mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-14 06:27:45 +08:00
2547 lines
296 KiB
HTML
2547 lines
296 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_args — 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" />
|
|
|
|
<!-- 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>DOCUMENTATION_OPTIONS.pagename = '_modules/tensorrt_llm/llmapi/llm_args';</script>
|
|
<script>
|
|
DOCUMENTATION_OPTIONS.theme_version = '0.16.1';
|
|
DOCUMENTATION_OPTIONS.theme_switcher_json_url = './_static/switcher.json';
|
|
DOCUMENTATION_OPTIONS.theme_switcher_version_match = '0.21.0rc0';
|
|
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="0.21.0rc0" />
|
|
|
|
|
|
</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/linux.html">Installing on Linux</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../installation/build-from-source-linux.html">Building from Source Code on Linux</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../installation/grace-hopper.html">Installing on Grace Hopper</a></li>
|
|
</ul>
|
|
<p aria-level="2" class="caption" role="heading"><span class="caption-text">LLM API</span></p>
|
|
<ul class="nav bd-sidenav">
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../llm-api/index.html">API Introduction</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../llm-api/reference.html">API Reference</a></li>
|
|
</ul>
|
|
<p aria-level="2" class="caption" role="heading"><span class="caption-text">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>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_medusa_decoding.html">Generate Text Using Medusa Decoding</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_eagle_decoding.html">Generate Text Using Eagle Decoding</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_distributed.html">Distributed LLM Generation</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_eagle2_decoding.html">Generate Text Using Eagle2 Decoding</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_inference_kv_events.html">Get KV Cache Events</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_lookahead_decoding.html">Generate Text Using Lookahead Decoding</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_quantization.html">Generation with Quantization</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_guided_decoding.html">Generate text with guided decoding</a></li>
|
|
<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_customize.html">Generate text with customization</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_auto_parallel.html">Automatic Parallelism with LLM</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_mgmn_llm_distributed.html">Llm Mgmn Llm Distributed</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_mgmn_trtllm_bench.html">Llm Mgmn Trtllm Bench</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_mgmn_trtllm_serve.html">Llm Mgmn Trtllm Serve</a></li>
|
|
</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_medusa_decoding.html">Generate Text Using Medusa Decoding</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_eagle_decoding.html">Generate Text Using Eagle Decoding</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_distributed.html">Distributed LLM Generation</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_eagle2_decoding.html">Generate Text Using Eagle2 Decoding</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_inference_kv_events.html">Get KV Cache Events</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_lookahead_decoding.html">Generate Text Using Lookahead Decoding</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_quantization.html">Generation with Quantization</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_guided_decoding.html">Generate text with guided decoding</a></li>
|
|
<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_customize.html">Generate text with customization</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_auto_parallel.html">Automatic Parallelism with LLM</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_mgmn_llm_distributed.html">Llm Mgmn Llm Distributed</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_mgmn_trtllm_bench.html">Llm Mgmn Trtllm Bench</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../examples/llm_mgmn_trtllm_serve.html">Llm Mgmn Trtllm Serve</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</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../examples/openai_completion_client.html">OpenAI Completion Client</a></li>
|
|
</ul>
|
|
</details></li>
|
|
</ul>
|
|
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Model Definition API</span></p>
|
|
<ul class="nav bd-sidenav">
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../python-api/tensorrt_llm.layers.html">Layers</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../python-api/tensorrt_llm.functional.html">Functionals</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../python-api/tensorrt_llm.models.html">Models</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../python-api/tensorrt_llm.plugin.html">Plugin</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../python-api/tensorrt_llm.quantization.html">Quantization</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../python-api/tensorrt_llm.runtime.html">Runtime</a></li>
|
|
</ul>
|
|
<p aria-level="2" class="caption" role="heading"><span class="caption-text">C++ API</span></p>
|
|
<ul class="nav bd-sidenav">
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../_cpp_gen/executor.html">Executor</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../_cpp_gen/runtime.html">Runtime</a></li>
|
|
</ul>
|
|
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Command-Line Reference</span></p>
|
|
<ul class="nav bd-sidenav">
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../commands/trtllm-build.html">trtllm-build</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../commands/trtllm-serve.html">trtllm-serve</a></li>
|
|
</ul>
|
|
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Architecture</span></p>
|
|
<ul class="nav bd-sidenav">
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../architecture/overview.html">TensorRT-LLM Architecture</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../architecture/core-concepts.html">Model Definition</a></li>
|
|
|
|
|
|
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../architecture/checkpoint.html">TensorRT-LLM Checkpoint</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../architecture/workflow.html">TensorRT-LLM Build Workflow</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../architecture/add-model.html">Adding a Model</a></li>
|
|
</ul>
|
|
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Advanced</span></p>
|
|
<ul class="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 (experimental)</a></li>
|
|
</ul>
|
|
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Performance</span></p>
|
|
<ul class="nav bd-sidenav">
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../performance/perf-overview.html">Overview</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../performance/perf-benchmarking.html">Benchmarking</a></li>
|
|
<li class="toctree-l1 has-children"><a class="reference internal" href="../../../performance/performance-tuning-guide/index.html">Performance Tuning Guide</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../performance/performance-tuning-guide/benchmarking-default-performance.html">Benchmarking Default Performance</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../performance/performance-tuning-guide/useful-build-time-flags.html">Useful Build-Time Flags</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../performance/performance-tuning-guide/tuning-max-batch-size-and-max-num-tokens.html">Tuning Max Batch Size and Max Num Tokens</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../performance/performance-tuning-guide/deciding-model-sharding-strategy.html">Deciding Model Sharding Strategy</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../performance/performance-tuning-guide/fp8-quantization.html">FP8 Quantization</a></li>
|
|
<li class="toctree-l2"><a class="reference internal" href="../../../performance/performance-tuning-guide/useful-runtime-flags.html">Useful Runtime Options</a></li>
|
|
</ul>
|
|
</details></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../performance/perf-analysis.html">Performance Analysis</a></li>
|
|
</ul>
|
|
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Reference</span></p>
|
|
<ul class="nav bd-sidenav">
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../reference/troubleshooting.html">Troubleshooting</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../reference/support-matrix.html">Support Matrix</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../reference/precision.html">Numerical Precision</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../reference/memory.html">Memory Usage of TensorRT-LLM</a></li>
|
|
</ul>
|
|
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Blogs</span></p>
|
|
<ul class="nav bd-sidenav">
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../blogs/H100vsA100.html">H100 has 4.6x A100 Performance in TensorRT-LLM, achieving 10,000 tok/s at 100ms to first token</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../blogs/H200launch.html">H200 achieves nearly 12,000 tokens/sec on Llama2-13B with TensorRT-LLM</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../blogs/Falcon180B-H200.html">Falcon-180B on a single H200 GPU with INT4 AWQ, and 6.7x faster Llama-70B over A100</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../blogs/quantization-in-TRT-LLM.html">Speed up inference with SOTA quantization techniques in TRT-LLM</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../blogs/XQA-kernel.html">New XQA-kernel provides 2.4x more Llama-70B throughput within the same latency budget</a></li>
|
|
<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>
|
|
</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_args</span></li>
|
|
</ul>
|
|
</nav>
|
|
</div>
|
|
|
|
</div>
|
|
|
|
|
|
</div>
|
|
</div>
|
|
|
|
|
|
|
|
|
|
<div id="searchbox"></div>
|
|
<article class="bd-article">
|
|
|
|
<h1>Source code for tensorrt_llm.llmapi.llm_args</h1><div class="highlight"><pre>
|
|
<span></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">math</span>
|
|
<span class="kn">import</span><span class="w"> </span><span class="nn">os</span>
|
|
<span class="kn">from</span><span class="w"> </span><span class="nn">abc</span><span class="w"> </span><span class="kn">import</span> <span class="n">ABC</span><span class="p">,</span> <span class="n">abstractmethod</span>
|
|
<span class="kn">from</span><span class="w"> </span><span class="nn">dataclasses</span><span class="w"> </span><span class="kn">import</span> <span class="n">dataclass</span><span class="p">,</span> <span class="n">field</span>
|
|
<span class="kn">from</span><span class="w"> </span><span class="nn">enum</span><span class="w"> </span><span class="kn">import</span> <span class="n">Enum</span><span class="p">,</span> <span class="n">EnumMeta</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="p">(</span><span class="n">TYPE_CHECKING</span><span class="p">,</span> <span class="n">Any</span><span class="p">,</span> <span class="n">ClassVar</span><span class="p">,</span> <span class="n">Dict</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">Union</span><span class="p">)</span>
|
|
|
|
<span class="kn">import</span><span class="w"> </span><span class="nn">torch</span>
|
|
<span class="kn">import</span><span class="w"> </span><span class="nn">yaml</span>
|
|
<span class="kn">from</span><span class="w"> </span><span class="nn">pydantic</span><span class="w"> </span><span class="kn">import</span> <span class="p">(</span><span class="n">BaseModel</span><span class="p">,</span> <span class="n">Field</span><span class="p">,</span> <span class="n">PrivateAttr</span><span class="p">,</span> <span class="n">field_validator</span><span class="p">,</span>
|
|
<span class="n">model_validator</span><span class="p">)</span>
|
|
<span class="kn">from</span><span class="w"> </span><span class="nn">strenum</span><span class="w"> </span><span class="kn">import</span> <span class="n">StrEnum</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.lora_manager</span><span class="w"> </span><span class="kn">import</span> <span class="p">(</span><span class="n">LoraConfig</span><span class="p">,</span>
|
|
<span class="n">get_default_trtllm_modules_to_hf_modules</span><span class="p">)</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">mpi_rank</span>
|
|
<span class="kn">from</span><span class="w"> </span><span class="nn">..auto_parallel</span><span class="w"> </span><span class="kn">import</span> <span class="n">AutoParallelConfig</span><span class="p">,</span> <span class="n">infer_cluster_config</span>
|
|
|
|
<span class="k">if</span> <span class="n">TYPE_CHECKING</span><span class="p">:</span>
|
|
<span class="kn">from</span><span class="w"> </span><span class="nn">tensorrt_llm._torch.pyexecutor.config</span><span class="w"> </span><span class="kn">import</span> <span class="n">PyTorchConfig</span>
|
|
|
|
<span class="c1"># yapf: disable</span>
|
|
<span class="c1"># isort: off</span>
|
|
<span class="kn">from</span><span class="w"> </span><span class="nn">..bindings.executor</span><span class="w"> </span><span class="kn">import</span> <span class="p">(</span>
|
|
<span class="n">BatchingType</span> <span class="k">as</span> <span class="n">_BatchingType</span><span class="p">,</span>
|
|
<span class="n">CacheTransceiverConfig</span> <span class="k">as</span> <span class="n">_CacheTransceiverConfig</span><span class="p">,</span>
|
|
<span class="n">CapacitySchedulerPolicy</span> <span class="k">as</span> <span class="n">_CapacitySchedulerPolicy</span><span class="p">,</span>
|
|
<span class="n">ContextChunkingPolicy</span> <span class="k">as</span> <span class="n">_ContextChunkingPolicy</span><span class="p">,</span>
|
|
<span class="n">DecodingConfig</span><span class="p">,</span>
|
|
<span class="n">DecodingMode</span><span class="p">,</span>
|
|
<span class="n">DynamicBatchConfig</span> <span class="k">as</span> <span class="n">_DynamicBatchConfig</span><span class="p">,</span>
|
|
<span class="n">EagleConfig</span> <span class="k">as</span> <span class="n">_EagleConfig</span><span class="p">,</span>
|
|
<span class="n">ExecutorConfig</span> <span class="k">as</span> <span class="n">_ExecutorConfig</span><span class="p">,</span>
|
|
<span class="n">ExtendedRuntimePerfKnobConfig</span> <span class="k">as</span> <span class="n">_ExtendedRuntimePerfKnobConfig</span><span class="p">,</span>
|
|
<span class="n">KvCacheConfig</span> <span class="k">as</span> <span class="n">_KvCacheConfig</span><span class="p">,</span>
|
|
<span class="n">LookaheadDecodingConfig</span> <span class="k">as</span> <span class="n">_LookaheadDecodingConfig</span><span class="p">,</span>
|
|
<span class="n">PeftCacheConfig</span> <span class="k">as</span> <span class="n">_PeftCacheConfig</span><span class="p">,</span>
|
|
<span class="n">SchedulerConfig</span> <span class="k">as</span> <span class="n">_SchedulerConfig</span><span class="p">)</span> <span class="c1"># isort: skip</span>
|
|
<span class="c1"># isort: on</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="c1"># yapf: enable</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">BuildConfig</span><span class="p">,</span> <span class="n">EngineConfig</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">..mapping</span><span class="w"> </span><span class="kn">import</span> <span class="n">Mapping</span>
|
|
<span class="kn">from</span><span class="w"> </span><span class="nn">..models.automodel</span><span class="w"> </span><span class="kn">import</span> <span class="n">AutoConfig</span>
|
|
<span class="kn">from</span><span class="w"> </span><span class="nn">..models.modeling_utils</span><span class="w"> </span><span class="kn">import</span> <span class="p">(</span><span class="n">PretrainedConfig</span><span class="p">,</span> <span class="n">QuantAlgo</span><span class="p">,</span> <span class="n">QuantConfig</span><span class="p">,</span>
|
|
<span class="n">SpeculativeDecodingMode</span><span class="p">)</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">BatchedLogitsProcessor</span>
|
|
<span class="kn">from</span><span class="w"> </span><span class="nn">.build_cache</span><span class="w"> </span><span class="kn">import</span> <span class="n">BuildCacheConfig</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">tokenizer_factory</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">generate_api_docs_as_docstring</span><span class="p">,</span> <span class="n">get_type_repr</span><span class="p">,</span>
|
|
<span class="n">print_traceback_on_error</span><span class="p">)</span>
|
|
|
|
<span class="c1"># TODO[chunweiy]: move the following symbols back to utils scope, and remove the following import</span>
|
|
|
|
|
|
<span class="nd">@dataclass</span>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">_ParallelConfig</span><span class="p">:</span>
|
|
<span class="w"> </span><span class="sd">''' The model distribution configs for LLM. '''</span>
|
|
<span class="n">tp_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">1</span>
|
|
<span class="n">pp_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">1</span>
|
|
<span class="n">cp_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">1</span>
|
|
<span class="n">gpus_per_node</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">8</span>
|
|
<span class="n">moe_cluster_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">1</span>
|
|
<span class="n">moe_tp_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">1</span>
|
|
<span class="n">moe_ep_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">1</span>
|
|
<span class="n">cp_config</span><span class="p">:</span> <span class="nb">dict</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span><span class="n">default_factory</span><span class="o">=</span><span class="nb">dict</span><span class="p">)</span>
|
|
<span class="n">enable_attention_dp</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span>
|
|
<span class="n">auto_parallel</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span>
|
|
|
|
<span class="n">_world_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
|
|
<span class="n">_devices</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">devices</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]:</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_devices</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="nb">list</span><span class="p">(</span><span class="nb">range</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">world_size</span><span class="p">))</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_devices</span>
|
|
|
|
<span class="nd">@devices</span><span class="o">.</span><span class="n">setter</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">devices</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">devices</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]):</span>
|
|
<span class="k">if</span> <span class="nb">len</span><span class="p">(</span><span class="n">devices</span><span class="p">)</span> <span class="o">!=</span> <span class="bp">self</span><span class="o">.</span><span class="n">world_size</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">"devices </span><span class="si">{</span><span class="n">devices</span><span class="si">}</span><span class="s2"> should have the same length as world_size </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">world_size</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">_devices</span> <span class="o">=</span> <span class="n">devices</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">world_size</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">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">auto_parallel</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">tp_size</span> <span class="o">></span> <span class="mi">1</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">pp_size</span> <span class="o">></span> <span class="mi">1</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">cp_size</span> <span class="o">></span> <span class="mi">1</span><span class="p">:</span>
|
|
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span>
|
|
<span class="s2">"manually TP and PP are not supported in auto parallel mode."</span>
|
|
<span class="p">)</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_world_size</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_world_size</span> <span class="o">></span> <span class="mi">1</span><span class="p">:</span>
|
|
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span>
|
|
