mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-14 06:27:45 +08:00
1701 lines
196 KiB
HTML
1701 lines
196 KiB
HTML
|
||
|
||
<!DOCTYPE html>
|
||
|
||
|
||
<html lang="en" data-content_root="../../../" >
|
||
|
||
<head>
|
||
<meta charset="utf-8" />
|
||
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
|
||
<title>tensorrt_llm.llmapi.llm — 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=8f2a1f02" />
|
||
<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" />
|
||
<link rel="stylesheet" type="text/css" href="../../../_static/autodoc_pydantic.css" />
|
||
<link rel="stylesheet" type="text/css" href="../../../_static/togglebutton.css?v=13237357" />
|
||
<link rel="stylesheet" type="text/css" href="../../../_static/custom.css?v=95073da6" />
|
||
|
||
<!-- 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=9a2dae69"></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>let toggleHintShow = 'Click to show';</script>
|
||
<script>let toggleHintHide = 'Click to hide';</script>
|
||
<script>let toggleOpenOnPrint = 'true';</script>
|
||
<script src="../../../_static/togglebutton.js?v=4a39c7ea"></script>
|
||
<script>var togglebuttonSelector = '.toggle, .admonition.dropdown';</script>
|
||
<script>var togglebuttonSelector = '.toggle, .admonition.dropdown';</script>
|
||
<script>DOCUMENTATION_OPTIONS.pagename = '_modules/tensorrt_llm/llmapi/llm';</script>
|
||
<script>
|
||
DOCUMENTATION_OPTIONS.theme_version = '0.16.1';
|
||
DOCUMENTATION_OPTIONS.theme_switcher_json_url = './_static/switcher.json';
|
||
DOCUMENTATION_OPTIONS.theme_switcher_version_match = '1.1.0rc3';
|
||
DOCUMENTATION_OPTIONS.show_version_warning_banner =
|
||
false;
|
||
</script>
|
||
<link rel="icon" href="../../../_static/favicon.png"/>
|
||
<link rel="index" title="Index" href="../../../genindex.html" />
|
||
<link rel="search" title="Search" href="../../../search.html" />
|
||
|
||
<meta name="viewport" content="width=device-width, initial-scale=1"/>
|
||
<meta name="docsearch:language" content="en"/>
|
||
<meta name="docsearch:version" content="1.1.0rc3" />
|
||
|
||
|
||
</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="me-auto navbar-header-items__center">
|
||
|
||
<div class="navbar-item">
|
||
|
||
|
||
<div class="version-switcher__container dropdown pst-js-only">
|
||
<button id="pst-version-switcher-button-2"
|
||
type="button"
|
||
class="version-switcher__button btn btn-sm dropdown-toggle"
|
||
data-bs-toggle="dropdown"
|
||
aria-haspopup="listbox"
|
||
aria-controls="pst-version-switcher-list-2"
|
||
aria-label="Version switcher list"
|
||
>
|
||
Choose version <!-- this text may get changed later by javascript -->
|
||
<span class="caret"></span>
|
||
</button>
|
||
<div id="pst-version-switcher-list-2"
|
||
class="version-switcher__menu dropdown-menu list-group-flush py-0"
|
||
role="listbox" aria-labelledby="pst-version-switcher-button-2">
|
||
<!-- dropdown will be populated by javascript on page load -->
|
||
</div>
|
||
</div></div>
|
||
|
||
</div>
|
||
|
||
|
||
<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>
|
||
|
||
|
||
|
||
</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__center">
|
||
|
||
|
||
|
||
<div class="navbar-item">
|
||
|
||
|
||
<div class="version-switcher__container dropdown pst-js-only">
|
||
<button id="pst-version-switcher-button-3"
|
||
type="button"
|
||
class="version-switcher__button btn btn-sm dropdown-toggle"
|
||
data-bs-toggle="dropdown"
|
||
aria-haspopup="listbox"
|
||
aria-controls="pst-version-switcher-list-3"
|
||
aria-label="Version switcher list"
|
||
>
|
||
Choose version <!-- this text may get changed later by javascript -->
|
||
<span class="caret"></span>
|
||
</button>
|
||
<div id="pst-version-switcher-list-3"
|
||
class="version-switcher__menu dropdown-menu list-group-flush py-0"
|
||
role="listbox" aria-labelledby="pst-version-switcher-button-3">
|
||
<!-- dropdown will be populated by javascript on page load -->
|
||
</div>
|
||
</div></div>
|
||
|
||
|
||
</div>
|
||
|
||
|
||
|
||
<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/containers.html">Pre-built release container images on NGC</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../installation/linux.html">Installing on Linux via <code class="docutils literal notranslate"><span class="pre">pip</span></code></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>
|
||
</ul>
|
||
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Deployment Guide</span></p>
|
||
<ul class="nav bd-sidenav">
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../deployment-guide/quick-start-recipe-for-llama4-scout-on-trtllm.html">Quick Start Recipe for Llama4 Scout 17B on TensorRT-LLM - Blackwell & Hopper Hardware</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../deployment-guide/quick-start-recipe-for-deepseek-r1-on-trtllm.html">Quick Start Recipe for DeepSeek R1 on TensorRT-LLM - Blackwell & Hopper Hardware</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../deployment-guide/quick-start-recipe-for-llama3.3-70b-on-trtllm.html">Quick Start Recipe for Llama3.3 70B on TensorRT-LLM - Blackwell & Hopper Hardware</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../deployment-guide/quick-start-recipe-for-gpt-oss-on-trtllm.html">Quick Start Recipe for GPT-OSS on TensorRT-LLM - Blackwell Hardware</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">LLM 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">Examples</span></p>
|
||
<ul class="nav bd-sidenav">
|
||
<li class="toctree-l1 has-children"><a class="reference internal" href="../../../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 class="simple">
|
||
</ul>
|
||
</details></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../examples/customization.html">LLM Common Customizations</a></li>
|
||
<li class="toctree-l1 has-children"><a class="reference internal" href="../../../examples/llm_api_examples.html">LLM 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="../../../examples/llm_inference.html">Generate text</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_inference_async.html">Generate text asynchronously</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_inference_async_streaming.html">Generate text in streaming</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_inference_distributed.html">Distributed LLM Generation</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_guided_decoding.html">Generate text with guided decoding</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_logits_processor.html">Control generated text using logits processor</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_multilora.html">Generate text with multiple LoRA adapters</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_speculative_decoding.html">Speculative Decoding</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_kv_cache_connector.html">KV Cache Connector</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_runtime.html">Runtime Configuration Examples</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_sampling.html">Sampling Techniques Showcase</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_mgmn_llm_distributed.html">Run LLM-API with pytorch backend on Slurm</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_mgmn_trtllm_bench.html">Run trtllm-bench with pytorch backend on Slurm</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_mgmn_trtllm_serve.html">Run trtllm-serve with pytorch backend on Slurm</a></li>
|
||
</ul>
|
||
</details></li>
|
||
<li class="toctree-l1 has-children"><a class="reference internal" href="../../../examples/trtllm_serve_examples.html">Online Serving 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="../../../examples/curl_chat_client.html">Curl Chat Client</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/curl_chat_client_for_multimodal.html">Curl Chat Client For Multimodal</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/curl_completion_client.html">Curl Completion Client</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/deepseek_r1_reasoning_parser.html">Deepseek R1 Reasoning Parser</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/genai_perf_client.html">Genai Perf Client</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/genai_perf_client_for_multimodal.html">Genai Perf Client For Multimodal</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/openai_chat_client.html">OpenAI Chat Client</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/openai_chat_client_for_multimodal.html">OpenAI Chat Client for Multimodal</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/openai_completion_client.html">OpenAI Completion Client</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/openai_completion_client_for_lora.html">Openai Completion Client For Lora</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../examples/openai_completion_client_json_schema.html">OpenAI Completion Client with JSON Schema</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-bench.html">trtllm-bench</a></li>
|
||
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../commands/trtllm-build.html">trtllm-build</a></li>
|
||
<li class="toctree-l1 has-children"><a class="reference internal" href="../../../commands/trtllm-serve/index.html">trtllm-serve</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="../../../commands/trtllm-serve/trtllm-serve.html">trtllm-serve</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../commands/trtllm-serve/run-benchmark-with-trtllm-serve.html">Run benchmarking with <code class="docutils literal notranslate"><span class="pre">trtllm-serve</span></code></a></li>
|
||
</ul>
|
||
</details></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="nav bd-sidenav">
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../advanced/gpt-attention.html">Multi-Head, Multi-Query, and Group-Query Attention</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../advanced/gpt-runtime.html">C++ GPT Runtime</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../advanced/executor.html">Executor API</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../advanced/graph-rewriting.html">Graph Rewriting Module</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../advanced/lora.html">Run gpt-2b + LoRA using Executor / cpp runtime</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../advanced/expert-parallelism.html">Expert Parallelism in TensorRT-LLM</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../advanced/kv-cache-management.html">KV Cache Management: Pools, Blocks, and Events</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../advanced/kv-cache-reuse.html">KV cache reuse</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../advanced/speculative-decoding.html">Speculative Sampling</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../advanced/disaggregated-service.html">Disaggregated-Service (Prototype)</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>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../reference/ci-overview.html">Continuous Integration Overview</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../reference/dev-containers.html">Using Dev Containers</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>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../blogs/tech_blog/blog10_ADP_Balance_Strategy.html">ADP Balance Strategy</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../blogs/tech_blog/blog1_Pushing_Latency_Boundaries_Optimizing_DeepSeek-R1_Performance_on_NVIDIA_B200_GPUs.html">Pushing Latency Boundaries: Optimizing DeepSeek-R1 Performance on NVIDIA B200 GPUs</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../blogs/tech_blog/blog2_DeepSeek_R1_MTP_Implementation_and_Optimization.html">DeepSeek R1 MTP Implementation and Optimization</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../blogs/tech_blog/blog3_Optimizing_DeepSeek_R1_Throughput_on_NVIDIA_Blackwell_GPUs.html">Optimizing DeepSeek R1 Throughput on NVIDIA Blackwell GPUs: A Deep Dive for Developers</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../blogs/tech_blog/blog4_Scaling_Expert_Parallelism_in_TensorRT-LLM.html">Scaling Expert Parallelism in TensorRT-LLM (Part 1: Design and Implementation of Large-scale EP)</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../blogs/tech_blog/blog5_Disaggregated_Serving_in_TensorRT-LLM.html">Disaggregated Serving in TensorRT-LLM</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../blogs/tech_blog/blog6_Llama4_maverick_eagle_guide.html">How to launch Llama4 Maverick + Eagle3 TensorRT-LLM server</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../blogs/tech_blog/blog7_NGram_performance_Analysis_And_Auto_Enablement.html">N-Gram Speculative Decoding in TensorRT‑LLM</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../blogs/tech_blog/blog8_Scaling_Expert_Parallelism_in_TensorRT-LLM_part2.html">Scaling Expert Parallelism in TensorRT-LLM (Part 2: Performance Status and Optimization)</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../blogs/tech_blog/blog9_Deploying_GPT_OSS_on_TRTLLM.html">Running a High Performance GPT-OSS-120B Inference Server with TensorRT-LLM</a></li>
|
||
</ul>
|
||
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Use TensorRT Engine</span></p>
|
||
<ul class="nav bd-sidenav">
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../legacy/tensorrt_quickstart.html">LLM API with TensorRT Engine</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"><a href="../../index.html" class="nav-link">Module code</a></li>
|
||
|
||
<li class="breadcrumb-item active" aria-current="page"><span class="ellipsis">tensorrt_llm.llmapi.llm</span></li>
|
||
</ul>
|
||
</nav>
|
||
</div>
|
||
|
||
</div>
|
||
|
||
|
||
</div>
|
||
</div>
|
||
|
||
|
||
|
||
|
||
<div id="searchbox"></div>
|
||
<article class="bd-article">
|
||
|
||
<h1>Source code for tensorrt_llm.llmapi.llm</h1><div class="highlight"><pre>
|
||
<span></span><span class="kn">import</span><span class="w"> </span><span class="nn">atexit</span>
|
||
<span class="kn">import</span><span class="w"> </span><span class="nn">json</span>
|
||
<span class="kn">import</span><span class="w"> </span><span class="nn">os</span>
|
||
<span class="kn">import</span><span class="w"> </span><span class="nn">shutil</span>
|
||
<span class="kn">import</span><span class="w"> </span><span class="nn">socket</span>
|
||
<span class="kn">import</span><span class="w"> </span><span class="nn">tempfile</span>
|
||
<span class="kn">import</span><span class="w"> </span><span class="nn">time</span>
|
||
<span class="kn">import</span><span class="w"> </span><span class="nn">weakref</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">pathlib</span><span class="w"> </span><span class="kn">import</span> <span class="n">Path</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">typing</span><span class="w"> </span><span class="kn">import</span> <span class="n">Any</span><span class="p">,</span> <span class="n">List</span><span class="p">,</span> <span class="n">Literal</span><span class="p">,</span> <span class="n">Optional</span><span class="p">,</span> <span class="n">Sequence</span><span class="p">,</span> <span class="n">Union</span>
|
||
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">tqdm</span><span class="w"> </span><span class="kn">import</span> <span class="n">tqdm</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">transformers</span><span class="w"> </span><span class="kn">import</span> <span class="n">PreTrainedTokenizerBase</span>
|
||
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">tensorrt_llm.inputs.data</span><span class="w"> </span><span class="kn">import</span> <span class="n">TextPrompt</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">tensorrt_llm.inputs.multimodal</span><span class="w"> </span><span class="kn">import</span> <span class="n">MultimodalParams</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">tensorrt_llm.inputs.registry</span><span class="w"> </span><span class="kn">import</span> <span class="n">DefaultInputProcessor</span>
|
||
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">.._utils</span><span class="w"> </span><span class="kn">import</span> <span class="n">nvtx_range_debug</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">..bindings</span><span class="w"> </span><span class="kn">import</span> <span class="n">executor</span> <span class="k">as</span> <span class="n">tllm</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">..builder</span><span class="w"> </span><span class="kn">import</span> <span class="n">EngineConfig</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">..disaggregated_params</span><span class="w"> </span><span class="kn">import</span> <span class="n">DisaggregatedParams</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">..executor</span><span class="w"> </span><span class="kn">import</span> <span class="p">(</span><span class="n">DetokenizedGenerationResultBase</span><span class="p">,</span> <span class="n">GenerationExecutor</span><span class="p">,</span>
|
||
<span class="n">GenerationResult</span><span class="p">,</span> <span class="n">IterationResult</span><span class="p">,</span> <span class="n">LoRARequest</span><span class="p">,</span>
|
||
