils literal notranslate"><span class="pre">LoraConfig</span></code> is a configuration tensor which identifies the moduleId, layerId, and adapter size of each element of <code class="docutils literal notranslate"><span class="pre">LoraWeights</span></code>. It has the shape <code class="docutils literal notranslate"><span class="pre">[num_lora_modules_layers,</span> <span class="pre">3]</span></code>. The last dimension holds <code class="docutils literal notranslate"><span class="pre">[module_id,</span> <span class="pre">layer_idx,</span> <span class="pre">adapter_size</span> <span class="pre">D</span> <span class="pre">(i.e.</span> <span class="pre">R</span> <span class="pre">value)]</span></code>.</p>
<p>This feature supports LoRAs as described in https://arxiv.org/pdf/2106.09685.pdf</p>
<section id="example-lora-tensors">
<h3>Example LoRA tensors<a class="headerlink" href="#example-lora-tensors" title="Link to this heading">#</a></h3>
<p>Here is an example of <code class="docutils literal notranslate"><span class="pre">LoraWeights</span></code> and <code class="docutils literal notranslate"><span class="pre">LoraConfig</span></code> tensors for a model with tp=1, pp=1, 4 layers, and a hidden size of 4.
The following tensors are for a LoRA which has a <code class="docutils literal notranslate"><span class="pre">q</span></code> and <code class="docutils literal notranslate"><span class="pre">k</span></code> adapter.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1"># loraConfig</span>
<span class="p">[</span>
<span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">0</span><span class="p">,</span> <span class="mi">2</span><span class="p">]</span>
<span class="p">[</span><span class="mi">2</span><span class="p">,</span> <span class="mi">0</span><span class="p">,</span> <span class="mi">4</span><span class="p">]</span>
<span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">]</span>
<span class="p">[</span><span class="mi">2</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="mi">4</span><span class="p">]</span>
<span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">]</span> <span class="c1"># Note that the final 2 layers only adapt `q`</span>
<span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">8</span><span class="p">]</span>
<span class="p">]</span>
<span class="c1"># Note: The loraConfig tensor configures the loraWeights tensor.</span>
<span class="c1"># The contents of each row of loraWeights is specified be the corresponding row in loraConfig</span>
<span class="c1"># loraWeights</span>
<span class="c1"># Note: that &#39;in weights&#39; and &#39;out weights&#39; are &#39;A&#39; and &#39;B&#39; in the LoRA paper.</span>
<span class="p">[</span>
<span class="p">[</span> <span class="o">&lt;</span><span class="mi">2</span> <span class="n">x</span> <span class="mi">4</span> <span class="ow">in</span> <span class="n">weights</span><span class="o">&gt;</span><span class="p">,</span> <span class="o">&lt;</span><span class="mi">4</span> <span class="n">x</span> <span class="mi">2</span> <span class="n">out</span> <span class="n">weights</span><span class="o">&gt;</span> <span class="o">&lt;</span><span class="n">padding</span><span class="o">&gt;</span> <span class="p">]</span> <span class="c1"># `q` adapter for layer 0</span>
<span class="p">[</span> <span class="o">&lt;</span><span class="mi">4</span> <span class="n">x</span> <span class="mi">4</span> <span class="ow">in</span> <span class="n">weights</span><span class="o">&gt;</span><span class="p">,</span> <span class="o">&lt;</span><span class="mi">4</span> <span class="n">x</span> <span class="mi">4</span> <span class="n">out</span> <span class="n">weights</span><span class="o">&gt;</span> <span class="o">&lt;</span><span class="n">padding</span><span class="o">&gt;</span> <span class="p">]</span> <span class="c1"># `k` adapter for layer 0</span>
<span class="p">[</span> <span class="o">&lt;</span><span class="mi">2</span> <span class="n">x</span> <span class="mi">4</span> <span class="ow">in</span> <span class="n">weights</span><span class="o">&gt;</span><span class="p">,</span> <span class="o">&lt;</span><span class="mi">4</span> <span class="n">x</span> <span class="mi">2</span> <span class="n">out</span> <span class="n">weights</span><span class="o">&gt;</span> <span class="o">&lt;</span><span class="n">padding</span><span class="o">&gt;</span> <span class="p">]</span> <span class="c1"># `q` adapter for layer 1</span>