<span class="s2">"world_size > 1 is only supported in auto parallel mode."</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">tp_size</span> <span class="o">*</span> <span class="bp">self</span><span class="o">.</span><span class="n">pp_size</span> <span class="o">*</span> <span class="bp">self</span><span class="o">.</span><span class="n">cp_size</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">world_size_per_node</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="nb">int</span><span class="p">:</span>
|
|
<span class="n">world_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">world_size</span>
|
|
<span class="n">total_nodes</span> <span class="o">=</span> <span class="n">math</span><span class="o">.</span><span class="n">ceil</span><span class="p">(</span><span class="n">world_size</span> <span class="o">/</span> <span class="bp">self</span><span class="o">.</span><span class="n">gpus_per_node</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="n">world_size</span> <span class="o">//</span> <span class="n">total_nodes</span> <span class="c1">#TODO is this right?</span>
|
|
|
|
<span class="nd">@world_size</span><span class="o">.</span><span class="n">setter</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">world_size</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">world_size</span><span class="p">:</span> <span class="nb">int</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">auto_parallel</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_world_size</span> <span class="o">=</span> <span class="n">world_size</span>
|
|
<span class="k">elif</span> <span class="p">(</span><span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">auto_parallel</span>
|
|
<span class="p">)</span> <span class="ow">and</span> <span class="n">world_size</span> <span class="o">!=</span> <span class="bp">self</span><span class="o">.</span><span class="n">tp_size</span> <span class="o">*</span> <span class="bp">self</span><span class="o">.</span><span class="n">pp_size</span> <span class="o">*</span> <span class="bp">self</span><span class="o">.</span><span class="n">cp_size</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">"world_size </span><span class="si">{</span><span class="n">world_size</span><span class="si">}</span><span class="s2"> should be equal to tp_size * pp_size </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">tp_size</span><span class="w"> </span><span class="o">*</span><span class="w"> </span><span class="bp">self</span><span class="o">.</span><span class="n">pp_size</span><span class="w"> </span><span class="o">*</span><span class="w"> </span><span class="bp">self</span><span class="o">.</span><span class="n">cp_size</span><span class="si">}</span><span class="s2"> "</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">is_multi_gpu</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="bp">self</span><span class="o">.</span><span class="n">world_size</span> <span class="o">></span> <span class="mi">1</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">to_mapping</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="n">Mapping</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="n">Mapping</span><span class="p">(</span><span class="n">world_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">world_size</span><span class="p">,</span>
|
|
<span class="n">rank</span><span class="o">=</span><span class="n">mpi_rank</span><span class="p">(),</span>
|
|
<span class="n">gpus_per_node</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">gpus_per_node</span><span class="p">,</span>
|
|
<span class="n">tp_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">pp_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">pp_size</span><span class="p">,</span>
|
|
<span class="n">cp_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">cp_size</span><span class="p">,</span>
|
|
<span class="n">cp_config</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">cp_config</span><span class="p">,</span>
|
|
<span class="n">enable_attention_dp</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">enable_attention_dp</span><span class="p">,</span>
|
|
<span class="n">moe_cluster_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">moe_cluster_size</span><span class="p">,</span>
|
|
<span class="n">moe_tp_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">moe_tp_size</span><span class="p">,</span>
|
|
<span class="n">moe_ep_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">moe_ep_size</span><span class="p">,</span>
|
|
<span class="n">auto_parallel</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">auto_parallel</span><span class="p">)</span>
|
|
|
|
|
|
<div class="viewcode-block" id="CalibConfig">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.CalibConfig">[docs]</a>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">CalibConfig</span><span class="p">(</span><span class="n">BaseModel</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">"""</span>
|
|
<span class="sd"> Calibration configuration.</span>
|
|
<span class="sd"> """</span>
|
|
<span class="n">device</span><span class="p">:</span> <span class="n">Literal</span><span class="p">[</span><span class="s1">'cuda'</span><span class="p">,</span>
|
|
<span class="s1">'cpu'</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="s1">'cuda'</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The device to run calibration."</span><span class="p">)</span>
|
|
<span class="n">calib_dataset</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="s1">'cnn_dailymail'</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The name or local path of calibration dataset."</span><span class="p">)</span>
|
|
<span class="n">calib_batches</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">512</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The number of batches that the calibration runs."</span><span class="p">)</span>
|
|
<span class="n">calib_batch_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"The batch size that the calibration runs."</span><span class="p">)</span>
|
|
<span class="n">calib_max_seq_length</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">512</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The maximum sequence length that the calibration runs."</span><span class="p">)</span>
|
|
<span class="n">random_seed</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">1234</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"The random seed used for calibration."</span><span class="p">)</span>
|
|
<span class="n">tokenizer_max_seq_length</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">2048</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"The maximum sequence length to initialize tokenizer for calibration."</span><span class="p">)</span>
|
|
|
|
<div class="viewcode-block" id="CalibConfig.from_dict">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.CalibConfig.from_dict">[docs]</a>
|
|
<span class="nd">@classmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">from_dict</span><span class="p">(</span><span class="bp">cls</span><span class="p">,</span> <span class="n">config</span><span class="p">:</span> <span class="nb">dict</span><span class="p">)</span> <span class="o">-></span> <span class="s1">'CalibConfig'</span><span class="p">:</span>
|
|
<span class="w"> </span><span class="sd">"""Create a CalibConfig instance from a dict.</span>
|
|
|
|
<span class="sd"> Args:</span>
|
|
<span class="sd"> config (dict): The dict used to create CalibConfig.</span>
|
|
|
|
<span class="sd"> Returns:</span>
|
|
<span class="sd"> tensorrt_llm.llmapi.CalibConfig: The CalibConfig created from dict.</span>
|
|
<span class="sd"> """</span>
|
|
<span class="k">return</span> <span class="bp">cls</span><span class="p">(</span><span class="o">**</span><span class="n">config</span><span class="p">)</span></div>
|
|
|
|
|
|
<div class="viewcode-block" id="CalibConfig.to_dict">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.CalibConfig.to_dict">[docs]</a>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">to_dict</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="nb">dict</span><span class="p">:</span>
|
|
<span class="w"> </span><span class="sd">"""Dump a CalibConfig instance to a dict.</span>
|
|
|
|
<span class="sd"> Returns:</span>
|
|
<span class="sd"> dict: The dict dumped from CalibConfig.</span>
|
|
<span class="sd"> """</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">model_dump</span><span class="p">()</span></div>
|
|
</div>
|
|
|
|
|
|
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">_ModelFormatKind</span><span class="p">(</span><span class="n">Enum</span><span class="p">):</span>
|
|
<span class="n">HF</span> <span class="o">=</span> <span class="mi">0</span>
|
|
<span class="n">TLLM_CKPT</span> <span class="o">=</span> <span class="mi">1</span>
|
|
<span class="n">TLLM_ENGINE</span> <span class="o">=</span> <span class="mi">2</span>
|
|
|
|
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">DecodingBaseConfig</span><span class="p">(</span><span class="n">BaseModel</span><span class="p">):</span>
|
|
<span class="n">max_draft_len</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="n">speculative_model</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="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="nd">@classmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">from_dict</span><span class="p">(</span><span class="bp">cls</span><span class="p">,</span> <span class="n">data</span><span class="p">:</span> <span class="nb">dict</span><span class="p">):</span>
|
|
<span class="c1"># dispatch to the correct decoding config</span>
|
|
<span class="n">decoding_type</span> <span class="o">=</span> <span class="n">data</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s2">"decoding_type"</span><span class="p">)</span>
|
|
<span class="n">config_classes</span> <span class="o">=</span> <span class="p">{</span>
|
|
<span class="s2">"MTP"</span><span class="p">:</span> <span class="n">MTPDecodingConfig</span><span class="p">,</span>
|
|
<span class="s2">"Medusa"</span><span class="p">:</span> <span class="n">MedusaDecodingConfig</span><span class="p">,</span>
|
|
<span class="s2">"Eagle"</span><span class="p">:</span> <span class="n">EagleDecodingConfig</span><span class="p">,</span>
|
|
<span class="s2">"Lookahead"</span><span class="p">:</span> <span class="n">LookaheadDecodingConfig</span><span class="p">,</span>
|
|
<span class="s2">"NGram"</span><span class="p">:</span> <span class="n">NGramDecodingConfig</span><span class="p">,</span>
|
|
<span class="p">}</span>
|
|
|
|
<span class="n">config_class</span> <span class="o">=</span> <span class="n">config_classes</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="n">decoding_type</span><span class="p">)</span>
|
|
<span class="k">if</span> <span class="n">config_class</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="sa">f</span><span class="s2">"Invalid decoding type: </span><span class="si">{</span><span class="n">decoding_type</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span>
|
|
|
|
<span class="k">return</span> <span class="n">config_class</span><span class="p">(</span><span class="o">**</span><span class="n">data</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_check_fields</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">pass</span>
|
|
|
|
|
|
<div class="viewcode-block" id="MedusaDecodingConfig">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.MedusaDecodingConfig">[docs]</a>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">MedusaDecodingConfig</span><span class="p">(</span><span class="n">DecodingBaseConfig</span><span class="p">):</span>
|
|
<span class="n">medusa_choices</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]]]</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="n">num_medusa_heads</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<div class="viewcode-block" id="MedusaDecodingConfig.from_dict">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.MedusaDecodingConfig.from_dict">[docs]</a>
|
|
<span class="nd">@classmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">from_dict</span><span class="p">(</span><span class="bp">cls</span><span class="p">,</span> <span class="n">data</span><span class="p">:</span> <span class="nb">dict</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="bp">cls</span><span class="p">(</span><span class="o">**</span><span class="n">data</span><span class="p">)</span></div>
|
|
|
|
|
|
<span class="n">decoding_type</span><span class="p">:</span> <span class="n">ClassVar</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="s2">"Medusa"</span></div>
|
|
|
|
|
|
|
|
<div class="viewcode-block" id="EagleDecodingConfig">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.EagleDecodingConfig">[docs]</a>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">EagleDecodingConfig</span><span class="p">(</span><span class="n">DecodingBaseConfig</span><span class="p">):</span>
|
|
<span class="n">eagle_choices</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]]]</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="n">greedy_sampling</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">bool</span><span class="p">]</span> <span class="o">=</span> <span class="kc">True</span>
|
|
<span class="n">posterior_threshold</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="kc">None</span>
|
|
<span class="n">use_dynamic_tree</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">bool</span><span class="p">]</span> <span class="o">=</span> <span class="kc">False</span>
|
|
<span class="n">dynamic_tree_max_topK</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="n">num_eagle_layers</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="n">max_non_leaves_per_layer</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="n">pytorch_eagle_weights_path</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="n">eagle3_one_model</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">bool</span><span class="p">]</span> <span class="o">=</span> <span class="kc">True</span>
|
|
|
|
<div class="viewcode-block" id="EagleDecodingConfig.from_dict">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.EagleDecodingConfig.from_dict">[docs]</a>
|
|
<span class="nd">@classmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">from_dict</span><span class="p">(</span><span class="bp">cls</span><span class="p">,</span> <span class="n">data</span><span class="p">:</span> <span class="nb">dict</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="bp">cls</span><span class="p">(</span><span class="o">**</span><span class="n">data</span><span class="p">)</span></div>
|
|
|
|
|
|
<span class="n">decoding_type</span><span class="p">:</span> <span class="n">ClassVar</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="s2">"Eagle"</span></div>
|
|
|
|
|
|
|
|
<div class="viewcode-block" id="NGramDecodingConfig">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.NGramDecodingConfig">[docs]</a>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">NGramDecodingConfig</span><span class="p">(</span><span class="n">DecodingBaseConfig</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">"""</span>
|
|
<span class="sd"> Configuration for NGram drafter speculative decoding.</span>
|
|
|
|
<span class="sd"> Arguments:</span>
|
|
<span class="sd"> prompt_lookup_num_tokens: int</span>
|
|
<span class="sd"> The length maximum of draft tokens (can be understood as length maximum of output draft tokens).</span>
|
|
|
|
<span class="sd"> max_matching_ngram_size: int</span>
|
|
<span class="sd"> The length maximum of searching tokens (can be understood as length maximum of input tokens to search).</span>
|
|
|
|
<span class="sd"> is_keep_all: bool = True</span>
|
|
<span class="sd"> Whether to keep all candidate pattern-matches pairs, only one match is kept for each pattern if False.</span>
|
|
|
|
<span class="sd"> is_use_oldest: bool = True</span>
|
|
<span class="sd"> Whether to provide the oldest match when pattern is hit, the newest one is provided if False.</span>
|
|
|
|
<span class="sd"> is_public_pool: bool = True</span>
|
|
<span class="sd"> Whether to use a common pool for all requests, or the pool is private for each request if False.</span>
|
|
<span class="sd"> """</span>
|
|
|
|
<span class="n">prompt_lookup_num_tokens</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">2</span>
|
|
<span class="n">max_matching_ngram_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">4</span>
|
|
<span class="n">is_keep_all</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span>
|
|
<span class="n">is_use_oldest</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span>
|
|
<span class="n">is_public_pool</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span>
|
|
|
|
<div class="viewcode-block" id="NGramDecodingConfig.from_dict">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.NGramDecodingConfig.from_dict">[docs]</a>
|
|
<span class="nd">@classmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">from_dict</span><span class="p">(</span><span class="bp">cls</span><span class="p">,</span> <span class="n">data</span><span class="p">:</span> <span class="nb">dict</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="bp">cls</span><span class="p">(</span><span class="o">**</span><span class="n">data</span><span class="p">)</span></div>
|
|
|
|
|
|
<span class="n">decoding_type</span><span class="p">:</span> <span class="n">ClassVar</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="s2">"NGram"</span></div>
|
|
|
|
|
|
|
|
<div class="viewcode-block" id="MTPDecodingConfig">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.MTPDecodingConfig">[docs]</a>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">MTPDecodingConfig</span><span class="p">(</span><span class="n">DecodingBaseConfig</span><span class="p">):</span>
|
|
<span class="n">num_nextn_predict_layers</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="mi">1</span>
|
|
<span class="n">use_relaxed_acceptance_for_thinking</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">bool</span><span class="p">]</span> <span class="o">=</span> <span class="kc">False</span>
|
|
<span class="n">relaxed_topk</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="mi">1</span>
|
|
<span class="n">relaxed_delta</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="mf">0.</span>
|
|
|
|
<div class="viewcode-block" id="MTPDecodingConfig.from_dict">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.MTPDecodingConfig.from_dict">[docs]</a>
|
|
<span class="nd">@classmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">from_dict</span><span class="p">(</span><span class="bp">cls</span><span class="p">,</span> <span class="n">data</span><span class="p">:</span> <span class="nb">dict</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="bp">cls</span><span class="p">(</span><span class="o">**</span><span class="n">data</span><span class="p">)</span></div>
|
|
|
|
|
|
<span class="n">decoding_type</span><span class="p">:</span> <span class="n">ClassVar</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="s2">"MTP"</span></div>
|
|
|
|
|
|
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">PybindMirror</span><span class="p">(</span><span class="n">ABC</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">''' A class containing the utilities for mirroring Python classes to</span>
|
|
<span class="sd"> pybinding classes.</span>
|
|
<span class="sd"> '''</span>
|
|
|
|
<span class="nd">@abstractmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_to_pybind</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">pass</span>
|
|
|
|
<span class="nd">@staticmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">maybe_to_pybind</span><span class="p">(</span><span class="n">ins</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span>
|
|
<span class="n">ins</span><span class="p">,</span>
|
|
<span class="n">PybindMirror</span><span class="p">)</span> <span class="ow">or</span> <span class="nb">type</span><span class="p">(</span><span class="n">ins</span><span class="p">)</span><span class="o">.</span><span class="vm">__class__</span> <span class="o">==</span> <span class="n">PybindMirrorEnumMeta</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="n">ins</span><span class="o">.</span><span class="n">_to_pybind</span><span class="p">()</span>
|
|
<span class="k">return</span> <span class="n">ins</span>
|
|
|
|
<span class="nd">@staticmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">mirror_pybind_fields</span><span class="p">(</span><span class="n">pybind_class</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">"""</span>
|
|
<span class="sd"> Class decorator that ensures Python class fields mirror those of a C++ class.</span>
|
|
|
|