<span class="n">PostprocWorkerConfig</span><span class="p">,</span> <span class="n">PromptAdapterRequest</span><span class="p">)</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">..executor.postproc_worker</span><span class="w"> </span><span class="kn">import</span> <span class="n">PostprocParams</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">..executor.utils</span><span class="w"> </span><span class="kn">import</span> <span class="p">(</span><span class="n">create_mpi_comm_session</span><span class="p">,</span>
|
||
<span class="n">get_spawn_proxy_process_env</span><span class="p">)</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">..inputs</span><span class="w"> </span><span class="kn">import</span> <span class="p">(</span><span class="n">PromptInputs</span><span class="p">,</span> <span class="n">create_input_processor</span><span class="p">,</span>
|
||
<span class="n">create_input_processor_with_hash</span><span class="p">,</span> <span class="n">prompt_inputs</span><span class="p">)</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">..logger</span><span class="w"> </span><span class="kn">import</span> <span class="n">logger</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">..sampling_params</span><span class="w"> </span><span class="kn">import</span> <span class="n">SamplingParams</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">..scheduling_params</span><span class="w"> </span><span class="kn">import</span> <span class="n">SchedulingParams</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">.llm_args</span><span class="w"> </span><span class="kn">import</span> <span class="p">(</span><span class="n">TORCH_LLMARGS_EXPLICIT_DOCSTRING</span><span class="p">,</span>
|
||
<span class="n">TRT_LLMARGS_EXPLICIT_DOCSTRING</span><span class="p">,</span> <span class="n">PeftCacheConfig</span><span class="p">,</span>
|
||
<span class="n">PybindMirror</span><span class="p">,</span> <span class="n">TorchLlmArgs</span><span class="p">,</span> <span class="n">TrtLlmArgs</span><span class="p">)</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">.llm_utils</span><span class="w"> </span><span class="kn">import</span> <span class="p">(</span><span class="n">CachedModelLoader</span><span class="p">,</span> <span class="n">KvCacheRetentionConfig</span><span class="p">,</span>
|
||
<span class="n">LlmBuildStats</span><span class="p">,</span> <span class="n">ModelLoader</span><span class="p">,</span> <span class="n">_ModelRuntimeContext</span><span class="p">)</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">.mpi_session</span><span class="w"> </span><span class="kn">import</span> <span class="n">MpiPoolSession</span><span class="p">,</span> <span class="n">external_mpi_comm_available</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">.tokenizer</span><span class="w"> </span><span class="kn">import</span> <span class="n">TokenizerBase</span><span class="p">,</span> <span class="n">_xgrammar_tokenizer_info</span>
|
||
<span class="c1"># TODO[chunweiy]: move the following symbols back to utils scope, and remove the following import</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">.utils</span><span class="w"> </span><span class="kn">import</span> <span class="p">(</span><span class="n">append_docstring</span><span class="p">,</span> <span class="n">exception_handler</span><span class="p">,</span> <span class="n">get_device_count</span><span class="p">,</span>
|
||
<span class="n">print_colored_debug</span><span class="p">,</span> <span class="n">set_api_status</span><span class="p">)</span>
|
||
|
||
|
||
<div class="viewcode-block" id="RequestOutput">
|
||
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.RequestOutput">[docs]</a>
|
||
<span class="k">class</span><span class="w"> </span><span class="nc">RequestOutput</span><span class="p">(</span><span class="n">DetokenizedGenerationResultBase</span><span class="p">,</span> <span class="n">GenerationResult</span><span class="p">):</span>
|
||
<span class="w"> </span><span class="sd">"""The output data of a completion request to the LLM.</span>
|
||
|
||
<span class="sd"> Attributes:</span>
|
||
<span class="sd"> request_id (int): The unique ID of the request.</span>
|
||
<span class="sd"> prompt (str, optional): The prompt string of the request.</span>
|
||
<span class="sd"> prompt_token_ids (List[int]): The token ids of the prompt.</span>
|
||
<span class="sd"> outputs (List[CompletionOutput]): The output sequences of the request.</span>
|
||
<span class="sd"> context_logits (torch.Tensor, optional): The logits on the prompt token ids.</span>
|
||
<span class="sd"> mm_embedding_handle (Dict[str, Any], optional): The multimodal embedding handle of the request.</span>
|
||
<span class="sd"> finished (bool): Whether the whole request is finished.</span>
|
||
<span class="sd"> """</span>
|
||
|
||
<div class="viewcode-block" id="RequestOutput.__init__">
|
||
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.RequestOutput.__init__">[docs]</a>
|
||
<span class="k">def</span><span class="w"> </span><span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="vm">__class__</span><span class="o">.</span><span class="vm">__name__</span><span class="si">}</span><span class="s2"> is designed to be instantiated using </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="vm">__class__</span><span class="o">.</span><span class="vm">__name__</span><span class="si">}</span><span class="s2">._from_generation_result by GenerationExecutor. "</span>
|
||
<span class="sa">f</span><span class="s2">"Users are not expected to create </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="vm">__class__</span><span class="o">.</span><span class="vm">__name__</span><span class="si">}</span><span class="s2"> directly."</span>
|
||
<span class="p">)</span></div>
|
||
|
||
|
||
<span class="nd">@classmethod</span>
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">_from_generation_result</span><span class="p">(</span>
|
||
<span class="bp">cls</span><span class="p">,</span>
|
||
<span class="n">generation_result</span><span class="p">:</span> <span class="n">GenerationResult</span><span class="p">,</span>
|
||
<span class="n">prompt</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">tokenizer</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">TokenizerBase</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">)</span> <span class="o">-></span> <span class="s1">'RequestOutput'</span><span class="p">:</span>
|
||
<span class="n">inst</span> <span class="o">=</span> <span class="bp">cls</span><span class="o">.</span><span class="fm">__new__</span><span class="p">(</span><span class="bp">cls</span><span class="p">)</span>
|
||
<span class="n">inst</span><span class="o">.</span><span class="vm">__dict__</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="n">generation_result</span><span class="o">.</span><span class="vm">__dict__</span><span class="p">)</span>
|
||
<span class="n">inst</span><span class="o">.</span><span class="n">tokenizer</span> <span class="o">=</span> <span class="n">tokenizer</span>
|
||
<span class="n">inst</span><span class="o">.</span><span class="n">_streaming</span> <span class="o">=</span> <span class="n">generation_result</span><span class="o">.</span><span class="n">_streaming</span>
|
||
<span class="n">inst</span><span class="o">.</span><span class="n">_prompt</span> <span class="o">=</span> <span class="n">prompt</span>
|
||
<span class="k">return</span> <span class="n">inst</span>
|
||
|
||
<span class="nd">@property</span>
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">prompt</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]:</span>
|
||
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_prompt</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">_repr_fields</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
||
<span class="k">return</span> <span class="p">[</span>
|
||
<span class="s2">"request_id"</span><span class="p">,</span> <span class="s2">"prompt"</span><span class="p">,</span> <span class="s2">"prompt_token_ids"</span><span class="p">,</span> <span class="s2">"outputs"</span><span class="p">,</span> <span class="s2">"finished"</span><span class="p">,</span>
|
||
<span class="s2">"mm_embedding_handle"</span>
|
||
<span class="p">]</span></div>
|
||
|
||
|
||
|
||
<span class="n">TRT_LLM_DOCSTRING</span> <span class="o">=</span> <span class="n">TRT_LLMARGS_EXPLICIT_DOCSTRING</span> <span class="o">+</span> <span class="s2">"""</span>
|
||
|
||
<span class="s2"> Attributes:</span>
|
||
<span class="s2"> tokenizer (tensorrt_llm.llmapi.tokenizer.TokenizerBase, optional): The tokenizer loaded by LLM instance, if any.</span>
|
||
<span class="s2"> workspace (pathlib.Path): The directory to store intermediate files.</span>
|
||
<span class="s2"> llm_id (str): The unique ID of the LLM instance.</span>
|
||
<span class="s2">"""</span>
|
||
|
||
<span class="n">TORCH_LLM_DOCSTRING</span> <span class="o">=</span> <span class="n">TORCH_LLMARGS_EXPLICIT_DOCSTRING</span> <span class="o">+</span> <span class="s2">"""</span>
|
||
|
||
<span class="s2"> Attributes:</span>
|
||
<span class="s2"> tokenizer (tensorrt_llm.llmapi.tokenizer.TokenizerBase, optional): The tokenizer loaded by LLM instance, if any.</span>
|
||
<span class="s2"> llm_id (str): The unique ID of the LLM instance.</span>
|
||
<span class="s2">"""</span>
|
||
|
||
|
||
<span class="k">class</span><span class="w"> </span><span class="nc">BaseLLM</span><span class="p">:</span>
|
||
<span class="w"> </span><span class="sd">"""</span>
|
||
<span class="sd"> The base class for all LLM classes.</span>
|
||
<span class="sd"> """</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span>
|
||
<span class="n">model</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Path</span><span class="p">],</span>
|
||
<span class="n">tokenizer</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Path</span><span class="p">,</span> <span class="n">TokenizerBase</span><span class="p">,</span>
|
||
<span class="n">PreTrainedTokenizerBase</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">tokenizer_mode</span><span class="p">:</span> <span class="n">Literal</span><span class="p">[</span><span class="s1">'auto'</span><span class="p">,</span> <span class="s1">'slow'</span><span class="p">]</span> <span class="o">=</span> <span class="s1">'auto'</span><span class="p">,</span>
|
||
<span class="n">skip_tokenizer_init</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
|
||
<span class="n">trust_remote_code</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
|
||
<span class="n">tensor_parallel_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">1</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="s2">"auto"</span><span class="p">,</span>
|
||
<span class="n">revision</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">tokenizer_revision</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="o">**</span><span class="n">kwargs</span><span class="p">:</span> <span class="n">Any</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
|
||
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor_cls</span> <span class="o">=</span> <span class="n">kwargs</span><span class="o">.</span><span class="n">pop</span><span class="p">(</span><span class="s2">"executor_cls"</span><span class="p">,</span> <span class="n">GenerationExecutor</span><span class="p">)</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_llm_id</span> <span class="o">=</span> <span class="kc">None</span>
|
||
|
||
<span class="n">log_level</span> <span class="o">=</span> <span class="n">logger</span><span class="o">.</span><span class="n">level</span>
|
||
<span class="n">logger</span><span class="o">.</span><span class="n">set_level</span><span class="p">(</span><span class="s2">"info"</span><span class="p">)</span> <span class="c1"># force display the backend</span>
|
||
|
||
<span class="k">try</span><span class="p">:</span>
|
||
<span class="n">backend</span> <span class="o">=</span> <span class="n">kwargs</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s1">'backend'</span><span class="p">,</span> <span class="kc">None</span><span class="p">)</span>
|
||
<span class="k">if</span> <span class="n">backend</span> <span class="o">==</span> <span class="s2">"pytorch"</span><span class="p">:</span>
|
||
<span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s2">"Using LLM with PyTorch backend"</span><span class="p">)</span>
|
||
<span class="n">llm_args_cls</span> <span class="o">=</span> <span class="n">TorchLlmArgs</span>
|
||
<span class="k">elif</span> <span class="n">backend</span> <span class="o">==</span> <span class="s1">'_autodeploy'</span><span class="p">:</span>
|
||
<span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s2">"Using LLM with AutoDeploy backend"</span><span class="p">)</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">.._torch.auto_deploy.llm_args</span><span class="w"> </span><span class="kn">import</span> \
|
||
<span class="n">LlmArgs</span> <span class="k">as</span> <span class="n">AutoDeployLlmArgs</span>
|
||
<span class="n">llm_args_cls</span> <span class="o">=</span> <span class="n">AutoDeployLlmArgs</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="s2">"Using LLM with TensorRT backend"</span><span class="p">)</span>
|
||
<span class="n">llm_args_cls</span> <span class="o">=</span> <span class="n">TrtLlmArgs</span>
|
||
|
||
<span class="c1"># check the kwargs and raise ValueError directly</span>
|
||
<span class="n">valid_keys</span> <span class="o">=</span> <span class="nb">set</span><span class="p">(</span>
|
||
<span class="nb">list</span><span class="p">(</span><span class="n">llm_args_cls</span><span class="o">.</span><span class="n">model_fields</span><span class="o">.</span><span class="n">keys</span><span class="p">())</span> <span class="o">+</span>
|
||
<span class="p">[</span><span class="s1">'_mpi_session'</span><span class="p">,</span> <span class="s1">'backend'</span><span class="p">])</span>
|
||
<span class="k">for</span> <span class="n">key</span> <span class="ow">in</span> <span class="n">kwargs</span><span class="p">:</span>
|
||
<span class="k">if</span> <span class="n">key</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">valid_keys</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="vm">__class__</span><span class="o">.</span><span class="vm">__name__</span><span class="si">}</span><span class="s2"> got invalid argument: </span><span class="si">{</span><span class="n">key</span><span class="si">}</span><span class="s2">"</span>
|
||
<span class="p">)</span>
|
||
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">args</span> <span class="o">=</span> <span class="n">llm_args_cls</span><span class="o">.</span><span class="n">from_kwargs</span><span class="p">(</span>
|
||
<span class="n">model</span><span class="o">=</span><span class="n">model</span><span class="p">,</span>
|
||
<span class="n">tokenizer</span><span class="o">=</span><span class="n">tokenizer</span><span class="p">,</span>
|
||
<span class="n">tokenizer_mode</span><span class="o">=</span><span class="n">tokenizer_mode</span><span class="p">,</span>
|
||
<span class="n">skip_tokenizer_init</span><span class="o">=</span><span class="n">skip_tokenizer_init</span><span class="p">,</span>
|
||
<span class="n">trust_remote_code</span><span class="o">=</span><span class="n">trust_remote_code</span><span class="p">,</span>
|
||
<span class="n">tensor_parallel_size</span><span class="o">=</span><span class="n">tensor_parallel_size</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">,</span>
|
||
<span class="n">revision</span><span class="o">=</span><span class="n">revision</span><span class="p">,</span>
|
||
<span class="n">tokenizer_revision</span><span class="o">=</span><span class="n">tokenizer_revision</span><span class="p">,</span>
|
||
<span class="o">**</span><span class="n">kwargs</span><span class="p">)</span>
|
||
|
||
<span class="k">except</span> <span class="ne">Exception</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
|
||
<span class="n">logger</span><span class="o">.</span><span class="n">error</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"Failed to parse the arguments for the LLM constructor: </span><span class="si">{</span><span class="n">e</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span>
|
||
<span class="k">raise</span> <span class="n">e</span>
|
||
|
||
<span class="k">finally</span><span class="p">:</span>
|
||
<span class="n">logger</span><span class="o">.</span><span class="n">set_level</span><span class="p">(</span><span class="n">log_level</span><span class="p">)</span> <span class="c1"># restore the log level</span>
|
||
|
||
<span class="n">print_colored_debug</span><span class="p">(</span><span class="sa">f</span><span class="s2">"LLM.args.mpi_session: </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">mpi_session</span><span class="si">}</span><span class="se">\n</span><span class="s2">"</span><span class="p">,</span>
|
||
<span class="s2">"yellow"</span><span class="p">)</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">mpi_session</span>
|
||
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">is_multi_gpu</span><span class="p">:</span>
|