<span class="p">[</span> <span class="o">&lt;</span><span class="mi">4</span> <span class="n">x</span> <span class="mi">4</span> <span class="ow">in</span> <span class="n">weights</span><span class="o">&gt;</span><span class="p">,</span> <span class="o">&lt;</span><span class="mi">4</span> <span class="n">x</span> <span class="mi">4</span> <span class="n">out</span> <span class="n">weights</span><span class="o">&gt;</span> <span class="o">&lt;</span><span class="n">padding</span><span class="o">&gt;</span> <span class="p">]</span> <span class="c1"># `k` adapter for layer 1</span>
<span class="p">[</span> <span class="o">&lt;</span><span class="mi">2</span> <span class="n">x</span> <span class="mi">4</span> <span class="ow">in</span> <span class="n">weights</span><span class="o">&gt;</span><span class="p">,</span> <span class="o">&lt;</span><span class="mi">4</span> <span class="n">x</span> <span class="mi">2</span> <span class="n">out</span> <span class="n">weights</span><span class="o">&gt;</span> <span class="o">&lt;</span><span class="n">padding</span><span class="o">&gt;</span> <span class="p">]</span> <span class="c1"># `q` adapter for layer 2</span>
<span class="p">[</span> <span class="o">&lt;</span><span class="mi">8</span> <span class="n">x</span> <span class="mi">4</span> <span class="ow">in</span> <span class="n">weights</span><span class="o">&gt;</span><span class="p">,</span> <span class="o">&lt;</span><span class="mi">4</span> <span class="n">x</span> <span class="mi">8</span> <span class="n">out</span> <span class="n">weights</span><span class="o">&gt;</span> <span class="p">]</span> <span class="c1"># `q` adapter for layer 3. Note the final layer has a adapter size of 8</span>
<span class="p">]</span>
</pre></div>
</div>
</section>
<section id="lora-module-id-mapping">
<h3>LoRA Module id mapping<a class="headerlink" href="#lora-module-id-mapping" title="Link to this heading">#</a></h3>
<div class="pst-scrollable-table-container"><table class="table">
<thead>
<tr class="row-odd"><th class="head"><p>module name (as specified in <code class="docutils literal notranslate"><span class="pre">convert_checkpoint.py</span></code> scripts)</p></th>
<th class="head"><p>module id</p></th>
<th class="head"><p>description</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>attn_qkv</p></td>
<td><p>0</p></td>
<td><p>compbined qkv adapter</p></td>
</tr>
<tr class="row-odd"><td><p>attn_q</p></td>
<td><p>1</p></td>
<td><p>q adapter</p></td>
</tr>
<tr class="row-even"><td><p>attn_k</p></td>
<td><p>2</p></td>
<td><p>k adapter</p></td>
</tr>
<tr class="row-odd"><td><p>attn_v</p></td>
<td><p>3</p></td>
<td><p>v adapter</p></td>
</tr>
<tr class="row-even"><td><p>attn_dense</p></td>
<td><p>4</p></td>
<td><p>adapter for the dense layer in attention</p></td>
</tr>
<tr class="row-odd"><td><p>mlp_h_to_4h</p></td>
<td><p>5</p></td>
<td><p>for llama2 adapter for gated mlp layer after attention / RMSNorm: up projection</p></td>
</tr>
<tr class="row-even"><td><p>mlp_4h_to_h</p></td>
<td><p>6</p></td>
<td><p>for llama2 adapter for gated mlp layer after attention / RMSNorm: down projection</p></td>
</tr>
<tr class="row-odd"><td><p>mlp_gate</p></td>
<td><p>7</p></td>
<td><p>for llama2 adapter for gated mlp later after attention / RMSNorm: gate</p></td>
</tr>
<tr class="row-even"><td><p>cross_attn_qkv</p></td>
<td><p>8</p></td>
<td><p>compbined qkv adapter for cross attention</p></td>
</tr>
<tr class="row-odd"><td><p>cross_attn_q</p></td>
<td><p>9</p></td>
<td><p>q adapter for cross attention</p></td>
</tr>
<tr class="row-even"><td><p>cross_attn_k</p></td>
<td><p>10</p></td>
<td><p>k adapter for cross attention</p></td>
</tr>
<tr class="row-odd"><td><p>cross_attn_v</p></td>
<td><p>11</p></td>
<td><p>v adapter for cross attention</p></td>
</tr>
<tr class="row-even"><td><p>cross_attn_dense</p></td>
<td><p>12</p></td>
<td><p>adapter for the dense layer in cross attention</p></td>
</tr>
<tr class="row-odd"><td><p>moe_h_to_4h</p></td>
<td><p>13</p></td>
<td><p>for mixtral adapter for expert mlp layer: up projection</p></td>
</tr>
<tr class="row-even"><td><p>moe_4h_to_h</p></td>
<td><p>14</p></td>
<td><p>for mixtral adapter for expert mlp layer: down projection</p></td>
</tr>