<span class="sd"> Args:</span>
|
|
<span class="sd"> pybind_class: The C++ class whose fields should be mirrored</span>
|
|
|
|
<span class="sd"> Returns:</span>
|
|
<span class="sd"> A decorator function that validates field mirroring</span>
|
|
<span class="sd"> """</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">decorator</span><span class="p">(</span><span class="bp">cls</span><span class="p">):</span>
|
|
<span class="k">assert</span> <span class="nb">issubclass</span><span class="p">(</span><span class="bp">cls</span><span class="p">,</span> <span class="n">BaseModel</span><span class="p">)</span>
|
|
<span class="c1"># Get all non-private fields from the C++ class</span>
|
|
<span class="n">cpp_fields</span> <span class="o">=</span> <span class="n">PybindMirror</span><span class="o">.</span><span class="n">get_pybind_variable_fields</span><span class="p">(</span><span class="n">pybind_class</span><span class="p">)</span>
|
|
<span class="n">python_fields</span> <span class="o">=</span> <span class="nb">set</span><span class="p">(</span><span class="bp">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="c1"># Check if all C++ fields exist in the Python class</span>
|
|
<span class="k">for</span> <span class="n">field</span> <span class="ow">in</span> <span class="n">cpp_fields</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="n">field</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">python_fields</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">"Field </span><span class="si">{</span><span class="n">field</span><span class="si">}</span><span class="s2"> is not mirrored in Python class </span><span class="si">{</span><span class="bp">cls</span><span class="o">.</span><span class="vm">__name__</span><span class="si">}</span><span class="s2"> from C++ class </span><span class="si">{</span><span class="n">pybind_class</span><span class="o">.</span><span class="vm">__name__</span><span class="si">}</span><span class="s2">. Please update the class."</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="c1"># Return the original class</span>
|
|
<span class="k">return</span> <span class="bp">cls</span>
|
|
|
|
<span class="k">return</span> <span class="n">decorator</span>
|
|
|
|
<span class="nd">@staticmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">get_pybind_enum_fields</span><span class="p">(</span><span class="n">pybind_class</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">''' Get all the enum fields from the pybind class. '''</span>
|
|
<span class="k">return</span> <span class="p">[</span>
|
|
<span class="n">f</span> <span class="k">for</span> <span class="n">f</span> <span class="ow">in</span> <span class="n">pybind_class</span><span class="o">.</span><span class="n">__members__</span><span class="o">.</span><span class="n">keys</span><span class="p">()</span>
|
|
<span class="k">if</span> <span class="ow">not</span> <span class="n">f</span><span class="o">.</span><span class="n">startswith</span><span class="p">(</span><span class="s1">'_'</span><span class="p">)</span> <span class="ow">and</span> <span class="ow">not</span> <span class="nb">callable</span><span class="p">(</span><span class="nb">getattr</span><span class="p">(</span><span class="n">pybind_class</span><span class="p">,</span> <span class="n">f</span><span class="p">))</span>
|
|
<span class="p">]</span>
|
|
|
|
<span class="nd">@staticmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">mirror_pybind_enum</span><span class="p">(</span><span class="n">pybind_class</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">''' Mirror the enum fields from the pybind class to the Python class. '''</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">decorator</span><span class="p">(</span><span class="bp">cls</span><span class="p">):</span>
|
|
<span class="k">assert</span> <span class="nb">issubclass</span><span class="p">(</span><span class="bp">cls</span><span class="p">,</span> <span class="n">Enum</span><span class="p">)</span>
|
|
<span class="n">cpp_fields</span> <span class="o">=</span> <span class="n">PybindMirror</span><span class="o">.</span><span class="n">get_pybind_enum_fields</span><span class="p">(</span><span class="n">pybind_class</span><span class="p">)</span>
|
|
<span class="n">python_fields</span> <span class="o">=</span> <span class="nb">set</span><span class="p">(</span><span class="bp">cls</span><span class="o">.</span><span class="n">__members__</span><span class="o">.</span><span class="n">keys</span><span class="p">())</span>
|
|
|
|
<span class="k">for</span> <span class="n">field</span> <span class="ow">in</span> <span class="n">cpp_fields</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="n">field</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">python_fields</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">"Field </span><span class="si">{</span><span class="n">field</span><span class="si">}</span><span class="s2"> is not mirrored in Python class </span><span class="si">{</span><span class="bp">cls</span><span class="o">.</span><span class="vm">__name__</span><span class="si">}</span><span class="s2"> from C++ class </span><span class="si">{</span><span class="n">pybind_class</span><span class="o">.</span><span class="vm">__name__</span><span class="si">}</span><span class="s2">. Please update the class."</span>
|
|
<span class="p">)</span>
|
|
<span class="k">return</span> <span class="bp">cls</span>
|
|
|
|
<span class="k">return</span> <span class="n">decorator</span>
|
|
|
|
<span class="nd">@staticmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">get_pybind_variable_fields</span><span class="p">(</span><span class="n">config_cls</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">''' Get all the variable fields from the pybind class. '''</span>
|
|
<span class="k">return</span> <span class="p">[</span>
|
|
<span class="n">f</span> <span class="k">for</span> <span class="n">f</span> <span class="ow">in</span> <span class="nb">dir</span><span class="p">(</span><span class="n">config_cls</span><span class="p">)</span>
|
|
<span class="k">if</span> <span class="ow">not</span> <span class="n">f</span><span class="o">.</span><span class="n">startswith</span><span class="p">(</span><span class="s1">'_'</span><span class="p">)</span> <span class="ow">and</span> <span class="ow">not</span> <span class="nb">callable</span><span class="p">(</span><span class="nb">getattr</span><span class="p">(</span><span class="n">config_cls</span><span class="p">,</span> <span class="n">f</span><span class="p">))</span>
|
|
<span class="p">]</span>
|
|
|
|
<span class="nd">@staticmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">pybind_equals</span><span class="p">(</span><span class="n">obj0</span><span class="p">,</span> <span class="n">obj1</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">''' Check if two pybind objects are equal. '''</span>
|
|
<span class="k">assert</span> <span class="nb">type</span><span class="p">(</span><span class="n">obj0</span><span class="p">)</span> <span class="ow">is</span> <span class="nb">type</span><span class="p">(</span><span class="n">obj1</span><span class="p">)</span>
|
|
<span class="k">for</span> <span class="n">field</span> <span class="ow">in</span> <span class="n">PybindMirror</span><span class="o">.</span><span class="n">get_pybind_variable_fields</span><span class="p">(</span><span class="nb">type</span><span class="p">(</span><span class="n">obj0</span><span class="p">)):</span>
|
|
<span class="k">if</span> <span class="nb">getattr</span><span class="p">(</span><span class="n">obj0</span><span class="p">,</span> <span class="n">field</span><span class="p">)</span> <span class="o">!=</span> <span class="nb">getattr</span><span class="p">(</span><span class="n">obj1</span><span class="p">,</span> <span class="n">field</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="kc">False</span>
|
|
<span class="k">return</span> <span class="kc">True</span>
|
|
|
|
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">PybindMirrorMeta</span><span class="p">(</span><span class="nb">type</span><span class="p">(</span><span class="n">PybindMirror</span><span class="p">)):</span>
|
|
<span class="k">pass</span>
|
|
|
|
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">PybindMirrorEnumMeta</span><span class="p">(</span><span class="n">EnumMeta</span><span class="p">,</span> <span class="n">PybindMirrorMeta</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">"""</span>
|
|
<span class="sd"> Combined metaclass for Enum and PybindMirror. This is crucial.</span>
|
|
<span class="sd"> """</span>
|
|
|
|
|
|
<div class="viewcode-block" id="BatchingType">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.BatchingType">[docs]</a>
|
|
<span class="nd">@PybindMirror</span><span class="o">.</span><span class="n">mirror_pybind_enum</span><span class="p">(</span><span class="n">_BatchingType</span><span class="p">)</span>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">BatchingType</span><span class="p">(</span><span class="n">StrEnum</span><span class="p">,</span> <span class="n">metaclass</span><span class="o">=</span><span class="n">PybindMirrorEnumMeta</span><span class="p">):</span>
|
|
<span class="n">STATIC</span> <span class="o">=</span> <span class="s2">"STATIC"</span>
|
|
<span class="n">INFLIGHT</span> <span class="o">=</span> <span class="s2">"INFLIGHT"</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_to_pybind</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="nb">getattr</span><span class="p">(</span><span class="n">_BatchingType</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">value</span><span class="p">)</span></div>
|
|
|
|
|
|
|
|
<div class="viewcode-block" id="CapacitySchedulerPolicy">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.CapacitySchedulerPolicy">[docs]</a>
|
|
<span class="nd">@PybindMirror</span><span class="o">.</span><span class="n">mirror_pybind_enum</span><span class="p">(</span><span class="n">_CapacitySchedulerPolicy</span><span class="p">)</span>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">CapacitySchedulerPolicy</span><span class="p">(</span><span class="n">StrEnum</span><span class="p">,</span> <span class="n">metaclass</span><span class="o">=</span><span class="n">PybindMirrorEnumMeta</span><span class="p">):</span>
|
|
<span class="n">MAX_UTILIZATION</span> <span class="o">=</span> <span class="s2">"MAX_UTILIZATION"</span>
|
|
<span class="n">GUARANTEED_NO_EVICT</span> <span class="o">=</span> <span class="s2">"GUARANTEED_NO_EVICT"</span>
|
|
<span class="n">STATIC_BATCH</span> <span class="o">=</span> <span class="s2">"STATIC_BATCH"</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_to_pybind</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="nb">getattr</span><span class="p">(</span><span class="n">_CapacitySchedulerPolicy</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">value</span><span class="p">)</span></div>
|
|
|
|
|
|
|
|
<div class="viewcode-block" id="ContextChunkingPolicy">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.ContextChunkingPolicy">[docs]</a>
|
|
<span class="nd">@PybindMirror</span><span class="o">.</span><span class="n">mirror_pybind_enum</span><span class="p">(</span><span class="n">_ContextChunkingPolicy</span><span class="p">)</span>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">ContextChunkingPolicy</span><span class="p">(</span><span class="n">StrEnum</span><span class="p">,</span> <span class="n">metaclass</span><span class="o">=</span><span class="n">PybindMirrorEnumMeta</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">''' Context chunking policy. '''</span>
|
|
<span class="n">FIRST_COME_FIRST_SERVED</span> <span class="o">=</span> <span class="s2">"FIRST_COME_FIRST_SERVED"</span>
|
|
<span class="n">EQUAL_PROGRESS</span> <span class="o">=</span> <span class="s2">"EQUAL_PROGRESS"</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_to_pybind</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="nb">getattr</span><span class="p">(</span><span class="n">_ContextChunkingPolicy</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">value</span><span class="p">)</span></div>
|
|
|
|
|
|
|
|
<div class="viewcode-block" id="DynamicBatchConfig">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.DynamicBatchConfig">[docs]</a>
|
|
<span class="nd">@PybindMirror</span><span class="o">.</span><span class="n">mirror_pybind_fields</span><span class="p">(</span><span class="n">_DynamicBatchConfig</span><span class="p">)</span>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">DynamicBatchConfig</span><span class="p">(</span><span class="n">BaseModel</span><span class="p">,</span> <span class="n">PybindMirror</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">"""Dynamic batch configuration.</span>
|
|
|
|
<span class="sd"> Controls how batch size and token limits are dynamically adjusted at runtime.</span>
|
|
<span class="sd"> """</span>
|
|
<span class="n">enable_batch_size_tuning</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Controls if the batch size should be tuned dynamically"</span><span class="p">)</span>
|
|
|
|
<span class="n">enable_max_num_tokens_tuning</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Controls if the max num tokens should be tuned dynamically"</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="n">dynamic_batch_moving_average_window</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"The window size for moving average of input and output length which is used to calculate dynamic batch size and max num tokens"</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_to_pybind</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="n">_DynamicBatchConfig</span><span class="p">(</span>
|
|
<span class="n">enable_batch_size_tuning</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">enable_batch_size_tuning</span><span class="p">,</span>
|
|
<span class="n">enable_max_num_tokens_tuning</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">enable_max_num_tokens_tuning</span><span class="p">,</span>
|
|
<span class="n">dynamic_batch_moving_average_window</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span>
|
|
<span class="n">dynamic_batch_moving_average_window</span><span class="p">)</span></div>
|
|
|
|
|
|
|
|
<div class="viewcode-block" id="SchedulerConfig">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.SchedulerConfig">[docs]</a>
|
|
<span class="nd">@PybindMirror</span><span class="o">.</span><span class="n">mirror_pybind_fields</span><span class="p">(</span><span class="n">_SchedulerConfig</span><span class="p">)</span>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">SchedulerConfig</span><span class="p">(</span><span class="n">BaseModel</span><span class="p">,</span> <span class="n">PybindMirror</span><span class="p">):</span>
|
|
<span class="n">capacity_scheduler_policy</span><span class="p">:</span> <span class="n">CapacitySchedulerPolicy</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="n">CapacitySchedulerPolicy</span><span class="o">.</span><span class="n">GUARANTEED_NO_EVICT</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The capacity scheduler policy to use"</span><span class="p">)</span>
|
|
|
|
<span class="n">context_chunking_policy</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">ContextChunkingPolicy</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"The context chunking policy to use"</span><span class="p">)</span>
|
|
|
|
<span class="n">dynamic_batch_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">DynamicBatchConfig</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"The dynamic batch config to use"</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_to_pybind</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="n">_SchedulerConfig</span><span class="p">(</span>
|
|
<span class="n">capacity_scheduler_policy</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">capacity_scheduler_policy</span><span class="o">.</span><span class="n">_to_pybind</span><span class="p">(</span>
|
|
<span class="p">),</span>
|
|
<span class="n">context_chunking_policy</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">context_chunking_policy</span><span class="o">.</span><span class="n">_to_pybind</span><span class="p">()</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">context_chunking_policy</span> <span class="k">else</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">dynamic_batch_config</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">dynamic_batch_config</span><span class="o">.</span><span class="n">_to_pybind</span><span class="p">()</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">dynamic_batch_config</span> <span class="k">else</span> <span class="kc">None</span><span class="p">)</span></div>
|
|
|
|
|
|
|
|
<span class="nd">@PybindMirror</span><span class="o">.</span><span class="n">mirror_pybind_fields</span><span class="p">(</span><span class="n">_PeftCacheConfig</span><span class="p">)</span>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">PeftCacheConfig</span><span class="p">(</span><span class="n">BaseModel</span><span class="p">,</span> <span class="n">PybindMirror</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">"""</span>
|
|
<span class="sd"> Configuration for the PEFT cache.</span>
|
|
<span class="sd"> """</span>
|
|
<span class="n">num_host_module_layer</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"number of max sized 1-layer 1-module adapterSize=1 sets of weights that can be stored in host cache"</span>
|
|
<span class="p">)</span>
|
|
<span class="n">num_device_module_layer</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"number of max sized 1-layer 1-module sets of weights that can be stored in host cache"</span>
|
|
<span class="p">)</span>
|
|
<span class="n">optimal_adapter_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span>
|
|
<span class="mi">8</span><span class="p">,</span> <span class="c1"># There are tests to keep the default value consistent with the pybind default value</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"optimal adapter size used to set page width"</span><span class="p">)</span>
|
|
<span class="n">max_adapter_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">64</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"max supported adapter size. Used to compute minimum"</span><span class="p">)</span>
|
|
<span class="n">num_put_workers</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"number of worker threads used to put weights into host cache"</span><span class="p">)</span>
|
|
<span class="n">num_ensure_workers</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"number of worker threads used to copy weights from host to device"</span><span class="p">)</span>
|
|
<span class="n">num_copy_streams</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"number of streams used to copy weights from host to device"</span>
|
|
<span class="p">)</span>
|
|
<span class="n">max_pages_per_block_host</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">24</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Number of cache pages per allocation block (host)"</span><span class="p">)</span>
|
|
<span class="n">max_pages_per_block_device</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">8</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Number of cache pages per allocation block (device)"</span><span class="p">)</span>
|
|
<span class="n">device_cache_percent</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="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"percent of memory after engine load to use for cache"</span><span class="p">)</span>
|
|
<span class="n">host_cache_size</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"size in bytes to use for host cache"</span><span class="p">)</span>
|
|
<span class="n">lora_prefetch_dir</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="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"folder to store the LoRA weights we hope to load during engine initialization"</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_to_pybind</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="n">_PeftCacheConfig</span><span class="p">(</span>
|
|
<span class="n">num_host_module_layer</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">num_host_module_layer</span><span class="p">,</span>
|
|
<span class="n">num_device_module_layer</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">num_device_module_layer</span><span class="p">,</span>
|
|
<span class="n">optimal_adapter_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">optimal_adapter_size</span><span class="p">,</span>
|
|
<span class="n">max_adapter_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">max_adapter_size</span><span class="p">,</span>
|
|