||
<span class="k">if</span> <span class="n">get_device_count</span><span class="p">(</span>
|
||
<span class="p">)</span> <span class="o"><</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">world_size_per_node</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"Only </span><span class="si">{</span><span class="n">get_device_count</span><span class="p">()</span><span class="si">}</span><span class="s2"> GPUs are available, but </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">world_size</span><span class="si">}</span><span class="s2"> are required."</span>
|
||
<span class="p">)</span>
|
||
|
||
<span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s1">'start MpiSession with </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">world_size</span><span class="si">}</span><span class="s1"> workers'</span>
|
||
<span class="p">)</span>
|
||
<span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span><span class="p">:</span>
|
||
<span class="n">mpi_process_pre_spawned</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">get_spawn_proxy_process_env</span><span class="p">()</span>
|
||
<span class="k">if</span> <span class="ow">not</span> <span class="n">mpi_process_pre_spawned</span><span class="p">:</span>
|
||
<span class="n">print_colored_debug</span><span class="p">(</span><span class="sa">f</span><span class="s2">"LLM create MpiPoolSession</span><span class="se">\n</span><span class="s2">"</span><span class="p">,</span>
|
||
<span class="s2">"yellow"</span><span class="p">)</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span> <span class="o">=</span> <span class="n">MpiPoolSession</span><span class="p">(</span>
|
||
<span class="n">n_workers</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">world_size</span><span class="p">)</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="n">print_colored_debug</span><span class="p">(</span><span class="sa">f</span><span class="s2">"LLM create MpiCommSession</span><span class="se">\n</span><span class="s2">"</span><span class="p">,</span>
|
||
<span class="s2">"yellow"</span><span class="p">)</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span> <span class="o">=</span> <span class="n">create_mpi_comm_session</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">world_size</span><span class="p">)</span>
|
||
|
||
<span class="k">try</span><span class="p">:</span>
|
||
<span class="c1"># Due to the Executor can only accept a engine path, we need to save the engine to a directory</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_engine_dir</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Path</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">GenerationExecutor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_on_trt_backend</span><span class="p">:</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_workspace</span> <span class="o">=</span> <span class="n">tempfile</span><span class="o">.</span><span class="n">TemporaryDirectory</span><span class="p">(</span>
|
||
<span class="n">suffix</span><span class="o">=</span><span class="s2">"-llm-workspace"</span><span class="p">,</span> <span class="nb">dir</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">workspace</span><span class="p">)</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_workspace</span> <span class="o">=</span> <span class="kc">None</span>
|
||
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_hf_model_dir</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Path</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
|
||
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">runtime_context</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">_ModelRuntimeContext</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">llm_build_stats</span> <span class="o">=</span> <span class="n">LlmBuildStats</span><span class="p">()</span>
|
||
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_build_model</span><span class="p">()</span>
|
||
|
||
<span class="k">except</span> <span class="ne">Exception</span><span class="p">:</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
|
||
<span class="k">raise</span>
|
||
|
||
<span class="n">exception_handler</span><span class="o">.</span><span class="n">register</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="s1">'shutdown'</span><span class="p">)</span>
|
||
<span class="n">atexit</span><span class="o">.</span><span class="n">register</span><span class="p">(</span><span class="n">LLM</span><span class="o">.</span><span class="n">_shutdown_wrapper</span><span class="p">,</span> <span class="n">weakref</span><span class="o">.</span><span class="n">ref</span><span class="p">(</span><span class="bp">self</span><span class="p">))</span>
|
||
|
||
<span class="nd">@property</span>
|
||
<span class="nd">@set_api_status</span><span class="p">(</span><span class="s2">"beta"</span><span class="p">)</span>
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">llm_id</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="nb">str</span><span class="p">:</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_llm_id</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="n">hostname</span> <span class="o">=</span> <span class="n">socket</span><span class="o">.</span><span class="n">gethostname</span><span class="p">()</span>
|
||
<span class="n">pid</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">getpid</span><span class="p">()</span>
|
||
<span class="n">timestamp</span> <span class="o">=</span> <span class="nb">int</span><span class="p">(</span><span class="n">time</span><span class="o">.</span><span class="n">time</span><span class="p">()</span> <span class="o">*</span> <span class="mi">1000</span><span class="p">)</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_llm_id</span> <span class="o">=</span> <span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="n">hostname</span><span class="si">}</span><span class="s2">-</span><span class="si">{</span><span class="n">pid</span><span class="si">}</span><span class="s2">-</span><span class="si">{</span><span class="n">timestamp</span><span class="si">}</span><span class="s2">"</span>
|
||
|
||
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_llm_id</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">generate</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="p">,</span>
|
||
<span class="n">inputs</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">PromptInputs</span><span class="p">,</span> <span class="n">Sequence</span><span class="p">[</span><span class="n">PromptInputs</span><span class="p">]],</span>
|
||
<span class="n">sampling_params</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span><span class="n">SamplingParams</span><span class="p">,</span>
|
||
<span class="n">List</span><span class="p">[</span><span class="n">SamplingParams</span><span class="p">]]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">use_tqdm</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span><span class="p">,</span>
|
||
<span class="n">lora_request</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span><span class="n">LoRARequest</span><span class="p">,</span>
|
||
<span class="n">Sequence</span><span class="p">[</span><span class="n">LoRARequest</span><span class="p">]]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">prompt_adapter_request</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span>
|
||
<span class="n">PromptAdapterRequest</span><span class="p">,</span> <span class="n">Sequence</span><span class="p">[</span><span class="n">PromptAdapterRequest</span><span class="p">]]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">kv_cache_retention_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span>
|
||
<span class="n">KvCacheRetentionConfig</span><span class="p">,</span> <span class="n">Sequence</span><span class="p">[</span><span class="n">KvCacheRetentionConfig</span><span class="p">]]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">disaggregated_params</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span>
|
||
<span class="n">DisaggregatedParams</span><span class="p">,</span> <span class="n">Sequence</span><span class="p">[</span><span class="n">DisaggregatedParams</span><span class="p">]]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">scheduling_params</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span><span class="n">SchedulingParams</span><span class="p">,</span>
|
||
<span class="n">List</span><span class="p">[</span><span class="n">SchedulingParams</span><span class="p">]]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="p">)</span> <span class="o">-></span> <span class="n">Union</span><span class="p">[</span><span class="n">RequestOutput</span><span class="p">,</span> <span class="n">List</span><span class="p">[</span><span class="n">RequestOutput</span><span class="p">]]:</span>
|
||
<span class="w"> </span><span class="sd">"""Generate output for the given prompts in the synchronous mode.</span>
|
||
<span class="sd"> Synchronous generation accepts either single prompt or batched prompts.</span>
|
||
|
||
<span class="sd"> Args:</span>
|
||
<span class="sd"> inputs (tensorrt_llm.inputs.data.PromptInputs, Sequence[tensorrt_llm.inputs.data.PromptInputs]): The prompt text or token ids.</span>
|
||
<span class="sd"> It can be single prompt or batched prompts.</span>
|
||
<span class="sd"> sampling_params (tensorrt_llm.sampling_params.SamplingParams, List[tensorrt_llm.sampling_params.SamplingParams], optional): The sampling params for the generation. Defaults to None.</span>
|
||
<span class="sd"> A default one will be used if not provided.</span>
|
||
<span class="sd"> use_tqdm (bool): Whether to use tqdm to display the progress bar. Defaults to True.</span>
|
||
<span class="sd"> lora_request (tensorrt_llm.executor.request.LoRARequest, Sequence[tensorrt_llm.executor.request.LoRARequest], optional):</span>
|
||
<span class="sd"> LoRA request to use for generation, if any. Defaults to None.</span>
|
||
<span class="sd"> prompt_adapter_request (tensorrt_llm.executor.request.PromptAdapterRequest, Sequence[tensorrt_llm.executor.request.PromptAdapterRequest], optional):</span>
|
||
<span class="sd"> Prompt Adapter request to use for generation, if any. Defaults to None.</span>
|
||
<span class="sd"> kv_cache_retention_config (tensorrt_llm.bindings.executor.KvCacheRetentionConfig, Sequence[tensorrt_llm.bindings.executor.KvCacheRetentionConfig], optional):</span>
|
||
<span class="sd"> Configuration for the request's retention in the KV Cache. Defaults to None.</span>
|
||
<span class="sd"> disaggregated_params (tensorrt_llm.disaggregated_params.DisaggregatedParams, Sequence[tensorrt_llm.disaggregated_params.DisaggregatedParams], optional):</span>
|
||
<span class="sd"> Disaggregated parameters. Defaults to None.</span>
|
||
<span class="sd"> scheduling_params (tensorrt_llm.scheduling_params.SchedulingParams, List[tensorrt_llm.scheduling_params.SchedulingParams], optional):</span>
|
||
<span class="sd"> Scheduling parameters. Defaults to None.</span>
|
||
<span class="sd"> Returns:</span>
|
||
<span class="sd"> Union[tensorrt_llm.llmapi.RequestOutput, List[tensorrt_llm.llmapi.RequestOutput]]: The output data of the completion request to the LLM.</span>
|
||
<span class="sd"> """</span>
|
||
<span class="n">unbatched</span> <span class="o">=</span> <span class="ow">not</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">inputs</span><span class="p">,</span> <span class="nb">list</span><span class="p">)</span>
|
||
<span class="k">if</span> <span class="ow">not</span> <span class="n">unbatched</span><span class="p">:</span>
|
||
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">inputs</span><span class="p">[</span><span class="mi">0</span><span class="p">],</span> <span class="nb">int</span><span class="p">):</span>
|
||
<span class="n">unbatched</span> <span class="o">=</span> <span class="kc">True</span>
|
||
|
||
<span class="k">if</span> <span class="n">unbatched</span><span class="p">:</span>
|
||
<span class="n">inputs</span> <span class="o">=</span> <span class="p">[</span><span class="n">inputs</span><span class="p">]</span>
|
||
|
||
<span class="n">inputs</span> <span class="o">=</span> <span class="p">[</span><span class="n">prompt_inputs</span><span class="p">(</span><span class="n">i</span><span class="p">)</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="n">inputs</span><span class="p">]</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">_item_at</span><span class="p">(</span><span class="n">maybe_batched</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">Any</span><span class="p">,</span> <span class="n">Sequence</span><span class="p">[</span><span class="n">Any</span><span class="p">]],</span> <span class="n">pos</span><span class="p">:</span> <span class="nb">int</span><span class="p">)</span> <span class="o">-></span> <span class="n">Any</span><span class="p">:</span>
|
||
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">maybe_batched</span><span class="p">,</span> <span class="nb">list</span><span class="p">):</span>
|
||
<span class="k">return</span> <span class="n">maybe_batched</span><span class="p">[</span><span class="n">pos</span><span class="p">]</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="k">return</span> <span class="n">maybe_batched</span>
|
||
|
||
<span class="n">futures</span> <span class="o">=</span> <span class="p">[]</span>
|
||
<span class="k">for</span> <span class="n">i</span><span class="p">,</span> <span class="n">request_inputs</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">inputs</span><span class="p">):</span>
|
||
<span class="n">future</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">generate_async</span><span class="p">(</span>
|
||
<span class="n">request_inputs</span><span class="p">,</span>
|
||
<span class="n">sampling_params</span><span class="o">=</span><span class="n">_item_at</span><span class="p">(</span><span class="n">sampling_params</span><span class="p">,</span> <span class="n">i</span><span class="p">),</span>
|
||
<span class="n">lora_request</span><span class="o">=</span><span class="n">_item_at</span><span class="p">(</span><span class="n">lora_request</span><span class="p">,</span> <span class="n">i</span><span class="p">),</span>
|
||
<span class="n">prompt_adapter_request</span><span class="o">=</span><span class="n">_item_at</span><span class="p">(</span><span class="n">prompt_adapter_request</span><span class="p">,</span> <span class="n">i</span><span class="p">),</span>
|
||
<span class="n">kv_cache_retention_config</span><span class="o">=</span><span class="n">_item_at</span><span class="p">(</span><span class="n">kv_cache_retention_config</span><span class="p">,</span>
|
||
<span class="n">i</span><span class="p">),</span>
|
||
<span class="n">disaggregated_params</span><span class="o">=</span><span class="n">_item_at</span><span class="p">(</span><span class="n">disaggregated_params</span><span class="p">,</span> <span class="n">i</span><span class="p">),</span>
|
||
<span class="n">scheduling_params</span><span class="o">=</span><span class="n">_item_at</span><span class="p">(</span><span class="n">scheduling_params</span><span class="p">,</span> <span class="n">i</span><span class="p">),</span>
|
||
<span class="n">streaming</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
|
||
<span class="n">futures</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">future</span><span class="p">)</span>
|
||
|
||
<span class="k">for</span> <span class="n">future</span> <span class="ow">in</span> <span class="n">tqdm</span><span class="p">(</span><span class="n">futures</span><span class="p">,</span>
|
||
<span class="n">desc</span><span class="o">=</span><span class="s2">"Processed requests"</span><span class="p">,</span>
|
||
<span class="n">dynamic_ncols</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
||
<span class="n">disable</span><span class="o">=</span><span class="ow">not</span> <span class="n">use_tqdm</span><span class="p">):</span>
|
||
<span class="n">future</span><span class="o">.</span><span class="n">result</span><span class="p">()</span>
|
||
|
||
<span class="k">if</span> <span class="n">unbatched</span><span class="p">:</span>
|
||
<span class="n">futures</span> <span class="o">=</span> <span class="n">futures</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
|
||
|
||
<span class="k">return</span> <span class="n">futures</span>
|
||
|
||
<span class="nd">@nvtx_range_debug</span><span class="p">(</span><span class="s2">"LLM.generate_async"</span><span class="p">,</span> <span class="n">color</span><span class="o">=</span><span class="s2">"green"</span><span class="p">,</span> <span class="n">category</span><span class="o">=</span><span class="s2">"LLM"</span><span class="p">)</span>
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">generate_async</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="p">,</span>
|
||
<span class="n">inputs</span><span class="p">:</span> <span class="n">PromptInputs</span><span class="p">,</span>
|
||