<tr class="row-odd"><td><p>moe_gate</p></td>
<td><p>15</p></td>
<td><p>for mixtral adapter for expert mlp layer: gate</p></td>
</tr>
<tr class="row-even"><td><p>moe_router</p></td>
<td><p>16</p></td>
<td><p>for mixtral adapter for expert router layer</p></td>
</tr>
<tr class="row-odd"><td><p>mlp_router</p></td>
<td><p>17</p></td>
<td><p>for qwen2-moe adapter for shared expert gate layer</p></td>
</tr>
</tbody>
</table>
</div>
</section>
<section id="loracache-configuration">
<h3>LoraCache configuration<a class="headerlink" href="#loracache-configuration" title="Link to this heading">#</a></h3>
<p>The core idea is that we will have a fixed size, 2-level LoRA cache in TRT-LLM. The higher level cache resides on the host and the lower level is on GPU (distinct from the existing KV cache). Sizes of both are user configurable.</p>
<p>The CPU cache is configured to be a max size. The GPU cache is configured to a percentage of free GPU memory after engine load. As requests come in LoRAs are stored in the host cache.</p>
<p>As requests are scheduled for execution LoRAs are loaded into the GPU cache.</p>
</section>
<section id="lora-with-tensor-parallel">
<h3>LoRA with tensor parallel<a class="headerlink" href="#lora-with-tensor-parallel" title="Link to this heading">#</a></h3>
<p>The partition of tensor parallel for LoRA is special. There are two cases: <code class="docutils literal notranslate"><span class="pre">RowLinear</span></code> and <code class="docutils literal notranslate"><span class="pre">ColumnLinear</span></code>. Assume we have a linear layer and the input feature size is <code class="docutils literal notranslate"><span class="pre">K</span></code> and the output feature size is <code class="docutils literal notranslate"><span class="pre">N</span></code>. Then, the shape of the weight is <code class="docutils literal notranslate"><span class="pre">[K,</span> <span class="pre">N]</span></code>.</p>
<p>First, consider this linear layer is a <code class="docutils literal notranslate"><span class="pre">ColumnLinear</span></code> layer. When we partition the weight, we split the weight by column with <code class="docutils literal notranslate"><span class="pre">tp_size</span></code>. Then, there are <code class="docutils literal notranslate"><span class="pre">tp_size</span></code> split weights and the shapes of these weights are <code class="docutils literal notranslate"><span class="pre">[K,</span> <span class="pre">N</span> <span class="pre">//</span> <span class="pre">tp_size]</span></code>. When we apply LoRA adapter on such <code class="docutils literal notranslate"><span class="pre">ColumnLinear</span></code> layer, the shapes of original two weights are <code class="docutils literal notranslate"><span class="pre">[K,</span> <span class="pre">lora_rank]</span></code> and <code class="docutils literal notranslate"><span class="pre">[lora_rank,</span> <span class="pre">N]</span></code>. So, we only partition the second weight and get <code class="docutils literal notranslate"><span class="pre">tp_size</span></code> split weights with shapes <code class="docutils literal notranslate"><span class="pre">[lora_rank,</span> <span class="pre">N</span> <span class="pre">//</span> <span class="pre">tp_size]</span></code>. For the first weight, each GPU maintains the same entire weight (with shape <code class="docutils literal notranslate"><span class="pre">[K,</span> <span class="pre">lora_rank]</span></code>).</p>
<p>Next, consider this linear layer is a <code class="docutils literal notranslate"><span class="pre">RowLinear</span></code> layer. When we partition the weight, we split the weight by row with <code class="docutils literal notranslate"><span class="pre">tp_size</span></code>. Then, there are <code class="docutils literal notranslate"><span class="pre">tp_size</span></code> split weights and the shapes of these weights are <code class="docutils literal notranslate"><span class="pre">[K</span> <span class="pre">//</span> <span class="pre">tp_size,</span> <span class="pre">N]</span></code>. When we apply LoRA adapter on such <code class="docutils literal notranslate"><span class="pre">RowLinear</span></code> layer, the shapes of original two weights are <code class="docutils literal notranslate"><span class="pre">[K,</span> <span class="pre">lora_rank]</span></code> and <code class="docutils literal notranslate"><span class="pre">[lora_rank,</span> <span class="pre">N]</span></code>. So, we only partition the first weight and get <code class="docutils literal notranslate"><span class="pre">tp_size</span></code> split weights with shapes <code class="docutils literal notranslate"><span class="pre">[K</span> <span class="pre">//</span> <span class="pre">tp_size,</span> <span class="pre">lora_rank]</span></code>. For the second weight, each GPU maintains the same entire weight (with shape <code class="docutils literal notranslate"><span class="pre">[lora_rank,</span> <span class="pre">N]</span></code>).</p>