<span class="n">num_put_workers</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">num_put_workers</span><span class="p">,</span>
|
|
<span class="n">num_ensure_workers</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">num_ensure_workers</span><span class="p">,</span>
|
|
<span class="n">num_copy_streams</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">num_copy_streams</span><span class="p">,</span>
|
|
<span class="n">max_pages_per_block_host</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">max_pages_per_block_host</span><span class="p">,</span>
|
|
<span class="n">max_pages_per_block_device</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">max_pages_per_block_device</span><span class="p">,</span>
|
|
<span class="n">device_cache_percent</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">device_cache_percent</span><span class="p">,</span>
|
|
<span class="n">host_cache_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">host_cache_size</span><span class="p">,</span>
|
|
<span class="n">lora_prefetch_dir</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">lora_prefetch_dir</span><span class="p">)</span>
|
|
|
|
|
|
<div class="viewcode-block" id="LookaheadDecodingConfig">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.LookaheadDecodingConfig">[docs]</a>
|
|
<span class="nd">@PybindMirror</span><span class="o">.</span><span class="n">mirror_pybind_fields</span><span class="p">(</span><span class="n">_LookaheadDecodingConfig</span><span class="p">)</span>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">LookaheadDecodingConfig</span><span class="p">(</span><span class="n">DecodingBaseConfig</span><span class="p">,</span> <span class="n">PybindMirror</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">"""</span>
|
|
<span class="sd"> Configuration for lookahead speculative decoding.</span>
|
|
<span class="sd"> """</span>
|
|
|
|
<span class="n">max_window_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="n">_LookaheadDecodingConfig</span><span class="o">.</span><span class="n">get_default_lookahead_decoding_window</span><span class="p">(</span>
|
|
<span class="p">),</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Number of NGrams in lookahead branch per step."</span><span class="p">)</span>
|
|
<span class="n">max_ngram_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="n">_LookaheadDecodingConfig</span><span class="o">.</span><span class="n">get_default_lookahead_decoding_ngram</span><span class="p">(),</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Number of tokens per NGram."</span><span class="p">)</span>
|
|
<span class="n">max_verification_set_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="n">_LookaheadDecodingConfig</span><span class="o">.</span>
|
|
<span class="n">get_default_lookahead_decoding_verification_set</span><span class="p">(),</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Number of NGrams in verification branch per step."</span><span class="p">)</span>
|
|
|
|
<div class="viewcode-block" id="LookaheadDecodingConfig.validate_positive_values">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.LookaheadDecodingConfig.validate_positive_values">[docs]</a>
|
|
<span class="nd">@field_validator</span><span class="p">(</span><span class="s1">'max_window_size'</span><span class="p">,</span> <span class="s1">'max_ngram_size'</span><span class="p">,</span>
|
|
<span class="s1">'max_verification_set_size'</span><span class="p">)</span>
|
|
<span class="nd">@classmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">validate_positive_values</span><span class="p">(</span><span class="bp">cls</span><span class="p">,</span> <span class="n">v</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="n">v</span> <span class="o"><=</span> <span class="mi">0</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">"Value must be positive, got </span><span class="si">{</span><span class="n">v</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="n">v</span></div>
|
|
|
|
|
|
<div class="viewcode-block" id="LookaheadDecodingConfig.__init__">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.LookaheadDecodingConfig.__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="n">data</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="o">**</span><span class="n">data</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_check_fields</span><span class="p">()</span></div>
|
|
|
|
|
|
<div class="viewcode-block" id="LookaheadDecodingConfig.calculate_speculative_resource">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.LookaheadDecodingConfig.calculate_speculative_resource">[docs]</a>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">calculate_speculative_resource</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="n">_LookaheadDecodingConfig</span><span class="o">.</span><span class="n">calculate_speculative_resource_tuple</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">max_window_size</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_ngram_size</span><span class="p">,</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">max_verification_set_size</span><span class="p">)</span></div>
|
|
|
|
|
|
<div class="viewcode-block" id="LookaheadDecodingConfig.from_dict">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.LookaheadDecodingConfig.from_dict">[docs]</a>
|
|
<span class="nd">@classmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">from_dict</span><span class="p">(</span><span class="bp">cls</span><span class="p">,</span> <span class="n">data</span><span class="p">:</span> <span class="nb">dict</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="bp">cls</span><span class="p">(</span><span class="o">**</span><span class="n">data</span><span class="p">)</span></div>
|
|
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_to_pybind</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="n">_LookaheadDecodingConfig</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">max_window_size</span><span class="p">,</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">max_ngram_size</span><span class="p">,</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">max_verification_set_size</span><span class="p">)</span>
|
|
|
|
<span class="n">decoding_type</span><span class="p">:</span> <span class="n">ClassVar</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="s2">"Lookahead"</span></div>
|
|
|
|
|
|
|
|
<div class="viewcode-block" id="KvCacheConfig">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.KvCacheConfig">[docs]</a>
|
|
<span class="nd">@PybindMirror</span><span class="o">.</span><span class="n">mirror_pybind_fields</span><span class="p">(</span><span class="n">_KvCacheConfig</span><span class="p">)</span>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">KvCacheConfig</span><span class="p">(</span><span class="n">BaseModel</span><span class="p">,</span> <span class="n">PybindMirror</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">"""</span>
|
|
<span class="sd"> Configuration for the KV cache.</span>
|
|
<span class="sd"> """</span>
|
|
<span class="n">enable_block_reuse</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"Controls if KV cache blocks can be reused for different requests."</span><span class="p">)</span>
|
|
<span class="n">max_tokens</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"The maximum number of tokens that should be stored in the KV cache. If both `max_tokens` and `free_gpu_memory_fraction` are specified, memory corresponding to the minimum will be used."</span>
|
|
<span class="p">)</span>
|
|
<span class="n">max_attention_window</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"Size of the attention window for each sequence. Only the last tokens will be stored in the KV cache. If the number of elements in `max_attention_window` is less than the number of layers, `max_attention_window` will be repeated multiple times to the number of layers."</span>
|
|
<span class="p">)</span>
|
|
<span class="n">sink_token_length</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"Number of sink tokens (tokens to always keep in attention window)."</span><span class="p">)</span>
|
|
<span class="n">free_gpu_memory_fraction</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="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"The fraction of GPU memory fraction that should be allocated for the KV cache. Default is 90%. If both `max_tokens` and `free_gpu_memory_fraction` are specified, memory corresponding to the minimum will be used."</span>
|
|
<span class="p">)</span>
|
|
<span class="n">host_cache_size</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"Size of the host cache in bytes. If both `max_tokens` and `host_cache_size` are specified, memory corresponding to the minimum will be used."</span>
|
|
<span class="p">)</span>
|
|
<span class="n">onboard_blocks</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Controls if blocks are onboarded."</span><span class="p">)</span>
|
|
<span class="n">cross_kv_cache_fraction</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="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"The fraction of the KV Cache memory should be reserved for cross attention. If set to p, self attention will use 1-p of KV Cache memory and cross attention will use p of KV Cache memory. Default is 50%. Should only be set when using encoder-decoder model."</span>
|
|
<span class="p">)</span>
|
|
<span class="n">secondary_offload_min_priority</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"Only blocks with priority > mSecondaryOfflineMinPriority can be offloaded to secondary memory."</span>
|
|
<span class="p">)</span>
|
|
<span class="n">event_buffer_max_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"Maximum size of the event buffer. If set to 0, the event buffer will not be used."</span>
|
|
<span class="p">)</span>
|
|
<span class="n">enable_partial_reuse</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"Whether blocks that are only partially matched can be reused."</span><span class="p">)</span>
|
|
<span class="n">copy_on_partial_reuse</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"Whether partially matched blocks that are in use can be reused after copying them."</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_to_pybind</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="n">_KvCacheConfig</span><span class="p">(</span>
|
|
<span class="n">enable_block_reuse</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">enable_block_reuse</span><span class="p">,</span>
|
|
<span class="n">max_tokens</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">max_tokens</span><span class="p">,</span>
|
|
<span class="n">max_attention_window</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">max_attention_window</span><span class="p">,</span>
|
|
<span class="n">sink_token_length</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">sink_token_length</span><span class="p">,</span>
|
|
<span class="n">free_gpu_memory_fraction</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">free_gpu_memory_fraction</span><span class="p">,</span>
|
|
<span class="n">host_cache_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">host_cache_size</span><span class="p">,</span>
|
|
<span class="n">onboard_blocks</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">onboard_blocks</span><span class="p">,</span>
|
|
<span class="n">cross_kv_cache_fraction</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">cross_kv_cache_fraction</span><span class="p">,</span>
|
|
<span class="n">secondary_offload_min_priority</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">secondary_offload_min_priority</span><span class="p">,</span>
|
|
<span class="n">event_buffer_max_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">event_buffer_max_size</span><span class="p">,</span>
|
|
<span class="n">enable_partial_reuse</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">enable_partial_reuse</span><span class="p">,</span>
|
|
<span class="n">copy_on_partial_reuse</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">copy_on_partial_reuse</span><span class="p">)</span></div>
|
|
|
|
|
|
|
|
<div class="viewcode-block" id="ExtendedRuntimePerfKnobConfig">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.ExtendedRuntimePerfKnobConfig">[docs]</a>
|
|
<span class="nd">@PybindMirror</span><span class="o">.</span><span class="n">mirror_pybind_fields</span><span class="p">(</span><span class="n">_ExtendedRuntimePerfKnobConfig</span><span class="p">)</span>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">ExtendedRuntimePerfKnobConfig</span><span class="p">(</span><span class="n">BaseModel</span><span class="p">,</span> <span class="n">PybindMirror</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">"""</span>
|
|
<span class="sd"> Configuration for extended runtime performance knobs.</span>
|
|
<span class="sd"> """</span>
|
|
|
|
<span class="n">multi_block_mode</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Whether to use multi-block mode."</span><span class="p">)</span>
|
|
|
|
<span class="n">enable_context_fmha_fp32_acc</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Whether to enable context FMHA FP32 accumulation."</span><span class="p">)</span>
|
|
|
|
<span class="n">cuda_graph_mode</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Whether to use CUDA graph mode."</span><span class="p">)</span>
|
|
|
|
<span class="n">cuda_graph_cache_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"Number of cuda graphs to be cached in the runtime. The larger the cache, the better the perf, but more GPU memory is consumed."</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_to_pybind</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="n">res</span> <span class="o">=</span> <span class="n">_ExtendedRuntimePerfKnobConfig</span><span class="p">(</span>
|
|
<span class="n">multi_block_mode</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">multi_block_mode</span><span class="p">,</span>
|
|
<span class="n">enable_context_fmha_fp32_acc</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">enable_context_fmha_fp32_acc</span><span class="p">)</span>
|
|
<span class="n">res</span><span class="o">.</span><span class="n">cuda_graph_mode</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">cuda_graph_mode</span>
|
|
<span class="n">res</span><span class="o">.</span><span class="n">cuda_graph_cache_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">cuda_graph_cache_size</span>
|
|
<span class="k">return</span> <span class="n">res</span></div>
|
|
|
|
|
|
|
|
<div class="viewcode-block" id="CacheTransceiverConfig">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.CacheTransceiverConfig">[docs]</a>
|
|
<span class="nd">@PybindMirror</span><span class="o">.</span><span class="n">mirror_pybind_fields</span><span class="p">(</span><span class="n">_CacheTransceiverConfig</span><span class="p">)</span>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">CacheTransceiverConfig</span><span class="p">(</span><span class="n">BaseModel</span><span class="p">,</span> <span class="n">PybindMirror</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">"""</span>
|
|
<span class="sd"> Configuration for the cache transceiver.</span>
|
|
<span class="sd"> """</span>
|
|
<span class="n">max_num_tokens</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The max number of tokens the transfer buffer can fit."</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_to_pybind</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="n">_CacheTransceiverConfig</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">max_num_tokens</span><span class="p">)</span></div>
|
|
|
|
|
|
|
|
<span class="nd">@dataclass</span>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">_ModelWrapper</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="k">def</span><span class="w"> </span><span class="nf">__post_init__</span><span class="p">(</span><span class="bp">self</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">model</span><span class="p">:</span>
|
|
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s2">"model should be provided."</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">model</span><span class="p">,</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="sa">f</span><span class="s2">"Invalid model: </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">model</span><span class="si">}</span><span class="s2">"</span>
|
|
|
|
<span class="n">model_dir</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">model</span><span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="n">model_dir</span><span class="o">.</span><span class="n">exists</span><span class="p">()</span> <span class="ow">and</span> <span class="n">model_dir</span><span class="o">.</span><span class="n">is_dir</span><span class="p">():</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">model</span> <span class="o">=</span> <span class="n">model_dir</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">is_hub_model</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="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">is_local_model</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">is_local_model</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">model</span><span class="p">,</span> <span class="n">Path</span><span class="p">)</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">model_dir</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">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">is_local_model</span><span class="p">,</span> <span class="sa">f</span><span class="s2">"model_dir is only available for local model, </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">model</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">model</span>
|
|
|
|
<span class="nd">@model_dir</span><span class="o">.</span><span class="n">setter</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">model_dir</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">model_dir</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">model_dir</span> <span class="o">=</span> <span class="n">Path</span><span class="p">(</span><span class="n">model_dir</span><span class="p">)</span>
|
|
<span class="k">assert</span> <span class="n">model_dir</span><span class="o">.</span><span class="n">exists</span><span class="p">()</span> <span class="ow">and</span> <span class="n">model_dir</span><span class="o">.</span><span class="n">is_dir</span><span class="p">(</span>
|
|
<span class="p">),</span> <span class="sa">f</span><span class="s2">"model_dir is not a valid path, </span><span class="si">{</span><span class="n">model_dir</span><span class="si">}</span><span class="s2">"</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">model</span> <span class="o">=</span> <span class="n">model_dir</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">model_name</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></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="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">model</span> <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">model</span><span class="p">,</span> <span class="nb">str</span><span class="p">)</span> <span class="k">else</span> <span class="kc">None</span>
|
|
|
|
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">BaseLlmArgs</span><span class="p">(</span><span class="n">BaseModel</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">"""</span>
|
|
<span class="sd"> Base class for both TorchLlmArgs and TrtLlmArgs. It contains all the arguments that are common to both.</span>
|
|
<span class="sd"> """</span>
|
|
<span class="n">model_config</span> <span class="o">=</span> <span class="p">{</span>
|
|
<span class="s2">"arbitrary_types_allowed"</span><span class="p">:</span> <span class="kc">True</span><span class="p">,</span>
|
|
<span class="s2">"extra"</span><span class="p">:</span> <span class="s2">"allow"</span><span class="p">,</span>
|
|
<span class="p">}</span>
|
|
|
|
<span class="c1"># Explicit arguments</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="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"The path to the model checkpoint or the model name from the Hugging Face Hub."</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="n">Field</span><span class="p">(</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"The path to the tokenizer checkpoint or the tokenizer name from the Hugging Face Hub."</span><span class="p">,</span>
|
|
<span class="n">default</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="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="s1">'auto'</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The mode to initialize the tokenizer."</span><span class="p">,</span>
|
|