<span class="n">sampling_params</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">SamplingParams</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">lora_request</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">LoRARequest</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">prompt_adapter_request</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">PromptAdapterRequest</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">streaming</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
|
||
<span class="n">kv_cache_retention_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">KvCacheRetentionConfig</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">disaggregated_params</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">DisaggregatedParams</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">_postproc_params</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">PostprocParams</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">scheduling_params</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">SchedulingParams</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="p">)</span> <span class="o">-></span> <span class="n">RequestOutput</span><span class="p">:</span>
|
||
<span class="w"> </span><span class="sd">"""Generate output for the given prompt in the asynchronous mode.</span>
|
||
<span class="sd"> Asynchronous generation accepts single prompt only.</span>
|
||
|
||
<span class="sd"> Args:</span>
|
||
<span class="sd"> inputs (tensorrt_llm.inputs.data.PromptInputs): The prompt text or token ids; it must be single prompt.</span>
|
||
<span class="sd"> sampling_params (tensorrt_llm.sampling_params.SamplingParams, optional): The sampling params for the generation. Defaults to None.</span>
|
||
<span class="sd"> A default one will be used if not provided.</span>
|
||
<span class="sd"> lora_request (tensorrt_llm.executor.request.LoRARequest, optional): LoRA request to use for generation, if any. Defaults to None.</span>
|
||
<span class="sd"> prompt_adapter_request (tensorrt_llm.executor.request.PromptAdapterRequest, optional): Prompt Adapter request to use for generation, if any. Defaults to None.</span>
|
||
<span class="sd"> streaming (bool): Whether to use the streaming mode for the generation. Defaults to False.</span>
|
||
<span class="sd"> kv_cache_retention_config (tensorrt_llm.bindings.executor.KvCacheRetentionConfig, optional): Configuration for the request's retention in the KV Cache. Defaults to None.</span>
|
||
<span class="sd"> disaggregated_params (tensorrt_llm.disaggregated_params.DisaggregatedParams, optional): Disaggregated parameters. Defaults to None.</span>
|
||
<span class="sd"> scheduling_params (tensorrt_llm.scheduling_params.SchedulingParams, optional): Scheduling parameters. Defaults to None.</span>
|
||
|
||
<span class="sd"> Returns:</span>
|
||
<span class="sd"> tensorrt_llm.llmapi.RequestOutput: The output data of the completion request to the LLM.</span>
|
||
<span class="sd"> """</span>
|
||
|
||
<span class="c1"># Check if the worker is shutting down</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_executor</span> <span class="ow">is</span> <span class="kc">None</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">_executor</span><span class="o">.</span><span class="n">is_shutdown</span><span class="p">():</span>
|
||
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="s2">"LLM is shutting down"</span><span class="p">)</span>
|
||
|
||
<span class="n">sampling_params</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_prepare_sampling_params</span><span class="p">(</span><span class="n">sampling_params</span><span class="p">)</span>
|
||
|
||
<span class="c1"># With pytorch backend, py_executor has logic to handle max_tokens of 1,</span>
|
||
<span class="c1"># so set to 1 to avoid allocating unnecessary KV cache blocks for single request</span>
|
||
<span class="c1"># TODO: Also support for trt backend</span>
|
||
<span class="n">is_ctx_only</span> <span class="o">=</span> <span class="n">disaggregated_params</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="n">disaggregated_params</span><span class="o">.</span><span class="n">request_type</span> <span class="o">==</span> <span class="s2">"context_only"</span>
|
||
<span class="n">is_gen_only</span> <span class="o">=</span> <span class="n">disaggregated_params</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="n">disaggregated_params</span><span class="o">.</span><span class="n">request_type</span> <span class="o">==</span> <span class="s2">"generation_only"</span>
|
||
<span class="k">if</span> <span class="n">is_ctx_only</span> <span class="ow">and</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">_on_trt_backend</span><span class="p">:</span>
|
||
<span class="n">sampling_params</span><span class="o">.</span><span class="n">max_tokens</span> <span class="o">=</span> <span class="mi">1</span>
|
||
|
||
<span class="n">inputs</span> <span class="o">=</span> <span class="n">prompt_inputs</span><span class="p">(</span><span class="n">inputs</span><span class="p">)</span>
|
||
|
||
<span class="k">if</span> <span class="ow">not</span> <span class="n">inputs</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s2">"prompt"</span><span class="p">)</span> <span class="ow">and</span> <span class="n">inputs</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s2">"prompt_token_ids"</span><span class="p">)</span> <span class="ow">and</span> <span class="p">(</span>
|
||
<span class="n">inputs</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s2">"multi_modal_data"</span><span class="p">)</span>
|
||
<span class="ow">or</span> <span class="n">inputs</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s2">"multi_modal_embeddings"</span><span class="p">))</span> <span class="ow">and</span> <span class="ow">not</span> <span class="nb">isinstance</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">input_processor</span><span class="p">,</span> <span class="n">DefaultInputProcessor</span><span class="p">):</span>
|
||
<span class="c1"># VLMs need to process/tokenize the prompt in their own way</span>
|
||
<span class="n">prompt</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">tokenizer</span><span class="o">.</span><span class="n">decode</span><span class="p">(</span><span class="n">inputs</span><span class="p">[</span><span class="s1">'prompt_token_ids'</span><span class="p">])</span>
|
||
<span class="n">inputs</span> <span class="o">=</span> <span class="n">TextPrompt</span><span class="p">(</span>
|
||
<span class="n">prompt</span><span class="o">=</span><span class="n">prompt</span><span class="p">,</span>
|
||
<span class="n">multi_modal_data</span><span class="o">=</span><span class="n">inputs</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s2">"multi_modal_data"</span><span class="p">),</span>
|
||
<span class="n">mm_processor_kwargs</span><span class="o">=</span><span class="n">inputs</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s2">"mm_processor_kwargs"</span><span class="p">))</span>
|
||
<span class="k">if</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">add_special_tokens</span><span class="p">:</span>
|
||
<span class="n">logger</span><span class="o">.</span><span class="n">debug</span><span class="p">(</span>
|
||
<span class="s2">"Setting add_special_tokens to False because prompt_token_ids were provided to generate. VLMs will re-encode the prompt."</span>
|
||
<span class="p">)</span>
|
||
<span class="n">sampling_params</span><span class="o">.</span><span class="n">add_special_tokens</span> <span class="o">=</span> <span class="kc">False</span>
|
||
|
||
<span class="n">query_token_ids</span> <span class="o">=</span> <span class="kc">None</span>
|
||
<span class="n">multimodal_params</span> <span class="o">=</span> <span class="kc">None</span>
|
||
|
||
<span class="k">if</span> <span class="s2">"prompt_token_ids"</span> <span class="ow">in</span> <span class="n">inputs</span><span class="p">:</span>
|
||
<span class="c1"># TODO: if specify prompt_token_ids, the mm hashing is not supported yet</span>
|
||
<span class="n">prompt_token_ids</span> <span class="o">=</span> <span class="n">inputs</span><span class="p">[</span><span class="s1">'prompt_token_ids'</span><span class="p">]</span>
|
||
<span class="n">prompt</span> <span class="o">=</span> <span class="kc">None</span>
|
||
<span class="n">query_token_ids</span> <span class="o">=</span> <span class="n">inputs</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s2">"query_token_ids"</span><span class="p">,</span> <span class="kc">None</span><span class="p">)</span>
|
||
<span class="k">elif</span> <span class="s2">"prompt"</span> <span class="ow">in</span> <span class="n">inputs</span><span class="p">:</span>
|
||
<span class="k">if</span> <span class="s1">'multi_modal_data'</span> <span class="ow">in</span> <span class="n">inputs</span><span class="p">:</span>
|
||
<span class="c1"># TODO: The current design uses a wrapper for existing input processor (input_processor_with_hash)</span>
|
||
<span class="c1"># to handle/add multimodal hashes, positions, and lengths. Now we only support image modality.</span>
|
||
<span class="c1"># In the future, we should refactor this to:</span>
|
||
<span class="c1"># 1. Extend support for more modalities and models</span>
|
||
<span class="c1"># 2. Decouple input processor into distinct phases (preprocessor (all preprocessing logics), vision model (fuse in model fwd), etc.</span>
|
||
<span class="n">input_processor_with_hash</span> <span class="o">=</span> <span class="n">create_input_processor_with_hash</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">input_processor</span><span class="p">)</span>
|
||
<span class="k">with</span> <span class="n">nvtx_range_debug</span><span class="p">(</span><span class="s2">"input_processor_with_hash"</span><span class="p">):</span>
|
||
<span class="n">prompt_token_ids</span><span class="p">,</span> <span class="n">extra_processed_inputs</span> <span class="o">=</span> <span class="n">input_processor_with_hash</span><span class="p">(</span>
|
||
<span class="n">inputs</span><span class="p">,</span> <span class="n">sampling_params</span><span class="p">)</span>
|
||
<span class="k">elif</span> <span class="s1">'multi_modal_embeddings'</span> <span class="ow">in</span> <span class="n">inputs</span><span class="p">:</span>
|
||
<span class="n">mm_embedding_info</span> <span class="o">=</span> <span class="n">inputs</span><span class="p">[</span><span class="s1">'multi_modal_embeddings'</span><span class="p">]</span>
|
||
<span class="n">prompt_token_ids</span><span class="p">,</span> <span class="n">extra_processed_inputs</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">input_processor</span><span class="o">.</span><span class="n">attach_multimodal_embeddings</span><span class="p">(</span>
|
||
<span class="n">inputs</span><span class="p">,</span> <span class="n">mm_embedding_info</span><span class="p">,</span> <span class="n">sampling_params</span><span class="p">)</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="k">with</span> <span class="n">nvtx_range_debug</span><span class="p">(</span><span class="s2">"input_processor"</span><span class="p">):</span>
|
||
<span class="n">prompt_token_ids</span><span class="p">,</span> <span class="n">extra_processed_inputs</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">input_processor</span><span class="p">(</span>
|
||
<span class="n">inputs</span><span class="p">,</span> <span class="n">sampling_params</span><span class="p">)</span>
|
||
<span class="n">prompt</span> <span class="o">=</span> <span class="n">inputs</span><span class="p">[</span><span class="s1">'prompt'</span><span class="p">]</span>
|
||
<span class="k">if</span> <span class="n">extra_processed_inputs</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="n">query_token_ids</span> <span class="o">=</span> <span class="n">extra_processed_inputs</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s1">'query_token_ids'</span><span class="p">)</span>
|
||
<span class="c1"># Create unified MultimodalParams</span>
|
||
<span class="n">multimodal_params</span> <span class="o">=</span> <span class="n">MultimodalParams</span><span class="p">(</span>
|
||
<span class="n">multimodal_input</span><span class="o">=</span><span class="n">extra_processed_inputs</span><span class="o">.</span><span class="n">get</span><span class="p">(</span>
|
||
<span class="s1">'multimodal_input'</span><span class="p">),</span>
|
||
<span class="n">multimodal_data</span><span class="o">=</span><span class="n">extra_processed_inputs</span><span class="o">.</span><span class="n">get</span><span class="p">(</span>
|
||
<span class="s1">'multimodal_data'</span><span class="p">))</span>
|
||
<span class="c1"># Only pass it if it has content</span>
|
||
<span class="k">if</span> <span class="ow">not</span> <span class="n">multimodal_params</span><span class="o">.</span><span class="n">has_content</span><span class="p">():</span>
|
||
<span class="n">multimodal_params</span> <span class="o">=</span> <span class="kc">None</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="c1"># Convert to shared tensor handle to reduce IPC overhead</span>
|
||
<span class="n">multimodal_params</span><span class="o">.</span><span class="n">to_handle</span><span class="p">(</span><span class="s2">"multimodal_data"</span><span class="p">)</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">TypeError</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"The inputs must be type str or list of int, but got </span><span class="si">{</span><span class="nb">type</span><span class="p">(</span><span class="n">inputs</span><span class="p">)</span><span class="si">}</span><span class="s2">"</span>
|
||
<span class="p">)</span>
|
||
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_check_arguments</span><span class="p">(</span>
|
||
<span class="nb">len</span><span class="p">(</span><span class="n">prompt_token_ids</span><span class="p">),</span>
|
||
<span class="nb">len</span><span class="p">(</span><span class="n">query_token_ids</span><span class="p">)</span> <span class="k">if</span> <span class="n">query_token_ids</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="k">else</span> <span class="mi">0</span><span class="p">,</span>
|
||
<span class="n">sampling_params</span><span class="p">,</span>
|
||
<span class="n">is_gen_only</span><span class="o">=</span><span class="n">is_gen_only</span><span class="p">)</span>
|
||
<span class="k">if</span> <span class="n">_postproc_params</span><span class="p">:</span>
|
||
<span class="n">_postproc_params</span><span class="o">.</span><span class="n">postproc_args</span><span class="o">.</span><span class="n">num_prompt_tokens</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span>
|
||
<span class="n">prompt_token_ids</span><span class="p">)</span>
|
||
<span class="n">result</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_executor</span><span class="o">.</span><span class="n">generate_async</span><span class="p">(</span>
|
||
<span class="n">prompt_token_ids</span><span class="p">,</span>
|
||
<span class="n">query_token_ids</span><span class="o">=</span><span class="n">query_token_ids</span><span class="p">,</span>
|
||
<span class="n">sampling_params</span><span class="o">=</span><span class="n">sampling_params</span><span class="p">,</span>
|
||
<span class="n">lora_request</span><span class="o">=</span><span class="n">lora_request</span><span class="p">,</span>
|
||
<span class="n">prompt_adapter_request</span><span class="o">=</span><span class="n">prompt_adapter_request</span><span class="p">,</span>
|
||
<span class="n">streaming</span><span class="o">=</span><span class="n">streaming</span><span class="p">,</span>
|
||
<span class="n">kv_cache_retention_config</span><span class="o">=</span><span class="n">kv_cache_retention_config</span><span class="p">,</span>
|
||
<span class="n">disaggregated_params</span><span class="o">=</span><span class="n">disaggregated_params</span><span class="p">,</span>
|
||
<span class="n">postproc_params</span><span class="o">=</span><span class="n">_postproc_params</span><span class="p">,</span>
|
||
<span class="n">multimodal_params</span><span class="o">=</span><span class="n">multimodal_params</span><span class="p">,</span>
|
||
<span class="n">scheduling_params</span><span class="o">=</span><span class="n">scheduling_params</span><span class="p">,</span>
|
||
<span class="p">)</span>
|
||
|
||
<span class="k">return</span> <span class="n">RequestOutput</span><span class="o">.</span><span class="n">_from_generation_result</span><span class="p">(</span><span class="n">result</span><span class="p">,</span> <span class="n">prompt</span><span class="p">,</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">tokenizer</span><span class="p">)</span>
|
||
|
||
<span class="nd">@set_api_status</span><span class="p">(</span><span class="s2">"beta"</span><span class="p">)</span>
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">get_stats</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">timeout</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">float</span><span class="p">]</span> <span class="o">=</span> <span class="mi">2</span><span class="p">)</span> <span class="o">-></span> <span class="n">List</span><span class="p">[</span><span class="nb">dict</span><span class="p">]:</span>