</section>
</section>
</section>
</article>
<footer class="prev-next-footer d-print-none">
<div class="prev-next-area">
<a class="left-prev"
href="inference-request.html"
title="previous page">
<i class="fa-solid fa-angle-left"></i>
<div class="prev-next-info">
<p class="prev-next-subtitle">previous</p>
<p class="prev-next-title">Inference Request</p>
</div>
</a>
<a class="right-next"
href="expert-parallelism.html"
title="next page">
<div class="prev-next-info">
<p class="prev-next-subtitle">next</p>
<p class="prev-next-title">Expert Parallelism in TensorRT-LLM</p>
</div>
<i class="fa-solid fa-angle-right"></i>
</a>
</div>
</footer>
</div>
<dialog id="pst-secondary-sidebar-modal"></dialog>
<div id="pst-secondary-sidebar" class="bd-sidebar-secondary bd-toc"><div class="sidebar-secondary-items sidebar-secondary__inner">
<div class="sidebar-secondary-item">
<div
id="pst-page-navigation-heading-2"
class="page-toc tocsection onthispage">
<i class="fa-solid fa-list"></i> On this page
</div>
<nav class="bd-toc-nav page-toc" aria-labelledby="pst-page-navigation-heading-2">
<ul class="visible nav section-nav flex-column">
<li class="toc-h2 nav-item toc-entry"><a class="reference internal nav-link" href="#lora-tensor-format-details">LoRA tensor format details</a><ul class="nav section-nav flex-column">
<li class="toc-h3 nav-item toc-entry"><a class="reference internal nav-link" href="#example-lora-tensors">Example LoRA tensors</a></li>
<li class="toc-h3 nav-item toc-entry"><a class="reference internal nav-link" href="#lora-module-id-mapping">LoRA Module id mapping</a></li>
<li class="toc-h3 nav-item toc-entry"><a class="reference internal nav-link" href="#loracache-configuration">LoraCache configuration</a></li>
<li class="toc-h3 nav-item toc-entry"><a class="reference internal nav-link" href="#lora-with-tensor-parallel">LoRA with tensor parallel</a></li>
</ul>
</li>
</ul>
</nav></div>
</div></div>
</div>
<footer class="bd-footer-content">
</footer>
</main>
</div>
</div>
<!-- Scripts loaded after <body> so the DOM is not blocked -->
<script defer src="../_static/scripts/bootstrap.js?digest=8878045cc6db502f8baf"></script>
<script defer src="../_static/scripts/pydata-sphinx-theme.js?digest=8878045cc6db502f8baf"></script>
<footer class="bd-footer">
<div class="bd-footer__inner bd-page-width">
<div class="footer-items__start">
<div class="footer-item">
<a class="footer-brand logo" href="https://www.nvidia.com">
<img src="../_static/nvidia-logo-horiz-rgb-1c-blk-for-screen.svg" class="logo__image only-light" alt="NVIDIA"/>
<img src="../_static/nvidia-logo-horiz-rgb-1c-wht-for-screen.svg" class="logo__image only-dark" alt="NVIDIA"/>
</a></div>
<div class="footer-item">
<div class="footer-links">
<a class="external" href="https://www.nvidia.com/en-us/about-nvidia/privacy-policy/">Privacy Policy</a>
|
<a class="external" href="https://www.nvidia.com/en-us/about-nvidia/privacy-center/">Manage My Privacy</a>
|
<a class="external" href="https://www.nvidia.com/en-us/preferences/start/">Do Not Sell or Share My Data</a>
|
<a class="external" href="https://www.nvidia.com/en-us/about-nvidia/terms-of-service/">Terms of Service</a>
|
<a class="external" href="https://www.nvidia.com/en-us/about-nvidia/accessibility/">Accessibility</a>
|
<a class="external" href="https://www.nvidia.com/en-us/about-nvidia/company-policies/">Corporate Policies</a>
|
<a class="external" href="https://www.nvidia.com/en-us/product-security/">Product Security</a>
|
<a class="external" href="https://www.nvidia.com/en-us/contact/">Contact</a>
</div>
</div>
<div class="footer-item">
<p class="copyright">
Copyright © 2024, NVidia.
<br/>
</p>
</div>
</div>
</div>
</footer>
</body>
</html>
Reference in New Issue View Git Blame Copy Permalink