<span class="n">json_schema_extra</span><span class="o">=</span><span class="p">{</span><span class="s2">"type"</span><span class="p">:</span> <span class="s2">"Literal['auto', 'slow']"</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="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Whether to skip the tokenizer initialization."</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="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Whether to trust the remote code."</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="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The tensor parallel size."</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="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="s2">"auto"</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The data type to use for the model."</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="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"The revision to use for the model."</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="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"The revision to use for the tokenizer."</span><span class="p">)</span>
|
|
|
|
<span class="c1"># Below are all remaining arguments</span>
|
|
|
|
<span class="n">pipeline_parallel_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"The pipeline parallel size."</span><span class="p">)</span>
|
|
|
|
<span class="n">context_parallel_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The context parallel size."</span><span class="p">)</span>
|
|
|
|
<span class="n">gpus_per_node</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"The number of GPUs per node."</span><span class="p">)</span>
|
|
|
|
<span class="n">moe_cluster_parallel_size</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The cluster parallel size for MoE models's expert weights."</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="n">moe_tensor_parallel_size</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The tensor parallel size for MoE models's expert weights."</span><span class="p">)</span>
|
|
|
|
<span class="n">moe_expert_parallel_size</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The expert parallel size for MoE models's expert weights."</span><span class="p">)</span>
|
|
|
|
<span class="n">enable_attention_dp</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Enable attention data parallel."</span><span class="p">)</span>
|
|
|
|
<span class="n">cp_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">dict</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default_factory</span><span class="o">=</span><span class="nb">dict</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Context parallel config."</span><span class="p">)</span>
|
|
|
|
<span class="n">load_format</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">'dummy'</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="s1">'auto'</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The format to load the model."</span><span class="p">,</span>
|
|
<span class="n">json_schema_extra</span><span class="o">=</span><span class="p">{</span><span class="s2">"type"</span><span class="p">:</span> <span class="s2">"Literal['auto', 'dummy']"</span><span class="p">})</span>
|
|
|
|
<span class="c1"># LoRA arguments</span>
|
|
<span class="n">enable_lora</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Enable LoRA."</span><span class="p">)</span>
|
|
|
|
<span class="n">max_lora_rank</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The maximum LoRA rank."</span><span class="p">,</span>
|
|
<span class="n">deprecated</span><span class="o">=</span><span class="s2">"Use lora_config.max_lora_rank instead."</span><span class="p">)</span>
|
|
|
|
<span class="n">max_loras</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="mi">4</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The maximum number of LoRA."</span><span class="p">,</span>
|
|
<span class="n">deprecated</span><span class="o">=</span><span class="s2">"Use lora_config.max_loras instead."</span><span class="p">)</span>
|
|
|
|
<span class="n">max_cpu_loras</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">4</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The maximum number of LoRA on CPU."</span><span class="p">,</span>
|
|
<span class="n">deprecated</span><span class="o">=</span><span class="s2">"Use lora_config.max_cpu_loras instead."</span><span class="p">)</span>
|
|
|
|
<span class="n">lora_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">LoraConfig</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"LoRA configuration for the model."</span><span class="p">)</span>
|
|
|
|
<span class="c1"># Prompt adapter arguments</span>
|
|
<span class="n">enable_prompt_adapter</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Enable prompt adapter."</span><span class="p">)</span>
|
|
|
|
<span class="n">max_prompt_adapter_token</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"The maximum number of prompt adapter tokens."</span><span class="p">)</span>
|
|
|
|
<span class="c1"># Quantization and calibration configurations</span>
|
|
<span class="n">quant_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">QuantConfig</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Quantization config."</span><span class="p">)</span>
|
|
|
|
<span class="c1"># Several options from ExecutorConfig, expanded here for less hierarchy</span>
|
|
<span class="n">kv_cache_config</span><span class="p">:</span> <span class="n">KvCacheConfig</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default_factory</span><span class="o">=</span><span class="n">KvCacheConfig</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"KV cache config."</span><span class="p">)</span>
|
|
|
|
<span class="n">enable_chunked_prefill</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Enable chunked prefill."</span><span class="p">)</span>
|
|
|
|
<span class="n">guided_decoding_backend</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="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Guided decoding backend."</span><span class="p">)</span>
|
|
|
|
<span class="n">batched_logits_processor</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">object</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Batched logits processor."</span><span class="p">,</span>
|
|
<span class="n">json_schema_extra</span><span class="o">=</span><span class="p">{</span>
|
|
<span class="s2">"type"</span><span class="p">:</span> <span class="sa">f</span><span class="s2">"Optional[</span><span class="si">{</span><span class="n">get_type_repr</span><span class="p">(</span><span class="n">BatchedLogitsProcessor</span><span class="p">)</span><span class="si">}</span><span class="s2">]"</span>
|
|
<span class="p">})</span>
|
|
|
|
<span class="n">iter_stats_max_iterations</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The maximum number of iterations for iter stats."</span><span class="p">)</span>
|
|
|
|
<span class="n">request_stats_max_iterations</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The maximum number of iterations for request stats."</span><span class="p">)</span>
|
|
|
|
<span class="c1"># A handful of options from PretrainedConfig</span>
|
|
<span class="n">peft_cache_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">PeftCacheConfig</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"PEFT cache config."</span><span class="p">)</span>
|
|
|
|
<span class="n">scheduler_config</span><span class="p">:</span> <span class="n">SchedulerConfig</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default_factory</span><span class="o">=</span><span class="n">SchedulerConfig</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Scheduler config."</span><span class="p">)</span>
|
|
|
|
<span class="n">cache_transceiver_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">CacheTransceiverConfig</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Cache transceiver config."</span><span class="p">)</span>
|
|
|
|
<span class="c1"># Speculative decoding parameters</span>
|
|
<span class="n">speculative_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">LookaheadDecodingConfig</span><span class="p">,</span> <span class="n">MedusaDecodingConfig</span><span class="p">,</span> <span class="n">EagleDecodingConfig</span><span class="p">,</span>
|
|
<span class="n">MTPDecodingConfig</span><span class="p">,</span> <span class="n">NGramDecodingConfig</span><span class="p">]]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Speculative decoding config."</span><span class="p">)</span>
|
|
|
|
<span class="n">batching_type</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">BatchingType</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Batching type."</span><span class="p">)</span>
|
|
|
|
<span class="n">normalize_log_probs</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Normalize log probabilities."</span><span class="p">)</span>
|
|
|
|
<span class="n">max_batch_size</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The maximum batch size."</span><span class="p">)</span>
|
|
|
|
<span class="c1"># generation constraints</span>
|
|
<span class="n">max_input_len</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="mi">1024</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The maximum input length."</span><span class="p">)</span>
|
|
|
|
<span class="n">max_seq_len</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"The maximum sequence length."</span><span class="p">)</span>
|
|
|
|
<span class="n">max_beam_width</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The maximum beam width."</span><span class="p">)</span>
|
|
|
|
<span class="n">max_num_tokens</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"The maximum number of tokens."</span><span class="p">)</span>
|
|
|
|
<span class="n">backend</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="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The backend to use."</span><span class="p">,</span>
|
|
<span class="n">exclude</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
|
|
|
|
<span class="n">gather_generation_logits</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Gather generation logits."</span><span class="p">)</span>
|
|
|
|
<span class="c1"># private fields those are unstable and just for internal use</span>
|
|
<span class="n">num_postprocess_workers</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The number of postprocess worker processes."</span><span class="p">,</span>
|
|
<span class="n">alias</span><span class="o">=</span><span class="s2">"_num_postprocess_workers"</span><span class="p">)</span>
|
|
|
|
<span class="n">postprocess_tokenizer_dir</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="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The postprocess tokenizer directory."</span><span class="p">,</span>
|
|
<span class="n">alias</span><span class="o">=</span><span class="s2">"_postprocess_tokenizer_dir"</span><span class="p">)</span>
|
|
|
|
<span class="n">reasoning_parser</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="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The parser to separate reasoning content from output."</span><span class="p">,</span>
|
|
<span class="n">alias</span><span class="o">=</span><span class="s2">"_reasoning_parser"</span><span class="p">)</span>
|
|
|
|
<span class="c1"># TODO[Superjomn]: To deprecate this config.</span>
|
|
<span class="n">decoding_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">object</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The decoding config."</span><span class="p">,</span>
|
|
<span class="n">json_schema_extra</span><span class="o">=</span><span class="p">{</span><span class="s2">"type"</span><span class="p">:</span> <span class="s2">"Optional[DecodingConfig]"</span><span class="p">},</span>
|
|
<span class="n">deprecated</span><span class="o">=</span><span class="s2">"Use speculative_config instead."</span><span class="p">,</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="n">mpi_session</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">object</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The optional MPI session to use for this LLM instance."</span><span class="p">,</span>
|
|
<span class="n">json_schema_extra</span><span class="o">=</span><span class="p">{</span><span class="s2">"type"</span><span class="p">:</span> <span class="s2">"Optional[MpiSession]"</span><span class="p">},</span>
|
|
<span class="n">exclude</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># exclude from serialization</span>
|
|
<span class="n">alias</span><span class="o">=</span><span class="s2">"_mpi_session"</span><span class="p">)</span>
|
|
|
|
<span class="nd">@print_traceback_on_error</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">model_post_init</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">__context</span><span class="p">:</span> <span class="n">Any</span><span class="p">):</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">skip_tokenizer_init</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="kc">None</span>
|
|
<span class="k">else</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_factory</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="n">trust_remote_code</span><span class="o">=</span><span class="bp">self</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">tokenizer_mode</span> <span class="o">!=</span> <span class="s1">'slow'</span><span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">get_device_properties</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span><span class="o">.</span><span class="n">major</span> <span class="o"><</span> <span class="mi">8</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">dtype</span> <span class="o">==</span> <span class="s1">'auto'</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">dtype</span> <span class="o">=</span> <span class="s1">'float16'</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">dtype</span> <span class="o">==</span> <span class="s1">'bfloat16'</span><span class="p">:</span>
|
|
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="s2">"Pre SM 80 GPUs do not support bfloat16"</span><span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">gpus_per_node</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span>
|
|
<span class="sa">f</span><span class="s2">"Using default gpus_per_node: </span><span class="si">{</span><span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">device_count</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">gpus_per_node</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">device_count</span><span class="p">()</span>
|
|
<span class="k">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">gpus_per_node</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">moe_cluster_parallel_size</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">moe_cluster_parallel_size</span> <span class="o">=</span> <span class="o">-</span><span class="mi">1</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">moe_tensor_parallel_size</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">moe_tensor_parallel_size</span> <span class="o">=</span> <span class="o">-</span><span class="mi">1</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">moe_expert_parallel_size</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">moe_expert_parallel_size</span> <span class="o">=</span> <span class="o">-</span><span class="mi">1</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">parallel_config</span> <span class="o">=</span> <span class="n">_ParallelConfig</span><span class="p">(</span>
|
|
<span class="n">tp_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">tensor_parallel_size</span><span class="p">,</span>
|
|
<span class="n">pp_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">pipeline_parallel_size</span><span class="p">,</span>
|
|
<span class="n">cp_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">context_parallel_size</span><span class="p">,</span>
|
|
<span class="n">gpus_per_node</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">gpus_per_node</span><span class="p">,</span>
|
|
<span class="n">moe_cluster_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">moe_cluster_parallel_size</span><span class="p">,</span>
|
|
<span class="n">moe_tp_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">moe_tensor_parallel_size</span><span class="p">,</span>
|
|
<span class="n">moe_ep_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">moe_expert_parallel_size</span><span class="p">,</span>
|
|
<span class="n">enable_attention_dp</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">enable_attention_dp</span><span class="p">,</span>
|
|
<span class="n">cp_config</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">cp_config</span><span class="p">)</span>
|
|
|
|
<span class="nd">@classmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">from_kwargs</span><span class="p">(</span><span class="bp">cls</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="s2">"BaseLlmArgs"</span><span class="p">:</span>
|
|
<span class="w"> </span><span class="sd">"""Create `LlmArgs` instance from kwargs.</span>
|
|
|
|
<span class="sd"> Args:</span>
|
|
<span class="sd"> kwargs (Any): Arguments passed to `LlmArgs` constructor.</span>
|
|
|
|
<span class="sd"> Returns:</span>
|
|
<span class="sd"> tensorrt_llm.llmapi.llm_utils.BaseLlmArgs: The `BaseLlmArgs` instance.</span>
|
|
<span class="sd"> """</span>
|
|
<span class="n">kwargs</span> <span class="o">=</span> <span class="n">BaseLlmArgs</span><span class="o">.</span><span class="n">_maybe_update_config_for_consistency</span><span class="p">(</span><span class="nb">dict</span><span class="p">(</span><span class="n">kwargs</span><span class="p">))</span>
|
|
<span class="n">ret</span> <span class="o">=</span> <span class="bp">cls</span><span class="p">(</span><span class="o">**</span><span class="n">kwargs</span><span class="p">)</span>
|
|
<span class="n">ret</span><span class="o">.</span><span class="n">_setup</span><span class="p">()</span>
|
|
<span class="k">return</span> <span class="n">ret</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">to_dict</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="nb">dict</span><span class="p">:</span>
|
|
<span class="w"> </span><span class="sd">"""Dump `LlmArgs` instance to a dict.</span>
|
|
|
|
<span class="sd"> Returns:</span>
|
|
<span class="sd"> dict: The dict that contains all fields of the `LlmArgs` instance.</span>
|
|
<span class="sd"> """</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">model_dump</span><span class="p">()</span>
|
|
|
|
<span class="nd">@staticmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_maybe_update_config_for_consistency</span><span class="p">(</span>
|
|
<span class="n">kwargs_dict</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Any</span><span class="p">])</span> <span class="o">-></span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Any</span><span class="p">]:</span>
|
|
<span class="c1"># max_beam_width is not included since vague behavior due to lacking the support for dynamic beam width during</span>
|
|
<span class="c1"># generation</span>
|
|
<span class="n">black_list</span> <span class="o">=</span> <span class="nb">set</span><span class="p">([</span><span class="s2">"max_beam_width"</span><span class="p">])</span>
|
|
<span class="n">executor_config_attrs</span> <span class="o">=</span> <span class="nb">set</span><span class="p">(</span>
|
|
<span class="n">attr</span> <span class="k">for</span> <span class="n">attr</span> <span class="ow">in</span> <span class="nb">dir</span><span class="p">(</span><span class="n">_ExecutorConfig</span><span class="p">)</span> <span class="k">if</span> <span class="ow">not</span> <span class="n">attr</span><span class="o">.</span><span class="n">startswith</span><span class="p">(</span><span class="s1">'_'</span><span class="p">)</span>
|
|
<span class="ow">and</span> <span class="nb">callable</span><span class="p">(</span><span class="nb">getattr</span><span class="p">(</span><span class="n">_ExecutorConfig</span><span class="p">,</span> <span class="n">attr</span><span class="p">)))</span>
|
|
<span class="n">executor_config_attrs</span> <span class="o">-=</span> <span class="n">black_list</span>
|
|
<span class="n">llm_args_attr</span> <span class="o">=</span> <span class="nb">set</span><span class="p">(</span><span class="n">BaseLlmArgs</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="c1"># NOTE: When cpp ExecutorConfig add new options, please add the new options into `LlmArgs` with docs as well</span>
|
|
<span class="c1"># ASK chunweiy for help if you are not sure about the new options.</span>
|
|
<span class="k">assert</span> <span class="n">executor_config_attrs</span><span class="o">.</span><span class="n">issubset</span><span class="p">(</span>