|
||
<span class="w"> </span><span class="sd">'''Get iteration statistics from the runtime.</span>
|
||
<span class="sd"> To collect statistics, call this function after prompts have been submitted with LLM().generate().</span>
|
||
|
||
<span class="sd"> Args:</span>
|
||
<span class="sd"> timeout (float, optional): Max wait time in seconds when retrieving stats from queue. Defaults to 2.</span>
|
||
|
||
<span class="sd"> Returns:</span>
|
||
<span class="sd"> List[dict]: A list of runtime stats as dict.</span>
|
||
<span class="sd"> e.g., ['{"cpuMemUsage": ..., "iter": 0, ...}', '{"cpuMemUsage": ..., "iter": 1, ...}']</span>
|
||
<span class="sd"> '''</span>
|
||
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_executor</span><span class="o">.</span><span class="n">get_stats</span><span class="p">(</span><span class="n">timeout</span><span class="o">=</span><span class="n">timeout</span><span class="p">)</span>
|
||
|
||
<span class="nd">@set_api_status</span><span class="p">(</span><span class="s2">"beta"</span><span class="p">)</span>
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">get_stats_async</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">timeout</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">float</span><span class="p">]</span> <span class="o">=</span> <span class="mi">2</span><span class="p">)</span> <span class="o">-></span> <span class="n">IterationResult</span><span class="p">:</span>
|
||
<span class="w"> </span><span class="sd">'''Get iteration statistics from the runtime.</span>
|
||
<span class="sd"> To collect statistics, you can call this function in an async coroutine or the /metrics endpoint (if you're using trtllm-serve)</span>
|
||
<span class="sd"> after prompts have been submitted.</span>
|
||
|
||
<span class="sd"> Args:</span>
|
||
<span class="sd"> timeout (float, optional): Max wait time in seconds when retrieving stats from queue. Defaults to 2.</span>
|
||
|
||
<span class="sd"> Returns:</span>
|
||
<span class="sd"> tensorrt_llm.executor.result.IterationResult: An async iterable object containing runtime stats.</span>
|
||
<span class="sd"> '''</span>
|
||
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_executor</span><span class="o">.</span><span class="n">aget_stats</span><span class="p">(</span><span class="n">timeout</span><span class="o">=</span><span class="n">timeout</span><span class="p">)</span>
|
||
|
||
<span class="nd">@set_api_status</span><span class="p">(</span><span class="s2">"beta"</span><span class="p">)</span>
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">get_kv_cache_events</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">timeout</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">float</span><span class="p">]</span> <span class="o">=</span> <span class="mi">2</span><span class="p">)</span> <span class="o">-></span> <span class="n">List</span><span class="p">[</span><span class="nb">dict</span><span class="p">]:</span>
|
||
<span class="w"> </span><span class="sd">'''Get iteration KV events from the runtime.</span>
|
||
|
||
<span class="sd"> KV events are used to track changes and operations within the KV Cache. Types of events:</span>
|
||
<span class="sd"> - KVCacheCreatedData: Indicates the creation of cache blocks.</span>
|
||
<span class="sd"> - KVCacheStoredData: Represents a sequence of stored blocks.</span>
|
||
<span class="sd"> - KVCacheRemovedData: Contains the hashes of blocks that are being removed from the cache.</span>
|
||
<span class="sd"> - KVCacheUpdatedData: Captures updates to existing cache blocks.</span>
|
||
|
||
<span class="sd"> To enable KV events:</span>
|
||
<span class="sd"> - set `event_buffer_max_size` to a positive integer in the `KvCacheConfig`.</span>
|
||
<span class="sd"> - set `enable_block_reuse` to True in the `KvCacheConfig`.</span>
|
||
|
||
<span class="sd"> Args:</span>
|
||
<span class="sd"> timeout (float, optional): Max wait time in seconds when retrieving events from queue. Defaults to 2.</span>
|
||
|
||
<span class="sd"> Returns:</span>
|
||
<span class="sd"> List[dict]: A list of runtime events as dict.</span>
|
||
<span class="sd"> '''</span>
|
||
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_executor</span><span class="o">.</span><span class="n">get_kv_events</span><span class="p">(</span><span class="n">timeout</span><span class="o">=</span><span class="n">timeout</span><span class="p">)</span>
|
||
|
||
<span class="nd">@set_api_status</span><span class="p">(</span><span class="s2">"beta"</span><span class="p">)</span>
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">get_kv_cache_events_async</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span>
|
||
<span class="n">timeout</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">float</span><span class="p">]</span> <span class="o">=</span> <span class="mi">2</span>
|
||
<span class="p">)</span> <span class="o">-></span> <span class="n">IterationResult</span><span class="p">:</span>
|
||
<span class="w"> </span><span class="sd">'''Get iteration KV events from the runtime.</span>
|
||
|
||
<span class="sd"> KV events are used to track changes and operations within the KV Cache. Types of events:</span>
|
||
<span class="sd"> - KVCacheCreatedData: Indicates the creation of cache blocks.</span>
|
||
<span class="sd"> - KVCacheStoredData: Represents a sequence of stored blocks.</span>
|
||
<span class="sd"> - KVCacheRemovedData: Contains the hashes of blocks that are being removed from the cache.</span>
|
||
<span class="sd"> - KVCacheUpdatedData: Captures updates to existing cache blocks.</span>
|
||
|
||
<span class="sd"> To enable KV events:</span>
|
||
<span class="sd"> - set `event_buffer_max_size` to a positive integer in the `KvCacheConfig`.</span>
|
||
<span class="sd"> - set `enable_block_reuse` to True in the `KvCacheConfig`.</span>
|
||
|
||
<span class="sd"> Args:</span>
|
||
<span class="sd"> timeout (float, optional): Max wait time in seconds when retrieving events from queue. . Defaults to 2.</span>
|
||
|
||
<span class="sd"> Returns:</span>
|
||
<span class="sd"> tensorrt_llm.executor.result.IterationResult: An async iterable object containing runtime events.</span>
|
||
<span class="sd"> '''</span>
|
||
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_executor</span><span class="o">.</span><span class="n">aget_kv_events</span><span class="p">(</span><span class="n">timeout</span><span class="o">=</span><span class="n">timeout</span><span class="p">)</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">_prepare_sampling_params</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="p">,</span>
|
||
<span class="n">sampling_params</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">SamplingParams</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">)</span> <span class="o">-></span> <span class="n">SamplingParams</span><span class="p">:</span>
|
||
<span class="k">if</span> <span class="n">sampling_params</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">tokenizer</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
|
||
<span class="s2">"tokenizer is required to initialize a default sampling_params, or you can explicitly specify a sampling_params"</span>
|
||
<span class="p">)</span>
|
||
<span class="n">sampling_params</span> <span class="o">=</span> <span class="n">SamplingParams</span><span class="p">(</span><span class="n">end_id</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">tokenizer</span><span class="o">.</span><span class="n">eos_token_id</span><span class="p">,</span>
|
||
<span class="n">pad_id</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">tokenizer</span><span class="o">.</span><span class="n">pad_token_id</span><span class="p">)</span>
|
||
<span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">sampling_params</span><span class="p">,</span> <span class="n">SamplingParams</span><span class="p">):</span>
|
||
<span class="k">if</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">end_id</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">tokenizer</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
|
||
<span class="s2">"tokenizer is required to reset end_id if it is None, or you can explicitly specify the end_id for sampling_params"</span>
|
||
<span class="p">)</span>
|
||
<span class="n">sampling_params</span><span class="o">.</span><span class="n">_setup</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">tokenizer</span><span class="p">)</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">TypeError</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"The sampling_params must be type SamplingParams or None, but got </span><span class="si">{</span><span class="nb">type</span><span class="p">(</span><span class="n">sampling_params</span><span class="p">)</span><span class="si">}</span><span class="s2">"</span>
|
||
<span class="p">)</span>
|
||
|
||
<span class="c1"># auto enabled context and/or generation logits flags, as they are required by logprob computation for TRT backend.</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">backend</span> <span class="ow">not</span> <span class="ow">in</span> <span class="p">[</span><span class="s2">"pytorch"</span><span class="p">,</span> <span class="s2">"_autodeploy"</span><span class="p">]:</span>
|
||
<span class="k">if</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">prompt_logprobs</span> <span class="ow">and</span> <span class="ow">not</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">return_context_logits</span><span class="p">:</span>
|
||
<span class="n">sampling_params</span><span class="o">.</span><span class="n">return_context_logits</span> <span class="o">=</span> <span class="kc">True</span>
|
||
<span class="n">sampling_params</span><span class="o">.</span><span class="n">_context_logits_auto_enabled</span> <span class="o">=</span> <span class="kc">True</span>
|
||
<span class="k">if</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">logprobs</span> <span class="ow">and</span> <span class="ow">not</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">return_generation_logits</span><span class="p">:</span>
|
||
<span class="n">sampling_params</span><span class="o">.</span><span class="n">return_generation_logits</span> <span class="o">=</span> <span class="kc">True</span>
|
||
<span class="n">sampling_params</span><span class="o">.</span><span class="n">_generation_logits_auto_enabled</span> <span class="o">=</span> <span class="kc">True</span>
|
||
|
||
<span class="k">if</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">_stream_interval</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="n">sampling_params</span><span class="o">.</span><span class="n">_stream_interval</span> <span class="o">=</span> <span class="nb">getattr</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="p">,</span>
|
||
<span class="s2">"stream_interval"</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>
|
||
<span class="n">sampling_params</span><span class="o">.</span><span class="n">return_perf_metrics</span> <span class="o">=</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">return_perf_metrics</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">return_perf_metrics</span>
|
||
<span class="k">return</span> <span class="n">sampling_params</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">_check_arguments</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">prompt_len</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">query_len</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span>
|
||
<span class="n">sampling_params</span><span class="p">:</span> <span class="n">SamplingParams</span><span class="p">,</span>
|
||
<span class="n">is_gen_only</span><span class="p">:</span> <span class="nb">bool</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
|
||
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">backend</span> <span class="ow">in</span> <span class="p">[</span><span class="s2">"pytorch"</span><span class="p">,</span> <span class="s2">"_autodeploy"</span><span class="p">]:</span>
|
||
<span class="c1"># TODO: remove these checks after PyTorch backend</span>
|
||
<span class="c1"># fully support TopK prompt and generation logprobs.</span>
|
||
<span class="k">if</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">prompt_logprobs</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"`prompt_logprobs` in sampling_params is not supported in the PyTorch backend yet. Received `prompt_logprobs=</span><span class="si">{</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">prompt_logprobs</span><span class="si">}</span><span class="s2">`. Please unset this field."</span>
|
||
<span class="p">)</span>
|
||
<span class="k">if</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">logprobs</span> <span class="ow">and</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">logprobs</span> <span class="o">></span> <span class="mi">1</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"PyTorch backend currently only supports `logprobs=1`. Received `logprobs=</span><span class="si">{</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">logprobs</span><span class="si">}</span><span class="s2">` (Top</span><span class="si">{</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">logprobs</span><span class="si">}</span><span class="s2"> logprobs). Please set `logprobs=1` in `sampling_params` instead."</span>
|
||
<span class="p">)</span>
|
||
<span class="c1"># Check prompt length and query length against max_num_tokens to filter illegal requests.</span>
|
||
<span class="c1"># Skip check for gen-only requests</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">backend</span> <span class="o">==</span> <span class="s2">"pytorch"</span> <span class="ow">and</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">enable_chunked_prefill</span> <span class="ow">and</span> <span class="ow">not</span> <span class="n">is_gen_only</span><span class="p">:</span>
|
||
<span class="n">max_num_tokens</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">max_num_tokens</span>
|
||
<span class="k">if</span> <span class="n">max_num_tokens</span> <span class="ow">and</span> <span class="n">prompt_len</span> <span class="o">/</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">cp_size</span> <span class="o">+</span> <span class="n">query_len</span> <span class="o">></span> <span class="n">max_num_tokens</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"The sum of prompt length (</span><span class="si">{</span><span class="n">prompt_len</span><span class="o">/</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">cp_size</span><span class="si">}</span><span class="s2">), query length (</span><span class="si">{</span><span class="n">query_len</span><span class="si">}</span><span class="s2">) should not exceed "</span>
|
||
<span class="sa">f</span><span class="s2">"max_num_tokens (</span><span class="si">{</span><span class="n">max_num_tokens</span><span class="si">}</span><span class="s2">)"</span><span class="p">)</span>
|
||
<span class="k">return</span>
|
||
|
||
<span class="n">build_config</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">build_config</span>
|
||
|
||
<span class="n">built_enging_cfg_file</span> <span class="o">=</span> <span class="n">Path</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">model</span><span class="p">)</span> <span class="o">/</span> <span class="s1">'config.json'</span>
|
||
<span class="k">with</span> <span class="nb">open</span><span class="p">(</span><span class="n">built_enging_cfg_file</span><span class="p">)</span> <span class="k">as</span> <span class="n">f</span><span class="p">:</span>
|
||
<span class="n">built_enging_cfg</span> <span class="o">=</span> <span class="n">json</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="n">f</span><span class="p">)</span>
|
||
<span class="n">max_seq_len</span> <span class="o">=</span> <span class="n">built_enging_cfg</span><span class="p">[</span><span class="s1">'build_config'</span><span class="p">][</span>
|
||
<span class="s1">'max_seq_len'</span><span class="p">]</span> <span class="k">if</span> <span class="s1">'build_config'</span> <span class="ow">in</span> <span class="n">built_enging_cfg</span> <span class="k">else</span> <span class="n">build_config</span><span class="o">.</span><span class="n">max_seq_len</span>
|
||
<span class="c1"># TODO: Remove this check and left the request verification to cpp runtime</span>
|
||
|
||
<span class="k">if</span> <span class="p">(</span><span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">enable_chunked_prefill</span><span class="p">)</span> <span class="ow">and</span> <span class="p">(</span>
|
||
<span class="n">prompt_len</span> <span class="o">/</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">cp_size</span> <span class="o">+</span> <span class="n">query_len</span> <span class="o">+</span>
|
||
<span class="p">(</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">max_tokens</span> <span class="ow">or</span> <span class="mi">0</span><span class="p">)</span> <span class="o">></span> <span class="n">max_seq_len</span><span class="p">):</span>
|
||
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"The sum of prompt length (</span><span class="si">{</span><span class="n">prompt_len</span><span class="o">/</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">cp_size</span><span class="si">}</span><span class="s2">) and query length (</span><span class="si">{</span><span class="n">query_len</span><span class="si">}</span><span class="s2">) max_tokens (</span><span class="si">{</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">max_tokens</span><span class="si">}</span><span class="s2">) should not exceed "</span>
|
||
<span class="sa">f</span><span class="s2">"max_seq_len (</span><span class="si">{</span><span class="n">max_seq_len</span><span class="si">}</span><span class="s2">)"</span><span class="p">)</span>
|
||
|
||
<span class="k">if</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">use_beam_search</span> <span class="ow">and</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">best_of</span> <span class="o">></span> <span class="n">build_config</span><span class="o">.</span><span class="n">max_beam_width</span><span class="p">:</span>
|
||
<span class="k">if</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">n</span> <span class="o">==</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">best_of</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"sampling_params.n (</span><span class="si">{</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">n</span><span class="si">}</span><span class="s2">) cannot exceed max_beam_width (</span><span class="si">{</span><span class="n">build_config</span><span class="o">.</span><span class="n">max_beam_width</span><span class="si">}</span><span class="s2">) when use_beam_search is True"</span>
|
||
<span class="p">)</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"sampling_params.best_of (</span><span class="si">{</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">best_of</span><span class="si">}</span><span class="s2">) cannot exceed max_beam_width (</span><span class="si">{</span><span class="n">build_config</span><span class="o">.</span><span class="n">max_beam_width</span><span class="si">}</span><span class="s2">) when use_beam_search is True"</span>
|
||
<span class="p">)</span>
|
||
|
||
<span class="n">max_batch_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">max_batch_size</span>
|
||
<span class="k">if</span> <span class="n">max_batch_size</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="n">max_batch_size</span> <span class="o">=</span> <span class="n">build_config</span><span class="o">.</span><span class="n">max_batch_size</span>
|
||
<span class="k">if</span> <span class="ow">not</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">use_beam_search</span> <span class="ow">and</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">best_of</span> <span class="o">></span> <span class="n">max_batch_size</span><span class="p">:</span>
|
||
<span class="k">if</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">n</span> <span class="o">==</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">best_of</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"sampling_params.n (</span><span class="si">{</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">n</span><span class="si">}</span><span class="s2">) cannot exceed max_batch_size (</span><span class="si">{</span><span class="n">max_batch_size</span><span class="si">}</span><span class="s2">) when use_beam_search is False"</span>
|
||
<span class="p">)</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"sampling_params.best_of (</span><span class="si">{</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">best_of</span><span class="si">}</span><span class="s2">) cannot exceed max_batch_size (</span><span class="si">{</span><span class="n">max_batch_size</span><span class="si">}</span><span class="s2">) when use_beam_search is False"</span>
|
||
<span class="p">)</span>
|
||
|
||
<span class="k">if</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">prompt_logprobs</span> <span class="ow">and</span> <span class="ow">not</span> <span class="n">build_config</span><span class="o">.</span><span class="n">gather_context_logits</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"`sampling_params's prompt_logprobs=</span><span class="si">{</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">prompt_logprobs</span><span class="si">}</span><span class="s2">` requires `gather_context_logits=True` "</span>
|
||
<span class="sa">f</span><span class="s2">"in the `BuildConfig` when constructing the LLM. "</span>
|
||
<span class="sa">f</span><span class="s2">"Example: LLM(..., build_config=BuildConfig(gather_context_logits=True))."</span>
|
||
<span class="p">)</span>
|
||
|
||
<span class="k">if</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">logprobs</span> <span class="ow">and</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">gather_generation_logits</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"`sampling_params.logprobs=</span><span class="si">{</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">logprobs</span><span class="si">}</span><span class="s2">` requires `gather_generation_logits=True` "</span>
|
||
<span class="sa">f</span><span class="s2">"to be passed explicitly to the `LLM()` constructor."</span><span class="p">)</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">_build_model</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
||
<span class="n">model_loader</span> <span class="o">=</span> <span class="n">CachedModelLoader</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="p">,</span>
|
||
<span class="n">mpi_session</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span><span class="p">,</span>
|
||
<span class="n">workspace</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_workspace</span><span class="p">,</span>
|
||
<span class="n">llm_build_stats</span><span class="o">=</span><span class="n">weakref</span><span class="o">.</span><span class="n">proxy</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">llm_build_stats</span><span class="p">))</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_engine_dir</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">_hf_model_dir</span> <span class="o">=</span> <span class="n">model_loader</span><span class="p">()</span>
|
||
|
||
<span class="nd">@property</span>
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">_on_trt_backend</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
|
||
<span class="k">return</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="p">,</span> <span class="n">TrtLlmArgs</span><span class="p">)</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">_try_load_tokenizer</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="n">Optional</span><span class="p">[</span><span class="n">TokenizerBase</span><span class="p">]:</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">skip_tokenizer_init</span><span class="p">:</span>
|
||
<span class="k">return</span> <span class="kc">None</span>
|
||
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">tokenizer</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">tokenizer</span><span class="p">,</span> <span class="n">TokenizerBase</span><span class="p">)</span>
|
||
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">tokenizer</span>
|
||
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">runtime_context</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">runtime_context</span><span class="o">.</span><span class="n">tokenizer</span>
|
||
|
||
<span class="c1"># TODO smor- need to refine what is the desired behavior if lora is enabled</span>
|
||
<span class="c1"># in terms of the tokenizer initialization process</span>
|
||
<span class="k">if</span> <span class="nb">hasattr</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="p">,</span> <span class="s2">"backend"</span><span class="p">)</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">backend</span> <span class="ow">in</span> <span class="p">[</span>
|
||
<span class="s2">"pytorch"</span><span class="p">,</span> <span class="s2">"_autodeploy"</span>
|
||
<span class="p">]</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">lora_config</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="n">num_lora_dirs</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">lora_config</span><span class="o">.</span><span class="n">lora_dir</span><span class="p">)</span>
|
||
<span class="k">if</span> <span class="n">num_lora_dirs</span> <span class="o">==</span> <span class="mi">1</span><span class="p">:</span>
|
||
<span class="n">tokenizer_path</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">lora_config</span><span class="o">.</span><span class="n">lora_dir</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
|
||
<span class="k">try</span><span class="p">:</span>
|
||
<span class="n">tokenizer</span> <span class="o">=</span> <span class="n">ModelLoader</span><span class="o">.</span><span class="n">load_hf_tokenizer</span><span class="p">(</span>
|
||
<span class="n">tokenizer_path</span><span class="p">,</span>
|
||
<span class="n">trust_remote_code</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">trust_remote_code</span><span class="p">,</span>
|
||
<span class="n">use_fast</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">tokenizer_mode</span> <span class="o">!=</span> <span class="s1">'slow'</span><span class="p">)</span>
|
||
<span class="k">if</span> <span class="n">tokenizer</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="n">tokenizer_path</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">model</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="k">return</span> <span class="n">tokenizer</span>
|
||
<span class="k">except</span> <span class="ne">Exception</span><span class="p">:</span>
|
||
<span class="n">tokenizer_path</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">model</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="n">tokenizer_path</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">model</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="n">tokenizer_path</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">model</span>
|
||
<span class="k">return</span> <span class="n">ModelLoader</span><span class="o">.</span><span class="n">load_hf_tokenizer</span><span class="p">(</span>
|
||
<span class="n">tokenizer_path</span><span class="p">,</span>
|
||
<span class="n">trust_remote_code</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">trust_remote_code</span><span class="p">,</span>
|
||
<span class="n">use_fast</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">tokenizer_mode</span> <span class="o">!=</span> <span class="s1">'slow'</span><span class="p">)</span>
|
||
|
||
<span class="nd">@property</span>
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">tokenizer</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="n">Optional</span><span class="p">[</span><span class="n">TokenizerBase</span><span class="p">]:</span>
|
||
<span class="k">if</span> <span class="nb">hasattr</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="s2">"input_processor"</span><span class="p">):</span>
|
||
<span class="k">if</span> <span class="nb">hasattr</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">input_processor</span><span class="p">,</span> <span class="s2">"tokenizer"</span><span class="p">):</span>
|
||
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">input_processor</span><span class="o">.</span><span class="n">tokenizer</span>
|
||
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_tokenizer</span>
|
||
|
||
<span class="nd">@tokenizer</span><span class="o">.</span><span class="n">setter</span>
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">tokenizer</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">tokenizer</span><span class="p">:</span> <span class="n">TokenizerBase</span><span class="p">):</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_tokenizer</span> <span class="o">=</span> <span class="n">tokenizer</span>
|
||
|
||
<span class="nd">@set_api_status</span><span class="p">(</span><span class="s2">"beta"</span><span class="p">)</span>
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">shutdown</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="k">if</span> <span class="nb">hasattr</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="s2">"_executor"</span><span class="p">)</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">_executor</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor</span> <span class="o">=</span> <span class="kc">None</span>
|
||
|
||
<span class="k">if</span> <span class="nb">hasattr</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="s1">'mpi_session'</span><span class="p">)</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span> <span class="o">=</span> <span class="kc">None</span>
|
||
|
||
<span class="nd">@staticmethod</span>
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">_shutdown_wrapper</span><span class="p">(</span><span class="n">self_ref</span><span class="p">):</span>
|
||
<span class="c1"># Retrieve the instance if it still exists</span>
|
||
<span class="n">instance</span> <span class="o">=</span> <span class="n">self_ref</span><span class="p">()</span>
|
||
<span class="k">if</span> <span class="n">instance</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="n">instance</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="fm">__enter__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
||
<span class="k">return</span> <span class="bp">self</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="fm">__exit__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">exc_type</span><span class="p">,</span> <span class="n">exc_value</span><span class="p">,</span> <span class="n">traceback</span><span class="p">)</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
|
||
<span class="k">del</span> <span class="n">exc_value</span><span class="p">,</span> <span class="n">traceback</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
|
||
<span class="k">return</span> <span class="kc">False</span> <span class="c1"># propagate exceptions</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">__getstate__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
||
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="s2">"LLM object can not be pickled."</span><span class="p">)</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="fm">__del__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
|
||
|
||
|
||
<span class="nd">@append_docstring</span><span class="p">(</span><span class="n">TRT_LLM_DOCSTRING</span><span class="p">)</span>
|
||
<span class="k">class</span><span class="w"> </span><span class="nc">_TrtLLM</span><span class="p">(</span><span class="n">BaseLLM</span><span class="p">):</span>
|
||
<span class="w"> </span><span class="sd">"""LLM class is the main class for running a LLM model using TensorRT-LLM backend.</span>
|
||
|
||
<span class="sd"> Parameters:</span>
|
||
<span class="sd">"""</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span>
|
||
<span class="n">model</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Path</span><span class="p">],</span>
|
||
<span class="n">tokenizer</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Path</span><span class="p">,</span> <span class="n">TokenizerBase</span><span class="p">,</span>
|
||
<span class="n">PreTrainedTokenizerBase</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">tokenizer_mode</span><span class="p">:</span> <span class="n">Literal</span><span class="p">[</span><span class="s1">'auto'</span><span class="p">,</span> <span class="s1">'slow'</span><span class="p">]</span> <span class="o">=</span> <span class="s1">'auto'</span><span class="p">,</span>
|
||
<span class="n">skip_tokenizer_init</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
|
||
<span class="n">trust_remote_code</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
|
||
<span class="n">tensor_parallel_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">1</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="s2">"auto"</span><span class="p">,</span>
|
||
<span class="n">revision</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">tokenizer_revision</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="o">**</span><span class="n">kwargs</span><span class="p">:</span> <span class="n">Any</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="c1"># TODO: deprecate backend in LLM kwargs</span>
|
||
|
||
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">tokenizer</span><span class="p">,</span> <span class="n">tokenizer_mode</span><span class="p">,</span> <span class="n">skip_tokenizer_init</span><span class="p">,</span>
|
||
<span class="n">trust_remote_code</span><span class="p">,</span> <span class="n">tensor_parallel_size</span><span class="p">,</span> <span class="n">dtype</span><span class="p">,</span>
|
||
<span class="n">revision</span><span class="p">,</span> <span class="n">tokenizer_revision</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">)</span>
|
||
|
||
<span class="nd">@property</span>