|
|
<span class="n">llm_args_attr</span>
|
|
<span class="p">),</span> <span class="sa">f</span><span class="s2">"New options found in underlying ExecutorConfig: </span><span class="si">{</span><span class="n">llm_args_attr</span><span class="w"> </span><span class="o">-</span><span class="w"> </span><span class="n">executor_config_attrs</span><span class="si">}</span><span class="s2">"</span>
|
|
|
|
<span class="c1"># ensure build_config and LlmArgsBase consistency</span>
|
|
<span class="k">if</span> <span class="n">kwargs_dict</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s2">"backend"</span><span class="p">)</span> <span class="o">!=</span> <span class="s2">"pytorch"</span> <span class="ow">and</span> <span class="n">kwargs_dict</span><span class="o">.</span><span class="n">get</span><span class="p">(</span>
|
|
<span class="s2">"build_config"</span><span class="p">):</span>
|
|
<span class="c1"># TODO: move this to _perform_config_arbitration() once it's default-on.</span>
|
|
<span class="k">for</span> <span class="n">field_name</span> <span class="ow">in</span> <span class="p">[</span>
|
|
<span class="s2">"max_input_len"</span><span class="p">,</span> <span class="s2">"max_seq_len"</span><span class="p">,</span> <span class="s2">"max_beam_width"</span>
|
|
<span class="p">]:</span>
|
|
<span class="n">build_val</span> <span class="o">=</span> <span class="nb">getattr</span><span class="p">(</span><span class="n">kwargs_dict</span><span class="p">[</span><span class="s2">"build_config"</span><span class="p">],</span> <span class="n">field_name</span><span class="p">,</span>
|
|
<span class="kc">None</span><span class="p">)</span>
|
|
<span class="n">llmargs_val</span> <span class="o">=</span> <span class="n">kwargs_dict</span><span class="o">.</span><span class="n">get</span><span class="p">(</span>
|
|
<span class="n">field_name</span><span class="p">)</span> <span class="ow">or</span> <span class="n">BaseLlmArgs</span><span class="o">.</span><span class="n">model_fields</span><span class="p">[</span><span class="n">field_name</span><span class="p">]</span>
|
|
|
|
<span class="k">if</span> <span class="n">build_val</span> <span class="o">!=</span> <span class="n">llmargs_val</span><span class="p">:</span>
|
|
<span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span>
|
|
<span class="sa">f</span><span class="s2">"Overriding LlmArgsBase.</span><span class="si">{</span><span class="n">field_name</span><span class="si">}</span><span class="s2"> (</span><span class="si">{</span><span class="n">llmargs_val</span><span class="si">}</span><span class="s2">) with build_config.</span><span class="si">{</span><span class="n">field_name</span><span class="si">}</span><span class="s2"> (</span><span class="si">{</span><span class="n">build_val</span><span class="si">}</span><span class="s2">)."</span>
|
|
<span class="p">)</span>
|
|
<span class="n">kwargs_dict</span><span class="p">[</span><span class="n">field_name</span><span class="p">]</span> <span class="o">=</span> <span class="n">build_val</span>
|
|
|
|
<span class="k">return</span> <span class="n">kwargs_dict</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_setup</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">''' This method will setup the configs right before building the model. '''</span>
|
|
|
|
<span class="n">is_trt_llm_args</span> <span class="o">=</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">TrtLlmArgs</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">model</span><span class="p">,</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="sa">f</span><span class="s2">"Invalid model: </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">model</span><span class="si">}</span><span class="s2">"</span>
|
|
|
|
<span class="k">if</span> <span class="n">is_trt_llm_args</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_setup_embedding_parallel_mode</span><span class="p">()</span>
|
|
|
|
<span class="k">if</span> <span class="n">is_trt_llm_args</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">enable_build_cache</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">enable_build_cache</span> <span class="o">=</span> <span class="n">BuildCacheConfig</span><span class="p">()</span> <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">enable_build_cache</span><span class="p">,</span> <span class="nb">bool</span><span class="p">)</span> <span class="k">else</span> <span class="bp">self</span><span class="o">.</span><span class="n">enable_build_cache</span>
|
|
<span class="k">if</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">enable_build_cache</span><span class="p">,</span> <span class="n">BuildCacheConfig</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">"Invalid build_cache_config: </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">enable_build_cache</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span>
|
|
<span class="n">model_obj</span> <span class="o">=</span> <span class="n">_ModelWrapper</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">model</span><span class="p">)</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">speculative_model</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">speculative_config</span><span class="p">,</span>
|
|
<span class="s2">"speculative_model"</span><span class="p">,</span> <span class="kc">None</span><span class="p">)</span>
|
|
<span class="n">speculative_model_obj</span> <span class="o">=</span> <span class="n">_ModelWrapper</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">speculative_model</span>
|
|
<span class="p">)</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">speculative_model</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="k">else</span> <span class="kc">None</span>
|
|
<span class="k">if</span> <span class="n">model_obj</span><span class="o">.</span><span class="n">is_local_model</span> <span class="ow">and</span> <span class="bp">self</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="s1">'pytorch'</span><span class="p">,</span> <span class="s1">'autodeploy'</span>
|
|
<span class="p">]:</span>
|
|
<span class="c1"># Load parallel_config from the engine.</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">model_format</span> <span class="o">=</span> <span class="n">get_model_format</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">model</span><span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">model_format</span> <span class="ow">is</span> <span class="n">_ModelFormatKind</span><span class="o">.</span><span class="n">TLLM_ENGINE</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">build_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">warning</span><span class="p">(</span>
|
|
<span class="s2">"The build_config is ignored for model format of TLLM_ENGINE."</span>
|
|
<span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_load_config_from_engine</span><span class="p">(</span><span class="n">model_obj</span><span class="o">.</span><span class="n">model_dir</span><span class="p">)</span>
|
|
<span class="n">runtime_defaults</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_pretrained_config</span><span class="o">.</span><span class="n">runtime_defaults</span>
|
|
<span class="k">if</span> <span class="n">runtime_defaults</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">kv_cache_config</span><span class="o">.</span><span class="n">fill_empty_fields_from_runtime_defaults</span><span class="p">(</span>
|
|
<span class="n">runtime_defaults</span><span class="p">)</span>
|
|
|
|
<span class="c1"># Load parallel_config from the checkpoint.</span>
|
|
<span class="k">elif</span> <span class="bp">self</span><span class="o">.</span><span class="n">model_format</span> <span class="ow">is</span> <span class="n">_ModelFormatKind</span><span class="o">.</span><span class="n">TLLM_CKPT</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_load_config_from_ckpt</span><span class="p">(</span><span class="n">model_obj</span><span class="o">.</span><span class="n">model_dir</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">model_format</span> <span class="o">=</span> <span class="n">_ModelFormatKind</span><span class="o">.</span><span class="n">HF</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">speculative_model</span> <span class="ow">and</span> <span class="n">speculative_model_obj</span><span class="o">.</span><span class="n">is_local_model</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">speculative_model_format</span> <span class="o">=</span> <span class="n">_ModelFormatKind</span><span class="o">.</span><span class="n">HF</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">quant_config</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">quant_config</span> <span class="ow">or</span> <span class="n">QuantConfig</span><span class="p">()</span>
|
|
|
|
<span class="k">if</span> <span class="n">is_trt_llm_args</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">calib_config</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">calib_config</span> <span class="ow">or</span> <span class="n">CalibConfig</span><span class="p">()</span>
|
|
|
|
<span class="c1"># Note: max_batch_size and max_num_tokens in LlmArgs are for runtime,</span>
|
|
<span class="c1"># which will be passed to the C++ Executor API, overwriting the values</span>
|
|
<span class="c1"># from an built engine. In order to set build configuration, it is</span>
|
|
<span class="c1"># recommended to use build_config instead.</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">build_config</span> <span class="ow">is</span> <span class="ow">not</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">max_batch_size</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">max_batch_size</span> <span class="o">!=</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_batch_size</span><span class="p">:</span>
|
|
<span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span>
|
|
<span class="sa">f</span><span class="s2">"Conflict detected in LlmArgs build_config.max_batch_size "</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="n">build_config</span><span class="o">.</span><span class="n">max_batch_size</span><span class="si">}</span><span class="s2">) != max_batch_size (</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">max_batch_size</span><span class="si">}</span><span class="s2">)."</span>
|
|
<span class="sa">f</span><span class="s2">"The 'max_batch_size' specified in LlmArgs is ignored at "</span>
|
|
<span class="sa">f</span><span class="s2">"engine build and will override at runtime."</span><span class="p">)</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_num_tokens</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">max_num_tokens</span> <span class="o">!=</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_num_tokens</span><span class="p">:</span>
|
|
<span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span>
|
|
<span class="sa">f</span><span class="s2">"Conflict detected in LlmArgs build_config.max_num_tokens "</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="n">build_config</span><span class="o">.</span><span class="n">max_num_tokens</span><span class="si">}</span><span class="s2">) != max_batch_size (</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">max_num_tokens</span><span class="si">}</span><span class="s2">)."</span>
|
|
<span class="sa">f</span><span class="s2">"The 'max_num_tokens' specified in LlmArgs is ignored at "</span>
|
|
<span class="sa">f</span><span class="s2">"engine build and will override at runtime."</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">build_config</span> <span class="o">=</span> <span class="n">BuildConfig</span><span class="p">()</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_batch_size</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">max_batch_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_batch_size</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_num_tokens</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">max_num_tokens</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_num_tokens</span>
|
|
|
|
<span class="c1"># TODO: remove the checker when manage weights support all data types</span>
|
|
<span class="k">if</span> <span class="n">is_trt_llm_args</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">fast_build</span> <span class="ow">and</span> <span class="p">(</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">quant_config</span><span class="o">.</span><span class="n">quant_algo</span> <span class="ow">is</span> <span class="n">QuantAlgo</span><span class="o">.</span><span class="n">FP8</span>
|
|
<span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">quant_config</span><span class="o">.</span><span class="n">quant_algo</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_update_plugin_config</span><span class="p">(</span><span class="s2">"manage_weights"</span><span class="p">,</span> <span class="kc">True</span><span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">_world_size</span> <span class="o">==</span> <span class="mi">1</span><span class="p">:</span>
|
|
<span class="bp">self</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">nccl_plugin</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_ensure_lora_config_consistency</span><span class="p">()</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">enable_lora</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_config</span> <span class="ow">is</span> <span class="kc">None</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">backend</span> <span class="o">!=</span> <span class="s1">'pytorch'</span><span class="p">:</span>
|
|
<span class="bp">self</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="o">=</span> <span class="s1">'auto'</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_lora_rank</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">build_config</span><span class="o">.</span><span class="n">lora_config</span><span class="o">.</span><span class="n">max_lora_rank</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_lora_rank</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_setup_speculative_config</span><span class="p">()</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">enable_prompt_adapter</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">max_prompt_embedding_table_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_prompt_adapter_token</span> <span class="o">*</span> <span class="bp">self</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">def</span><span class="w"> </span><span class="nf">_setup_speculative_config</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="p">,</span> <span class="n">LookaheadDecodingConfig</span><span class="p">):</span>
|
|
<span class="n">lookahead_config</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span>
|
|
<span class="c1"># Update the build config</span>
|
|
<span class="n">_</span><span class="p">,</span> <span class="n">_</span><span class="p">,</span> <span class="n">max_draft_tokens</span><span class="p">,</span> <span class="n">_</span> <span class="o">=</span> <span class="n">lookahead_config</span><span class="o">.</span><span class="n">calculate_speculative_resource</span><span class="p">(</span>
|
|
<span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">speculative_decoding_mode</span> <span class="o">=</span> <span class="n">SpeculativeDecodingMode</span><span class="o">.</span><span class="n">LOOKAHEAD_DECODING</span>
|
|
<span class="k">if</span> <span class="n">max_draft_tokens</span> <span class="o">></span> <span class="bp">self</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">max_draft_len</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">max_draft_len</span> <span class="o">=</span> <span class="n">max_draft_tokens</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">decoding_config</span> <span class="o">=</span> <span class="n">DecodingConfig</span><span class="p">(</span>
|
|
<span class="n">decoding_mode</span><span class="o">=</span><span class="n">DecodingMode</span><span class="o">.</span><span class="n">Lookahead</span><span class="p">(),</span>
|
|
<span class="n">lookahead_decoding_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="n">lookahead_config</span><span class="p">))</span>
|
|
<span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="p">,</span> <span class="n">MedusaDecodingConfig</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">speculative_decoding_mode</span> <span class="o">=</span> <span class="n">SpeculativeDecodingMode</span><span class="o">.</span><span class="n">MEDUSA</span>
|
|
|
|
<span class="k">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">max_draft_len</span> <span class="o">></span> <span class="mi">0</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">max_draft_len</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">max_draft_len</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">decoding_config</span> <span class="o">=</span> <span class="n">DecodingConfig</span><span class="p">(</span>
|
|
<span class="n">decoding_mode</span><span class="o">=</span><span class="n">DecodingMode</span><span class="o">.</span><span class="n">Medusa</span><span class="p">(),</span>
|
|
<span class="n">medusa_choices</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">medusa_choices</span><span class="p">)</span>
|
|
<span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="p">,</span> <span class="n">EagleDecodingConfig</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">speculative_decoding_mode</span> <span class="o">=</span> <span class="n">SpeculativeDecodingMode</span><span class="o">.</span><span class="n">EAGLE</span>
|
|
<span class="k">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">max_draft_len</span> <span class="o">></span> <span class="mi">0</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">max_draft_len</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">max_draft_len</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">backend</span> <span class="o">!=</span> <span class="s1">'pytorch'</span><span class="p">:</span>
|
|
<span class="n">eagle_config</span> <span class="o">=</span> <span class="n">_EagleConfig</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">eagle_choices</span><span class="p">,</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">greedy_sampling</span><span class="p">,</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">posterior_threshold</span><span class="p">,</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">use_dynamic_tree</span><span class="p">,</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">dynamic_tree_max_topK</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">decoding_config</span> <span class="o">=</span> <span class="n">DecodingConfig</span><span class="p">(</span>
|
|
<span class="n">decoding_mode</span><span class="o">=</span><span class="n">DecodingMode</span><span class="o">.</span><span class="n">Eagle</span><span class="p">(),</span>
|
|
<span class="n">eagle_config</span><span class="o">=</span><span class="n">eagle_config</span><span class="p">)</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="kn">from</span><span class="w"> </span><span class="nn">tensorrt_llm._torch.speculative</span><span class="w"> </span><span class="kn">import</span> <span class="n">Eagle3Config</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span> <span class="o">=</span> <span class="n">Eagle3Config</span><span class="p">(</span>
|
|
<span class="n">max_draft_tokens</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">max_draft_len</span><span class="p">,</span>
|
|
<span class="n">draft_model_path</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span>
|
|
<span class="n">pytorch_eagle_weights_path</span><span class="p">,</span>
|
|
<span class="n">eagle3_one_model</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span>
|
|
<span class="n">eagle3_one_model</span><span class="p">)</span>
|
|
<span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="p">,</span> <span class="n">NGramDecodingConfig</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">speculative_decoding_mode</span> <span class="o">=</span> <span class="n">SpeculativeDecodingMode</span><span class="o">.</span><span class="n">NGRAM</span>
|
|
<span class="k">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">backend</span> <span class="o">==</span> <span class="s1">'pytorch'</span>
|
|
<span class="k">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">prompt_lookup_num_tokens</span> <span class="o">></span> <span class="mi">0</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">max_matching_ngram_size</span> <span class="o">></span> <span class="mi">0</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">max_draft_len</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">max_draft_len</span>