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">workspace</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="n">Path</span><span class="p">:</span>
|
||
<span class="k">return</span> <span class="n">Path</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_workspace</span><span class="o">.</span><span class="n">name</span><span class="p">)</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_on_trt_backend</span> <span class="k">else</span> <span class="kc">None</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">save</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">engine_dir</span><span class="p">:</span> <span class="nb">str</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="w"> </span><span class="sd">"""Save the built engine to the given path.</span>
|
||
|
||
<span class="sd"> Args:</span>
|
||
<span class="sd"> engine_dir (str): The path to save the engine.</span>
|
||
<span class="sd"> """</span>
|
||
<span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="sa">f</span><span class="s2">"Save model to </span><span class="si">{</span><span class="n">engine_dir</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_engine_dir</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="s2">"The engine is not built yet."</span><span class="p">)</span>
|
||
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_engine_dir</span><span class="o">.</span><span class="n">absolute</span><span class="p">()</span> <span class="o">==</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">abspath</span><span class="p">(</span><span class="n">engine_dir</span><span class="p">):</span>
|
||
<span class="k">return</span>
|
||
|
||
<span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span> <span class="ow">or</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span><span class="o">.</span><span class="n">is_comm_session</span><span class="p">():</span>
|
||
<span class="n">shutil</span><span class="o">.</span><span class="n">copytree</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_engine_dir</span><span class="p">,</span> <span class="n">engine_dir</span><span class="p">,</span> <span class="n">dirs_exist_ok</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="c1"># NFS is fragile, so we copy files one by one</span>
|
||
<span class="n">target_engine_dir</span> <span class="o">=</span> <span class="n">Path</span><span class="p">(</span><span class="n">engine_dir</span><span class="p">)</span>
|
||
<span class="n">target_engine_dir</span><span class="o">.</span><span class="n">mkdir</span><span class="p">(</span><span class="n">parents</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">exist_ok</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
|
||
<span class="c1"># copy files one by one</span>
|
||
<span class="k">for</span> <span class="n">file</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">_engine_dir</span><span class="o">.</span><span class="n">iterdir</span><span class="p">():</span>
|
||
<span class="n">print_colored_debug</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"Copying </span><span class="si">{</span><span class="n">file</span><span class="si">}</span><span class="s2"> to </span><span class="si">{</span><span class="n">target_engine_dir</span><span class="w"> </span><span class="o">/</span><span class="w"> </span><span class="n">file</span><span class="o">.</span><span class="n">name</span><span class="si">}</span><span class="se">\n</span><span class="s2">"</span><span class="p">)</span>
|
||
<span class="n">shutil</span><span class="o">.</span><span class="n">copy</span><span class="p">(</span><span class="n">file</span><span class="p">,</span> <span class="n">target_engine_dir</span> <span class="o">/</span> <span class="n">file</span><span class="o">.</span><span class="n">name</span><span class="p">)</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">_build_model</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
||
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="n">_build_model</span><span class="p">()</span>
|
||
<span class="c1"># update the model_dir to a local dir for the runtime, such as tokenizer loading.</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_engine_dir</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">model</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_engine_dir</span>
|
||
|
||
<span class="c1"># Tokenizer loading should be after calling model_loader(), since model_loader() may download the model from HF hub.</span>
|
||
<span class="c1"># It should also be before bindings ExecutorConfig, which may depend on tokenizer info.</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_tokenizer</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_try_load_tokenizer</span><span class="p">()</span>
|
||
|
||
<span class="c1"># Multimodal special handling:</span>
|
||
<span class="c1"># 1. Default load_tokenizer may fail because MM has different tokenizer configuration. Hence we initialize it inside input processor</span>
|
||
<span class="c1"># 2. May need to modify model weights for MM (e.g., resize vocab embedding). We must do such operation via input processor's __init__</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">input_processor</span> <span class="o">=</span> <span class="n">create_input_processor</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_hf_model_dir</span><span class="p">,</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">tokenizer</span><span class="p">)</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_tokenizer</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">input_processor</span><span class="o">.</span><span class="n">tokenizer</span>
|
||
|
||
<span class="n">max_batch_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">max_batch_size</span>
|
||
<span class="n">max_num_tokens</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">max_num_tokens</span>
|
||
<span class="n">max_seq_len</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">max_seq_len</span>
|
||
|
||
<span class="n">build_config</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">build_config</span>
|
||
|
||
<span class="n">max_batch_size</span> <span class="o">=</span> <span class="n">max_batch_size</span> <span class="ow">or</span> <span class="n">build_config</span><span class="o">.</span><span class="n">max_batch_size</span>
|
||
<span class="n">max_num_tokens</span> <span class="o">=</span> <span class="n">max_num_tokens</span> <span class="ow">or</span> <span class="n">build_config</span><span class="o">.</span><span class="n">max_num_tokens</span>
|
||
<span class="n">max_seq_len</span> <span class="o">=</span> <span class="n">max_seq_len</span> <span class="ow">or</span> <span class="n">build_config</span><span class="o">.</span><span class="n">max_seq_len</span>
|
||
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span> <span class="o">=</span> <span class="n">tllm</span><span class="o">.</span><span class="n">ExecutorConfig</span><span class="p">(</span>
|
||
<span class="n">max_beam_width</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">max_beam_width</span><span class="p">,</span>
|
||
<span class="n">scheduler_config</span><span class="o">=</span><span class="n">PybindMirror</span><span class="o">.</span><span class="n">maybe_to_pybind</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">scheduler_config</span><span class="p">),</span>
|
||
<span class="n">batching_type</span><span class="o">=</span><span class="n">PybindMirror</span><span class="o">.</span><span class="n">maybe_to_pybind</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">batching_type</span><span class="p">)</span>
|
||
<span class="ow">or</span> <span class="n">tllm</span><span class="o">.</span><span class="n">BatchingType</span><span class="o">.</span><span class="n">INFLIGHT</span><span class="p">,</span>
|
||
<span class="n">max_batch_size</span><span class="o">=</span><span class="n">max_batch_size</span><span class="p">,</span>
|
||
<span class="n">max_num_tokens</span><span class="o">=</span><span class="n">max_num_tokens</span><span class="p">,</span>
|
||
<span class="n">gather_generation_logits</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">gather_generation_logits</span><span class="p">,</span>
|
||
<span class="n">fail_fast_on_attention_window_too_large</span><span class="o">=</span><span class="nb">getattr</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="p">,</span> <span class="s1">'fail_fast_on_attention_window_too_large'</span><span class="p">,</span> <span class="kc">False</span><span class="p">))</span>
|
||
|
||
<span class="c1"># also set executor_config.max_seq_len in TRT workflow, to deduce default max_tokens</span>
|
||
<span class="k">if</span> <span class="n">max_seq_len</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span><span class="o">.</span><span class="n">max_seq_len</span> <span class="o">=</span> <span class="n">max_seq_len</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="n">engine_config</span> <span class="o">=</span> <span class="n">EngineConfig</span><span class="o">.</span><span class="n">from_json_file</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_engine_dir</span> <span class="o">/</span>
|
||
<span class="s2">"config.json"</span><span class="p">)</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span><span class="o">.</span><span class="n">max_seq_len</span> <span class="o">=</span> <span class="n">engine_config</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">max_seq_len</span>
|
||
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">kv_cache_config</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span><span class="o">.</span><span class="n">kv_cache_config</span> <span class="o">=</span> <span class="n">PybindMirror</span><span class="o">.</span><span class="n">maybe_to_pybind</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">kv_cache_config</span><span class="p">)</span>
|
||
<span class="k">if</span> <span class="n">os</span><span class="o">.</span><span class="n">getenv</span><span class="p">(</span><span class="s2">"FORCE_DETERMINISTIC"</span><span class="p">,</span> <span class="s2">"0"</span><span class="p">)</span> <span class="o">==</span> <span class="s2">"1"</span><span class="p">:</span>
|
||
<span class="c1"># Disable KV cache reuse for deterministic mode</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span><span class="o">.</span><span class="n">kv_cache_config</span><span class="o">.</span><span class="n">enable_block_reuse</span> <span class="o">=</span> <span class="kc">False</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span><span class="o">.</span><span class="n">kv_cache_config</span><span class="o">.</span><span class="n">enable_partial_reuse</span> <span class="o">=</span> <span class="kc">False</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">peft_cache_config</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span><span class="o">.</span><span class="n">peft_cache_config</span> <span class="o">=</span> <span class="n">PybindMirror</span><span class="o">.</span><span class="n">maybe_to_pybind</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">peft_cache_config</span><span class="p">)</span>
|
||
|
||
<span class="n">lora_config</span> <span class="o">=</span> <span class="kc">None</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">plugin_config</span><span class="o">.</span><span class="n">lora_plugin</span><span class="p">:</span>
|
||
<span class="n">engine_config</span> <span class="o">=</span> <span class="n">EngineConfig</span><span class="o">.</span><span class="n">from_json_file</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_engine_dir</span> <span class="o">/</span>
|
||
<span class="s2">"config.json"</span><span class="p">)</span>
|
||
<span class="n">lora_config</span> <span class="o">=</span> <span class="n">engine_config</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">lora_config</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">lora_config</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span>
|
||
<span class="s2">"Overriding lora_config from engine with lora_config from LLM args"</span>
|
||
<span class="p">)</span>
|
||
<span class="n">lora_config</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">lora_config</span>
|
||
|
||
<span class="n">max_lora_rank</span> <span class="o">=</span> <span class="n">lora_config</span><span class="o">.</span><span class="n">max_lora_rank</span>
|
||
<span class="n">num_lora_modules</span> <span class="o">=</span> <span class="n">engine_config</span><span class="o">.</span><span class="n">pretrained_config</span><span class="o">.</span><span class="n">num_hidden_layers</span> <span class="o">*</span> \
|
||
<span class="nb">len</span><span class="p">(</span><span class="n">lora_config</span><span class="o">.</span><span class="n">lora_target_modules</span> <span class="o">+</span> <span class="n">lora_config</span><span class="o">.</span><span class="n">missing_qkv_modules</span><span class="p">)</span>
|
||
|
||
<span class="n">peft_cache_config_model</span> <span class="o">=</span> <span class="n">PeftCacheConfig</span><span class="o">.</span><span class="n">from_pybind</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span><span class="o">.</span><span class="n">peft_cache_config</span>
|
||
<span class="p">)</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span><span class="o">.</span><span class="n">peft_cache_config</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="k">else</span> <span class="n">PeftCacheConfig</span><span class="p">(</span>
|
||
<span class="p">)</span>
|
||
<span class="k">if</span> <span class="n">lora_config</span><span class="o">.</span><span class="n">max_loras</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="n">peft_cache_config_model</span><span class="o">.</span><span class="n">num_device_module_layer</span> <span class="o">=</span> \
|
||
<span class="n">max_lora_rank</span> <span class="o">*</span> <span class="n">num_lora_modules</span> <span class="o">*</span> <span class="n">lora_config</span><span class="o">.</span><span class="n">max_loras</span>
|
||
<span class="k">if</span> <span class="n">lora_config</span><span class="o">.</span><span class="n">max_cpu_loras</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="n">peft_cache_config_model</span><span class="o">.</span><span class="n">num_host_module_layer</span> <span class="o">=</span> \
|
||
<span class="n">max_lora_rank</span> <span class="o">*</span> <span class="n">num_lora_modules</span> <span class="o">*</span> <span class="n">lora_config</span><span class="o">.</span><span class="n">max_cpu_loras</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span><span class="o">.</span><span class="n">peft_cache_config</span> <span class="o">=</span> <span class="n">peft_cache_config_model</span><span class="o">.</span><span class="n">_to_pybind</span><span class="p">(</span>
|
||
<span class="p">)</span>
|
||
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">decoding_config</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span><span class="o">.</span><span class="n">decoding_config</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">decoding_config</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">guided_decoding_backend</span> <span class="o">==</span> <span class="s1">'xgrammar'</span><span class="p">:</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span><span class="o">.</span><span class="n">guided_decoding_config</span> <span class="o">=</span> <span class="n">tllm</span><span class="o">.</span><span class="n">GuidedDecodingConfig</span><span class="p">(</span>
|
||
<span class="n">backend</span><span class="o">=</span><span class="n">tllm</span><span class="o">.</span><span class="n">GuidedDecodingConfig</span><span class="o">.</span><span class="n">GuidedDecodingBackend</span><span class="o">.</span>
|
||
<span class="n">XGRAMMAR</span><span class="p">,</span>
|
||
<span class="o">**</span><span class="n">_xgrammar_tokenizer_info</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">tokenizer</span><span class="p">))</span>
|
||
<span class="k">elif</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">guided_decoding_backend</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"Unsupported guided decoding backend </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">guided_decoding_backend</span><span class="si">}</span><span class="s2">"</span>
|
||
<span class="p">)</span>
|
||