|
|
<span class="kn">from</span><span class="w"> </span><span class="nn">tensorrt_llm._torch.speculative</span><span class="w"> </span><span class="kn">import</span> <span class="n">NGramConfig</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span> <span class="o">=</span> <span class="n">NGramConfig</span><span class="p">(</span>
|
|
<span class="n">prompt_lookup_num_tokens</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span>
|
|
<span class="n">prompt_lookup_num_tokens</span><span class="p">,</span>
|
|
<span class="n">max_matching_ngram_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span>
|
|
<span class="n">max_matching_ngram_size</span><span class="p">,</span>
|
|
<span class="n">is_keep_all</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">is_keep_all</span><span class="p">,</span>
|
|
<span class="n">is_use_oldest</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">is_use_oldest</span><span class="p">,</span>
|
|
<span class="n">is_public_pool</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">is_public_pool</span><span class="p">,</span>
|
|
<span class="p">)</span>
|
|
<span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="p">,</span> <span class="n">MTPDecodingConfig</span><span class="p">):</span>
|
|
<span class="kn">from</span><span class="w"> </span><span class="nn">tensorrt_llm._torch.speculative</span><span class="w"> </span><span class="kn">import</span> <span class="n">MTPConfig</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span> <span class="o">=</span> <span class="n">MTPConfig</span><span class="p">(</span>
|
|
<span class="n">num_nextn_predict_layers</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span>
|
|
<span class="n">num_nextn_predict_layers</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">build_config</span><span class="o">.</span><span class="n">max_batch_size</span><span class="p">,</span>
|
|
<span class="n">use_relaxed_acceptance_for_thinking</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span>
|
|
<span class="n">use_relaxed_acceptance_for_thinking</span><span class="p">,</span>
|
|
<span class="n">relaxed_topk</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">relaxed_topk</span><span class="p">,</span>
|
|
<span class="n">relaxed_delta</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="o">.</span><span class="n">relaxed_delta</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">"Speculative config type not recognized: </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">speculative_config</span><span class="si">}</span><span class="s2">"</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">decoding_config</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_ensure_lora_config_consistency</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_config</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_lora_rank</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">warning</span><span class="p">(</span>
|
|
<span class="s2">"max_lora_rank is ignored when lora_config is provided."</span><span class="p">)</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_loras</span> <span class="o">!=</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_config</span><span class="o">.</span><span class="n">max_loras</span><span class="p">:</span>
|
|
<span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span>
|
|
<span class="s2">"max_loras is ignored when lora_config is provided."</span><span class="p">)</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_cpu_loras</span> <span class="o">!=</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_config</span><span class="o">.</span><span class="n">max_cpu_loras</span><span class="p">:</span>
|
|
<span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span>
|
|
<span class="s2">"max_cpu_loras is ignored when lora_config is provided."</span><span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="nb">len</span><span class="p">(</span><span class="bp">self</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="o">==</span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="c1"># TODO [TRTLLM-5173]</span>
|
|
<span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span>
|
|
<span class="s2">"lora_dir is empty, so custom embedding or lm head will not be applied."</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">enable_lora</span> <span class="ow">and</span> <span class="bp">self</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="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">backend</span> <span class="o">==</span> <span class="s1">'pytorch'</span><span class="p">:</span>
|
|
<span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span>
|
|
<span class="s2">"enable_lora is ignored when lora_config is provided for pytorch backend."</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</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="k">if</span> <span class="nb">len</span><span class="p">(</span><span class="bp">self</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="o">==</span> <span class="mi">0</span> <span class="ow">and</span> <span class="nb">len</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">lora_config</span><span class="o">.</span><span class="n">lora_target_modules</span><span class="p">)</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span>
|
|
<span class="s2">"Both lora_dir and lora_target_modules are empty, so all LoRA modules will be expected. "</span>
|
|
<span class="s2">"This will lead to serious memory consumption. Please provide either lora_dir or lora_target_modules if this behavior is not what you expect."</span>
|
|
<span class="p">)</span>
|
|
<span class="n">default_trtllm_modules_to_hf_modules</span> <span class="o">=</span> <span class="n">get_default_trtllm_modules_to_hf_modules</span><span class="p">(</span>
|
|
<span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">lora_config</span><span class="o">.</span><span class="n">lora_target_modules</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span>
|
|
<span class="n">default_trtllm_modules_to_hf_modules</span><span class="o">.</span><span class="n">keys</span><span class="p">())</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_build_config_mutable</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="bp">self</span><span class="o">.</span><span class="n">model_format</span> <span class="ow">is</span> <span class="ow">not</span> <span class="n">_ModelFormatKind</span><span class="o">.</span><span class="n">TLLM_ENGINE</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_update_plugin_config</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">key</span><span class="p">:</span> <span class="nb">str</span><span class="p">,</span> <span class="n">value</span><span class="p">:</span> <span class="n">Any</span><span class="p">):</span>
|
|
<span class="nb">setattr</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">plugin_config</span><span class="p">,</span> <span class="n">key</span><span class="p">,</span> <span class="n">value</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_load_config_from_engine</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="n">Path</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="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">_pretrained_config</span> <span class="o">=</span> <span class="n">engine_config</span><span class="o">.</span><span class="n">pretrained_config</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">build_config</span> <span class="o">=</span> <span class="n">engine_config</span><span class="o">.</span><span class="n">build_config</span>
|
|
|
|
<span class="c1"># load and check parallel_config</span>
|
|
<span class="n">mapping</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_pretrained_config</span><span class="o">.</span><span class="n">mapping</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">tp_size</span> <span class="ow">not</span> <span class="ow">in</span> <span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</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">"tp_size </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">tp_size</span><span class="si">}</span><span class="s2"> is not consistent with the engine's tp_size </span><span class="si">{</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</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">parallel_config</span><span class="o">.</span><span class="n">pp_size</span> <span class="ow">not</span> <span class="ow">in</span> <span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="n">mapping</span><span class="o">.</span><span class="n">pp_size</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">"pp_size </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">pp_size</span><span class="si">}</span><span class="s2"> is not consistent with the engine's pp_size </span><span class="si">{</span><span class="n">mapping</span><span class="o">.</span><span class="n">pp_size</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">parallel_config</span><span class="o">.</span><span class="n">cp_size</span> <span class="ow">not</span> <span class="ow">in</span> <span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="n">mapping</span><span class="o">.</span><span class="n">cp_size</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">"cp_size </span><span class="si">{</span><span class="bp">self</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"> is not consistent with the engine's cp_size </span><span class="si">{</span><span class="n">mapping</span><span class="o">.</span><span class="n">cp_size</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">parallel_config</span> <span class="o">=</span> <span class="n">_ParallelConfig</span><span class="p">(</span>
|
|
<span class="n">tp_size</span><span class="o">=</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">pp_size</span><span class="o">=</span><span class="n">mapping</span><span class="o">.</span><span class="n">pp_size</span><span class="p">,</span>
|
|
<span class="n">cp_size</span><span class="o">=</span><span class="n">mapping</span><span class="o">.</span><span class="n">cp_size</span><span class="p">,</span>
|
|
<span class="n">gpus_per_node</span><span class="o">=</span><span class="n">mapping</span><span class="o">.</span><span class="n">gpus_per_node</span><span class="p">,</span>
|
|
<span class="n">moe_cluster_size</span><span class="o">=</span><span class="n">mapping</span><span class="o">.</span><span class="n">moe_cluster_size</span><span class="p">,</span>
|
|
<span class="n">moe_tp_size</span><span class="o">=</span><span class="n">mapping</span><span class="o">.</span><span class="n">moe_tp_size</span><span class="p">,</span>
|
|
<span class="n">moe_ep_size</span><span class="o">=</span><span class="n">mapping</span><span class="o">.</span><span class="n">moe_ep_size</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_load_config_from_ckpt</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">ckpt_dir</span><span class="p">:</span> <span class="n">Path</span><span class="p">):</span>
|
|
<span class="n">pretrained_config</span> <span class="o">=</span> <span class="n">PretrainedConfig</span><span class="o">.</span><span class="n">from_json_file</span><span class="p">(</span><span class="n">ckpt_dir</span> <span class="o">/</span>
|
|
<span class="s2">"config.json"</span><span class="p">)</span>
|
|
<span class="n">tp_size</span> <span class="o">=</span> <span class="n">pretrained_config</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span>
|
|
<span class="n">pp_size</span> <span class="o">=</span> <span class="n">pretrained_config</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">pp_size</span>
|
|
<span class="n">cp_size</span> <span class="o">=</span> <span class="n">pretrained_config</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">cp_size</span>
|
|
<span class="n">moe_cluster_size</span> <span class="o">=</span> <span class="n">pretrained_config</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">moe_cluster_size</span>
|
|
<span class="n">moe_tp_size</span> <span class="o">=</span> <span class="n">pretrained_config</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">moe_tp_size</span>
|
|
<span class="n">moe_ep_size</span> <span class="o">=</span> <span class="n">pretrained_config</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">moe_ep_size</span>
|
|
<span class="n">world_size</span> <span class="o">=</span> <span class="n">pretrained_config</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">world_size</span>
|
|
<span class="n">gpus_per_node</span> <span class="o">=</span> <span class="n">pretrained_config</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">gpus_per_node</span>
|
|
<span class="c1"># load parallel_config</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">tp_size</span> <span class="o">!=</span> <span class="mi">1</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">tp_size</span> <span class="o">!=</span> <span class="n">tp_size</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">"tp_size </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">tp_size</span><span class="si">}</span><span class="s2"> is not consistent with the checkpoint's tp_size </span><span class="si">{</span><span class="n">tp_size</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">parallel_config</span><span class="o">.</span><span class="n">pp_size</span> <span class="o">!=</span> <span class="mi">1</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">pp_size</span> <span class="o">!=</span> <span class="n">pp_size</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">"pp_size </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">pp_size</span><span class="si">}</span><span class="s2"> is not consistent with the checkpoint's pp_size </span><span class="si">{</span><span class="n">pp_size</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">parallel_config</span><span class="o">.</span><span class="n">cp_size</span> <span class="o">!=</span> <span class="mi">1</span> <span class="ow">and</span> <span class="bp">self</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">cp_size</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">"cp_size </span><span class="si">{</span><span class="bp">self</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"> is not consistent with the checkpoint's cp_size </span><span class="si">{</span><span class="n">cp_size</span><span class="si">}</span><span class="s2">"</span>
|
|
<span class="p">)</span>
|
|
<span class="k">if</span> <span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">auto_parallel</span>
|
|
<span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">world_size</span> <span class="o">!=</span> <span class="mi">1</span> <span class="ow">and</span> <span class="n">world_size</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">"auto parallel with world_size </span><span class="si">{</span><span class="bp">self</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"> does not support checkpoint with "</span>
|
|
<span class="s2">"world_size </span><span class="si">{world_size}</span><span class="s2"> > 1"</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">parallel_config</span><span class="o">.</span><span class="n">auto_parallel</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">parallel_config</span> <span class="o">=</span> <span class="n">_ParallelConfig</span><span class="p">(</span>
|
|
<span class="n">tp_size</span><span class="o">=</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">pp_size</span><span class="o">=</span><span class="n">pp_size</span><span class="p">,</span>
|
|
<span class="n">cp_size</span><span class="o">=</span><span class="n">cp_size</span><span class="p">,</span>
|
|
<span class="n">gpus_per_node</span><span class="o">=</span><span class="n">gpus_per_node</span><span class="p">,</span>
|
|
<span class="n">moe_cluster_size</span><span class="o">=</span><span class="n">moe_cluster_size</span><span class="p">,</span>
|
|
<span class="n">moe_tp_size</span><span class="o">=</span><span class="n">moe_tp_size</span><span class="p">,</span>
|
|
<span class="n">moe_ep_size</span><span class="o">=</span><span class="n">moe_ep_size</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_setup_embedding_parallel_mode</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">embedding_parallel_mode</span> <span class="o">==</span> <span class="s1">'NONE'</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_convert_checkpoint_options</span><span class="p">[</span><span class="s1">'use_parallel_embedding'</span><span class="p">]</span> <span class="o">=</span> <span class="kc">False</span>
|
|
<span class="k">elif</span> <span class="bp">self</span><span class="o">.</span><span class="n">embedding_parallel_mode</span> <span class="o">==</span> <span class="s1">'SHARDING_ALONG_VOCAB'</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_convert_checkpoint_options</span><span class="p">[</span><span class="s1">'use_parallel_embedding'</span><span class="p">]</span> <span class="o">=</span> <span class="kc">True</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_convert_checkpoint_options</span><span class="p">[</span><span class="s1">'embedding_sharding_dim'</span><span class="p">]</span> <span class="o">=</span> <span class="mi">0</span>
|
|
<span class="k">elif</span> <span class="bp">self</span><span class="o">.</span><span class="n">embedding_parallel_mode</span> <span class="o">==</span> <span class="s1">'SHARDING_ALONG_HIDDEN'</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_convert_checkpoint_options</span><span class="p">[</span><span class="s1">'use_parallel_embedding'</span><span class="p">]</span> <span class="o">=</span> <span class="kc">True</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_convert_checkpoint_options</span><span class="p">[</span><span class="s1">'embedding_sharding_dim'</span><span class="p">]</span> <span class="o">=</span> <span class="mi">1</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">"Invalid embedding_parallel_mode: </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">llm_args</span><span class="o">.</span><span class="n">embedding_parallel_mode</span><span class="si">}</span><span class="s2">"</span>
|
|
<span class="p">)</span>
|
|
|
|
|
|
<div class="viewcode-block" id="TrtLlmArgs">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.TrtLlmArgs">[docs]</a>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">TrtLlmArgs</span><span class="p">(</span><span class="n">BaseLlmArgs</span><span class="p">):</span>
|
|
|
|
<span class="n">auto_parallel</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Enable auto parallel mode."</span><span class="p">,</span>
|
|
<span class="n">deprecated</span><span class="o">=</span>
|
|
<span class="s2">"Use tensor_parallel_size/pipeline_parallel_size/xxx_parallel_size instead."</span><span class="p">,</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="n">auto_parallel_world_size</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The world size for auto parallel mode."</span><span class="p">,</span>
|
|
<span class="n">deprecated</span><span class="o">=</span>
|
|
<span class="s2">"Use tensor_parallel_size/pipeline_parallel_size/xxx_parallel_size instead."</span><span class="p">,</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="n">enable_tqdm</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Enable tqdm for progress bar."</span><span class="p">)</span>
|
|
|
|
<span class="c1"># BuildConfig is introduced to give users a familiar interface to configure the model building.</span>
|
|
<span class="n">build_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">object</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Build config."</span><span class="p">,</span>
|
|
<span class="n">json_schema_extra</span><span class="o">=</span><span class="p">{</span><span class="s2">"type"</span><span class="p">:</span> <span class="sa">f</span><span class="s2">"Optional[</span><span class="si">{</span><span class="n">get_type_repr</span><span class="p">(</span><span class="n">BuildConfig</span><span class="p">)</span><span class="si">}</span><span class="s2">]"</span><span class="p">})</span>
|
|
|
|
<span class="n">workspace</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="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The workspace for the model."</span><span class="p">)</span>
|
|
|
|
<span class="c1"># Once set, the model will reuse the build_cache</span>
|
|
<span class="n">enable_build_cache</span><span class="p">:</span> <span class="nb">object</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Enable build cache."</span><span class="p">,</span>