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span><span class="o">.</span><span class="n">normalize_log_probs</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">normalize_log_probs</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span><span class="o">.</span><span class="n">enable_chunked_context</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">enable_chunked_prefill</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span><span class="o">.</span><span class="n">max_beam_width</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">max_beam_width</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">max_beam_width</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">extended_runtime_perf_knob_config</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span><span class="o">.</span><span class="n">extended_runtime_perf_knob_config</span> <span class="o">=</span> <span class="n">PybindMirror</span><span class="o">.</span><span class="n">maybe_to_pybind</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">extended_runtime_perf_knob_config</span><span class="p">)</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">cache_transceiver_config</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span><span class="o">.</span><span class="n">cache_transceiver_config</span> <span class="o">=</span> <span class="n">PybindMirror</span><span class="o">.</span><span class="n">maybe_to_pybind</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">cache_transceiver_config</span><span class="p">)</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span><span class="o">.</span><span class="n">llm_parallel_config</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">parallel_config</span>
|
||
<span class="n">return_logits</span> <span class="o">=</span> <span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">gather_generation_logits</span>
|
||
<span class="ow">or</span> <span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">build_config</span>
|
||
<span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">gather_context_logits</span><span class="p">))</span>
|
||
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_executor_cls</span><span class="o">.</span><span class="n">create</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_engine_dir</span><span class="p">,</span>
|
||
<span class="n">executor_config</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_executor_config</span><span class="p">,</span>
|
||
<span class="n">batched_logits_processor</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">batched_logits_processor</span><span class="p">,</span>
|
||
<span class="n">model_world_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">world_size</span><span class="p">,</span>
|
||
<span class="n">mpi_session</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span><span class="p">,</span>
|
||
<span class="n">reuse_mpi_comm</span><span class="o">=</span><span class="n">external_mpi_comm_available</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">world_size</span><span class="p">),</span>
|
||
<span class="n">return_logits</span><span class="o">=</span><span class="n">return_logits</span><span class="p">,</span>
|
||
<span class="n">postproc_worker_config</span><span class="o">=</span><span class="n">PostprocWorkerConfig</span><span class="p">(</span>
|
||
<span class="n">num_postprocess_workers</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">num_postprocess_workers</span><span class="p">,</span>
|
||
<span class="n">postprocess_tokenizer_dir</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">postprocess_tokenizer_dir</span><span class="p">,</span>
|
||
<span class="p">),</span>
|
||
<span class="n">is_llm_executor</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
||
<span class="n">lora_config</span><span class="o">=</span><span class="n">lora_config</span><span class="p">)</span>
|
||
|
||
|
||
<span class="nd">@append_docstring</span><span class="p">(</span><span class="n">TORCH_LLM_DOCSTRING</span><span class="p">)</span>
|
||
<span class="k">class</span><span class="w"> </span><span class="nc">_TorchLLM</span><span class="p">(</span><span class="n">BaseLLM</span><span class="p">):</span>
|
||
<span class="w"> </span><span class="sd">"""LLM class is the main class for running a LLM model using PyTorch backend.</span>
|
||
|
||
<span class="sd"> Parameters:</span>
|
||
<span class="sd">"""</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span>
|
||
<span class="n">model</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Path</span><span class="p">],</span>
|
||
<span class="n">tokenizer</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Path</span><span class="p">,</span> <span class="n">TokenizerBase</span><span class="p">,</span>
|
||
<span class="n">PreTrainedTokenizerBase</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">tokenizer_mode</span><span class="p">:</span> <span class="n">Literal</span><span class="p">[</span><span class="s1">'auto'</span><span class="p">,</span> <span class="s1">'slow'</span><span class="p">]</span> <span class="o">=</span> <span class="s1">'auto'</span><span class="p">,</span>
|
||
<span class="n">skip_tokenizer_init</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
|
||
<span class="n">trust_remote_code</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
|
||
<span class="n">tensor_parallel_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">1</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="s2">"auto"</span><span class="p">,</span>
|
||
<span class="n">revision</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">tokenizer_revision</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="o">**</span><span class="n">kwargs</span><span class="p">:</span> <span class="n">Any</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
|
||
|
||
<span class="c1"># TODO: deprecate backend in LLM kwargs</span>
|
||
<span class="n">backend</span> <span class="o">=</span> <span class="n">kwargs</span><span class="o">.</span><span class="n">pop</span><span class="p">(</span><span class="s2">"backend"</span><span class="p">,</span> <span class="s2">"pytorch"</span><span class="p">)</span>
|
||
|
||
<span class="c1"># Validate that users don't pass TrtLlmArgs-specific arguments</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_validate_args_for_torch_backend</span><span class="p">(</span><span class="n">kwargs</span><span class="p">)</span>
|
||
|
||
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">(</span><span class="n">model</span><span class="p">,</span>
|
||
<span class="n">tokenizer</span><span class="p">,</span>
|
||
<span class="n">tokenizer_mode</span><span class="p">,</span>
|
||
<span class="n">skip_tokenizer_init</span><span class="p">,</span>
|
||
<span class="n">trust_remote_code</span><span class="p">,</span>
|
||
<span class="n">tensor_parallel_size</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="p">,</span>
|
||
<span class="n">revision</span><span class="p">,</span>
|
||
<span class="n">tokenizer_revision</span><span class="p">,</span>
|
||
<span class="n">backend</span><span class="o">=</span><span class="n">backend</span><span class="p">,</span>
|
||
<span class="o">**</span><span class="n">kwargs</span><span class="p">)</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">_build_model</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
||
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="n">_build_model</span><span class="p">()</span>
|
||
<span class="k">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">_engine_dir</span> <span class="ow">is</span> <span class="kc">None</span>
|
||
|
||
<span class="c1"># Tokenizer loading should be after calling model_loader(), since model_loader() may download the model from HF hub.</span>
|
||
<span class="c1"># It should also be before bindings ExecutorConfig, which may depend on tokenizer info.</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_tokenizer</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_try_load_tokenizer</span><span class="p">()</span>
|
||
|
||
<span class="c1"># Multimodal special handling:</span>
|
||
<span class="c1"># 1. Default load_tokenizer may fail because MM has different tokenizer configuration. Hence we initialize it inside input processor</span>
|
||
<span class="c1"># 2. May need to modify model weights for MM (e.g., resize vocab embedding). We must do such operation via input processor's __init__</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">input_processor</span> <span class="o">=</span> <span class="n">create_input_processor</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_hf_model_dir</span><span class="p">,</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">tokenizer</span><span class="p">)</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_tokenizer</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">input_processor</span><span class="o">.</span><span class="n">tokenizer</span>
|
||
|
||
<span class="c1"># TODO: revisit gather_context_logits</span>
|
||
<span class="n">return_logits</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">gather_generation_logits</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_executor</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_executor_cls</span><span class="o">.</span><span class="n">create</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">_engine_dir</span><span class="p">,</span>
|
||
<span class="n">executor_config</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">batched_logits_processor</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">batched_logits_processor</span><span class="p">,</span>
|
||
<span class="n">model_world_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">world_size</span><span class="p">,</span>
|
||
<span class="n">mpi_session</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span><span class="p">,</span>
|
||
<span class="n">reuse_mpi_comm</span><span class="o">=</span><span class="n">external_mpi_comm_available</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">world_size</span><span class="p">),</span>
|
||
<span class="n">return_logits</span><span class="o">=</span><span class="n">return_logits</span><span class="p">,</span>
|
||
<span class="n">postproc_worker_config</span><span class="o">=</span><span class="n">PostprocWorkerConfig</span><span class="p">(</span>
|
||
<span class="n">num_postprocess_workers</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">num_postprocess_workers</span><span class="p">,</span>
|
||
<span class="n">postprocess_tokenizer_dir</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">postprocess_tokenizer_dir</span><span class="p">,</span>
|
||
<span class="p">),</span>
|
||
<span class="n">is_llm_executor</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
||
<span class="n">lora_config</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="o">.</span><span class="n">lora_config</span><span class="p">,</span>
|
||
<span class="c1"># Autodeploy does not support kv_connector_config</span>
|
||
<span class="n">kv_connector_config</span><span class="o">=</span><span class="nb">getattr</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="p">,</span> <span class="s2">"kv_connector_config"</span><span class="p">,</span> <span class="kc">None</span><span class="p">),</span>
|
||
<span class="n">hf_model_dir</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_hf_model_dir</span><span class="p">,</span>
|
||
<span class="n">tokenizer</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">tokenizer</span><span class="p">,</span>
|
||
<span class="n">llm_args</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">args</span><span class="p">)</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">_validate_args_for_torch_backend</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">kwargs</span><span class="p">:</span> <span class="nb">dict</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="w"> </span><span class="sd">"""Validate that users don't pass TrtLlmArgs-specific arguments when using PyTorch backend.</span>
|
||
<span class="sd"> """</span>
|
||
<span class="n">trtllm_fields</span> <span class="o">=</span> <span class="nb">set</span><span class="p">(</span><span class="n">TrtLlmArgs</span><span class="o">.</span><span class="n">model_fields</span><span class="o">.</span><span class="n">keys</span><span class="p">())</span>
|
||
<span class="n">torchllm_fields</span> <span class="o">=</span> <span class="nb">set</span><span class="p">(</span><span class="n">TorchLlmArgs</span><span class="o">.</span><span class="n">model_fields</span><span class="o">.</span><span class="n">keys</span><span class="p">())</span>
|
||
|
||
<span class="n">trtllm_specific_fields</span> <span class="o">=</span> <span class="n">trtllm_fields</span> <span class="o">-</span> <span class="n">torchllm_fields</span>
|
||
|
||
<span class="c1"># Check if any TrtLlmArgs-specific arguments are passed</span>
|
||
<span class="n">trtllm_specific_args</span> <span class="o">=</span> <span class="p">[]</span>
|
||
<span class="k">for</span> <span class="n">key</span> <span class="ow">in</span> <span class="n">kwargs</span><span class="p">:</span>
|
||
<span class="k">if</span> <span class="n">key</span> <span class="ow">in</span> <span class="n">trtllm_specific_fields</span><span class="p">:</span>
|
||
<span class="n">trtllm_specific_args</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">key</span><span class="p">)</span>
|
||
|
||
<span class="k">if</span> <span class="n">trtllm_specific_args</span><span class="p">:</span>
|
||
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
|
||
<span class="sa">f</span><span class="s2">"The following arguments are specific to TensorRT backend and cannot be used with PyTorch backend: </span><span class="si">{</span><span class="n">trtllm_specific_args</span><span class="si">}</span><span class="s2">.</span><span class="se">\n</span><span class="s2">"</span>
|
||
<span class="sa">f</span><span class="s2">"Please use 'from tensorrt_llm._tensorrt_engine import LLM' instead to use the TensorRT backend."</span>
|
||
<span class="p">)</span>
|
||
|
||
|
||
<div class="viewcode-block" id="LLM">
|
||
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.LLM">[docs]</a>
|
||
<span class="k">class</span><span class="w"> </span><span class="nc">LLM</span><span class="p">(</span><span class="n">_TorchLLM</span><span class="p">):</span>
|
||
|
||
<div class="viewcode-block" id="LLM.__init__">
|
||
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.LLM.__init__">[docs]</a>
|
||
<span class="k">def</span><span class="w"> </span><span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span>
|
||
<span class="n">model</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Path</span><span class="p">],</span>
|
||
<span class="n">tokenizer</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Path</span><span class="p">,</span> <span class="n">TokenizerBase</span><span class="p">,</span>
|
||
<span class="n">PreTrainedTokenizerBase</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">tokenizer_mode</span><span class="p">:</span> <span class="n">Literal</span><span class="p">[</span><span class="s1">'auto'</span><span class="p">,</span> <span class="s1">'slow'</span><span class="p">]</span> <span class="o">=</span> <span class="s1">'auto'</span><span class="p">,</span>
|
||
<span class="n">skip_tokenizer_init</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
|
||
<span class="n">trust_remote_code</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
|
||
<span class="n">tensor_parallel_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">1</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="s2">"auto"</span><span class="p">,</span>
|
||
<span class="n">revision</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">tokenizer_revision</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="o">**</span><span class="n">kwargs</span><span class="p">:</span> <span class="n">Any</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">tokenizer</span><span class="p">,</span> <span class="n">tokenizer_mode</span><span class="p">,</span> <span class="n">skip_tokenizer_init</span><span class="p">,</span>
|
||
<span class="n">trust_remote_code</span><span class="p">,</span> <span class="n">tensor_parallel_size</span><span class="p">,</span> <span class="n">dtype</span><span class="p">,</span>
|
||
<span class="n">revision</span><span class="p">,</span> <span class="n">tokenizer_revision</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">)</span></div>
|
||
</div>
|
||
|
||
|
||
|
||
<span class="c1"># sphinx will ignore the LLM's docstring if it is not explicitly set</span>
|
||
<span class="n">LLM</span><span class="o">.</span><span class="vm">__doc__</span> <span class="o">=</span> \
|
||
<span class="sa">f</span><span class="s2">"""LLM class is the main class for running a LLM model.</span>
|
||
|
||
<span class="s2"> For more details about the arguments, please refer to :class:`TorchLlmArgs`.</span>
|
||
|
||
<span class="s2"> Parameters:</span>
|
||
<span class="s2">"""</span> <span class="o">+</span> <span class="n">TORCH_LLM_DOCSTRING</span>
|
||
</pre></div>
|
||
|
||
</article>
|
||
|
||
|
||
|
||
|
||
|
||
<footer class="prev-next-footer d-print-none">
|
||
|
||
<div class="prev-next-area">
|
||
</div>
|
||
</footer>
|
||
|
||
</div>
|
||
|
||
|
||
|
||
<div class="bd-sidebar-secondary"></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 © 2025, NVidia.
|
||
<br/>
|
||
|
||
</p>
|
||
</div>
|
||
|
||
<div class="footer-item">
|
||
<div class="extra_footer">
|
||
|
||
<p>Last updated on September 02, 2025.</p>
|
||
|
||
<p>This page is generated by TensorRT-LLM commit <a href="https://github.com/NVIDIA/TensorRT-LLM/tree/e81c50d">e81c50d</a>.</p>
|
||
|
||
</div></div>
|
||
|
||
</div>
|
||
|
||
|
||
|
||
</div>
|
||
|
||
</footer>
|
||
</body>
|
||
</html> |