|
|
<span class="n">json_schema_extra</span><span class="o">=</span><span class="p">{</span>
|
|
<span class="s2">"type"</span><span class="p">:</span> <span class="sa">f</span><span class="s2">"Union[</span><span class="si">{</span><span class="n">get_type_repr</span><span class="p">(</span><span class="n">BuildCacheConfig</span><span class="p">)</span><span class="si">}</span><span class="s2">, bool]"</span>
|
|
<span class="p">})</span>
|
|
|
|
<span class="n">extended_runtime_perf_knob_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span>
|
|
<span class="n">ExtendedRuntimePerfKnobConfig</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Extended runtime perf knob config."</span><span class="p">)</span>
|
|
|
|
<span class="n">calib_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">CalibConfig</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Calibration config."</span><span class="p">)</span>
|
|
|
|
<span class="n">embedding_parallel_mode</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="s1">'SHARDING_ALONG_VOCAB'</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"The embedding parallel mode."</span><span class="p">)</span>
|
|
|
|
<span class="n">fast_build</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Enable fast build."</span><span class="p">)</span>
|
|
|
|
<span class="c1"># Private attributes</span>
|
|
<span class="n">_auto_parallel_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">AutoParallelConfig</span><span class="p">]</span> <span class="o">=</span> <span class="n">PrivateAttr</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">)</span>
|
|
<span class="c1"># This is used to hold the options for convert_checkpoint</span>
|
|
<span class="n">_convert_checkpoint_options</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span>
|
|
<span class="n">Any</span><span class="p">]</span> <span class="o">=</span> <span class="n">PrivateAttr</span><span class="p">(</span><span class="n">default_factory</span><span class="o">=</span><span class="nb">dict</span><span class="p">)</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">auto_parallel_config</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="n">AutoParallelConfig</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_auto_parallel_config</span>
|
|
|
|
<div class="viewcode-block" id="TrtLlmArgs.model_post_init">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.TrtLlmArgs.model_post_init">[docs]</a>
|
|
<span class="nd">@print_traceback_on_error</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">model_post_init</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">__context</span><span class="p">):</span>
|
|
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="n">model_post_init</span><span class="p">(</span><span class="n">__context</span><span class="p">)</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_auto_parallel_config</span> <span class="o">=</span> <span class="n">AutoParallelConfig</span><span class="p">(</span>
|
|
<span class="n">sharded_io_allowlist</span><span class="o">=</span><span class="p">[</span>
|
|
<span class="s2">"past_key_value_</span><span class="se">\\</span><span class="s2">d+"</span><span class="p">,</span>
|
|
<span class="s2">"present_key_value_</span><span class="se">\\</span><span class="s2">d*"</span><span class="p">,</span>
|
|
<span class="p">],</span>
|
|
<span class="n">same_buffer_io</span><span class="o">=</span><span class="p">{</span>
|
|
<span class="s2">"past_key_value_(</span><span class="se">\\</span><span class="s2">d+)"</span><span class="p">:</span> <span class="s2">"present_key_value_</span><span class="se">\\</span><span class="s2">1"</span><span class="p">,</span>
|
|
<span class="p">},</span>
|
|
<span class="o">**</span><span class="n">infer_cluster_config</span><span class="p">(),</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">auto_parallel</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">auto_parallel</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">auto_parallel</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">parallel_config</span><span class="o">.</span><span class="n">world_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">auto_parallel_world_size</span></div>
|
|
</div>
|
|
|
|
|
|
|
|
<span class="n">LlmArgs</span> <span class="o">=</span> <span class="n">TrtLlmArgs</span>
|
|
|
|
<span class="n">LLMARGS_EXPLICIT_DOCSTRING</span> <span class="o">=</span> <span class="n">generate_api_docs_as_docstring</span><span class="p">(</span><span class="n">LlmArgs</span><span class="p">,</span>
|
|
<span class="n">indent</span><span class="o">=</span><span class="s1">' '</span> <span class="o">*</span> <span class="mi">4</span><span class="p">)</span>
|
|
|
|
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">LoadFormat</span><span class="p">(</span><span class="n">Enum</span><span class="p">):</span>
|
|
<span class="n">AUTO</span> <span class="o">=</span> <span class="mi">0</span>
|
|
<span class="c1"># Initialize all weights randomly.</span>
|
|
<span class="n">DUMMY</span> <span class="o">=</span> <span class="mi">1</span>
|
|
|
|
|
|
<div class="viewcode-block" id="TorchLlmArgs">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.TorchLlmArgs">[docs]</a>
|
|
<span class="k">class</span><span class="w"> </span><span class="nc">TorchLlmArgs</span><span class="p">(</span><span class="n">BaseLlmArgs</span><span class="p">):</span>
|
|
|
|
<span class="c1"># Just a dummy BuildConfig to allow code reuse with the TrtLlmArgs</span>
|
|
<span class="n">build_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">object</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Build config."</span><span class="p">,</span>
|
|
<span class="n">exclude_from_json</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
|
<span class="n">json_schema_extra</span><span class="o">=</span><span class="p">{</span><span class="s2">"type"</span><span class="p">:</span> <span class="sa">f</span><span class="s2">"Optional[</span><span class="si">{</span><span class="n">get_type_repr</span><span class="p">(</span><span class="n">BuildConfig</span><span class="p">)</span><span class="si">}</span><span class="s2">]"</span><span class="p">})</span>
|
|
|
|
<span class="c1"># PyTorch backend specific configurations</span>
|
|
|
|
<span class="n">use_cuda_graph</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"If true, use CUDA graphs for decoding. CUDA graphs are only created for the batch sizes in cuda_graph_batch_sizes, and are enabled for batches that consist of decoding requests *only* (the reason is that it's hard to capture a single graph with prefill requests since the input shapes are a function of the sequence lengths). Note that each CUDA graph can use up to 200 MB of extra memory."</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="n">cuda_graph_batch_sizes</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"List of batch sizes to create CUDA graphs for."</span><span class="p">)</span>
|
|
|
|
<span class="n">cuda_graph_max_batch_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Maximum batch size for CUDA graphs."</span><span class="p">)</span>
|
|
|
|
<span class="n">cuda_graph_padding_enabled</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"If true, batches are rounded up to the nearest cuda_graph_batch_size. This is usually a net win for performance."</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="n">disable_overlap_scheduler</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Disable the overlap scheduler."</span><span class="p">)</span>
|
|
|
|
<span class="n">moe_max_num_tokens</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"If set, at most moe_max_num_tokens tokens will be sent to torch.ops.trtllm.fused_moe at the same time. If the number of tokens exceeds moe_max_num_tokens, the input tensors will be split into chunks and a for loop will be used."</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="n">moe_load_balancer</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">object</span><span class="p">,</span> <span class="nb">str</span><span class="p">]]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Configuration for MoE load balancing."</span><span class="p">,</span>
|
|
<span class="n">json_schema_extra</span><span class="o">=</span><span class="p">{</span><span class="s2">"type"</span><span class="p">:</span> <span class="s2">"Union[MoeLoadBalancerConfig, str]"</span><span class="p">})</span>
|
|
|
|
<span class="n">attn_backend</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="s1">'TRTLLM'</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Attention backend to use."</span><span class="p">)</span>
|
|
|
|
<span class="n">moe_backend</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="s1">'CUTLASS'</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"MoE backend to use."</span><span class="p">)</span>
|
|
|
|
<span class="n">mixed_sampler</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"If true, will iterate over sampling_params of each request and use the corresponding sampling strategy, e.g. top-k, top-p, etc."</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="n">enable_trtllm_sampler</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"If true, will use the TRTLLM sampler instead of the PyTorch sampler. The TRTLLM sampler has a wide coverage of sampling strategies."</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="n">kv_cache_dtype</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="s2">"auto"</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Data type for KV cache."</span><span class="p">)</span>
|
|
|
|
<span class="n">use_kv_cache</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Whether to use KV cache."</span><span class="p">)</span>
|
|
|
|
<span class="n">enable_iter_perf_stats</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Enable iteration performance statistics."</span><span class="p">)</span>
|
|
|
|
<span class="n">enable_iter_req_stats</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"If true, enables per request stats per iteration. Must also set enable_iter_perf_stats to true to get request stats."</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="n">print_iter_log</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Print iteration logs."</span><span class="p">)</span>
|
|
|
|
<span class="n">torch_compile_enabled</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Enable torch.compile optimization."</span><span class="p">)</span>
|
|
|
|
<span class="n">torch_compile_fullgraph</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Enable full graph compilation in torch.compile."</span><span class="p">)</span>
|
|
|
|
<span class="n">torch_compile_inductor_enabled</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"Enable inductor backend in torch.compile."</span><span class="p">)</span>
|
|
|
|
<span class="n">torch_compile_piecewise_cuda_graph</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Enable piecewise CUDA graph in torch.compile."</span><span class="p">)</span>
|
|
|
|
<span class="n">torch_compile_enable_userbuffers</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"When torch compile is enabled, userbuffers is enabled by default."</span><span class="p">)</span>
|
|
|
|
<span class="n">autotuner_enabled</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Enable autotuner only when torch compile is enabled."</span><span class="p">)</span>
|
|
|
|
<span class="n">enable_layerwise_nvtx_marker</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">description</span><span class="o">=</span><span class="s2">"If true, enable layerwise nvtx marker."</span><span class="p">)</span>
|
|
|
|
<span class="n">auto_deploy_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">object</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span><span class="s2">"Auto deploy config."</span><span class="p">,</span>
|
|
<span class="n">exclude_from_json</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
|
<span class="n">json_schema_extra</span><span class="o">=</span><span class="p">{</span><span class="s2">"type"</span><span class="p">:</span> <span class="sa">f</span><span class="s2">"Optional[AutoDeployConfig]"</span><span class="p">})</span>
|
|
|
|
<span class="n">load_format</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">LoadFormat</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="n">LoadFormat</span><span class="o">.</span><span class="n">AUTO</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"How to load the model weights. By default, detect the weight type from the model checkpoint."</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="n">enable_min_latency</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span>
|
|
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">description</span><span class="o">=</span>
|
|
<span class="s2">"If true, enable min-latency mode. Currently only used for Llama4."</span><span class="p">,</span>
|
|
<span class="p">)</span>
|
|
|
|
<div class="viewcode-block" id="TorchLlmArgs.convert_load_format">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.TorchLlmArgs.convert_load_format">[docs]</a>
|
|
<span class="nd">@field_validator</span><span class="p">(</span><span class="s1">'load_format'</span><span class="p">,</span> <span class="n">mode</span><span class="o">=</span><span class="s1">'before'</span><span class="p">)</span>
|
|
<span class="nd">@classmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">convert_load_format</span><span class="p">(</span><span class="bp">cls</span><span class="p">,</span> <span class="n">v</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">v</span><span class="p">,</span> <span class="n">LoadFormat</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="n">v</span>
|
|
<span class="n">load_format</span> <span class="o">=</span> <span class="n">v</span><span class="o">.</span><span class="n">upper</span><span class="p">()</span>
|
|
<span class="k">if</span> <span class="n">load_format</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">LoadFormat</span><span class="o">.</span><span class="n">__members__</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">"Invalid LoadFormat: </span><span class="si">{</span><span class="n">v</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="n">LoadFormat</span><span class="p">[</span><span class="n">load_format</span><span class="p">]</span></div>
|
|
|
|
|
|
<span class="c1"># Extra resource managers to use in addition to the KV cache manager.</span>
|
|
<span class="c1"># Each manager's prepare_resources method is called before the forward pass,</span>
|
|
<span class="c1"># and update_resources() is called after the pass finishes. free_resources()</span>
|
|
<span class="c1"># is called when a request finishes. The KV cache manager is guaranteed to</span>
|
|
<span class="c1"># be invoked after all of these extra managers in all stages.</span>
|
|
<span class="n">_extra_resource_managers</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span>
|
|
<span class="nb">object</span><span class="p">]</span> <span class="o">=</span> <span class="n">PrivateAttr</span><span class="p">(</span><span class="n">default_factory</span><span class="o">=</span><span class="nb">dict</span><span class="p">,</span> <span class="p">)</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">extra_resource_managers</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">object</span><span class="p">]:</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_extra_resource_managers</span>
|
|
|
|
<span class="nd">@extra_resource_managers</span><span class="o">.</span><span class="n">setter</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">extra_resource_managers</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">value</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">object</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">_extra_resource_managers</span> <span class="o">=</span> <span class="n">value</span>
|
|
|
|
<div class="viewcode-block" id="TorchLlmArgs.model_post_init">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.TorchLlmArgs.model_post_init">[docs]</a>
|
|
<span class="nd">@print_traceback_on_error</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">model_post_init</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">__context</span><span class="p">):</span>
|
|
<span class="kn">from</span><span class="w"> </span><span class="nn">.._torch.model_config</span><span class="w"> </span><span class="kn">import</span> <span class="n">MoeLoadBalancerConfig</span>
|
|
|
|
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="n">model_post_init</span><span class="p">(</span><span class="n">__context</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">model_format</span> <span class="o">=</span> <span class="n">_ModelFormatKind</span><span class="o">.</span><span class="n">HF</span>
|
|
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">moe_load_balancer</span><span class="p">,</span> <span class="nb">str</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="ow">not</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">exists</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">moe_load_balancer</span><span class="p">):</span>
|
|
<span class="k">raise</span> <span class="ne">FileNotFoundError</span><span class="p">(</span>
|
|
<span class="sa">f</span><span class="s2">"MoE load balancer config file not found: </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">moe_load_balancer</span><span class="si">}</span><span class="s2">"</span>
|
|
<span class="p">)</span>
|
|
<span class="k">try</span><span class="p">:</span>
|
|
<span class="k">with</span> <span class="nb">open</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">moe_load_balancer</span><span class="p">)</span> <span class="k">as</span> <span class="n">f</span><span class="p">:</span>
|
|
<span class="n">moe_load_balancer_config</span> <span class="o">=</span> <span class="n">yaml</span><span class="o">.</span><span class="n">safe_load</span><span class="p">(</span><span class="n">f</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">moe_load_balancer</span> <span class="o">=</span> <span class="n">MoeLoadBalancerConfig</span><span class="p">(</span>
|
|
<span class="o">**</span><span class="n">moe_load_balancer_config</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="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
|
|
<span class="sa">f</span><span class="s2">"Failed to load MoE load balancer config file: </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">moe_load_balancer</span><span class="si">}</span><span class="s2">"</span>
|
|
<span class="p">)</span> <span class="kn">from</span><span class="w"> </span><span class="nn">e</span></div>
|
|
|
|
|
|
<span class="c1"># TODO: Remove this after the PyTorch backend is fully migrated to TorchLlmArgs from ExecutorConfig</span>
|
|
<div class="viewcode-block" id="TorchLlmArgs.get_pytorch_backend_config">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.TorchLlmArgs.get_pytorch_backend_config">[docs]</a>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">get_pytorch_backend_config</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="s2">"PyTorchConfig"</span><span class="p">:</span>
|
|
<span class="kn">from</span><span class="w"> </span><span class="nn">tensorrt_llm._torch.pyexecutor.config</span><span class="w"> </span><span class="kn">import</span> <span class="n">PyTorchConfig</span>
|
|
|
|
<span class="c1"># TODO: Remove this after the PyTorch backend is fully migrated to TorchLlmArgs from ExecutorConfig</span>
|
|
<span class="c1"># Just a WAR to support the auto_deploy</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">auto_deploy_config</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">auto_deploy_config</span>
|
|
|
|
<span class="k">return</span> <span class="n">PyTorchConfig</span><span class="p">(</span>
|
|
<span class="n">extra_resource_managers</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">extra_resource_managers</span><span class="p">,</span>
|
|
<span class="n">use_cuda_graph</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">use_cuda_graph</span><span class="p">,</span>
|
|
<span class="n">cuda_graph_batch_sizes</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">cuda_graph_batch_sizes</span><span class="p">,</span>
|
|
<span class="n">cuda_graph_max_batch_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">cuda_graph_max_batch_size</span><span class="p">,</span>
|
|
<span class="n">cuda_graph_padding_enabled</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">cuda_graph_padding_enabled</span><span class="p">,</span>
|
|
<span class="n">disable_overlap_scheduler</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">disable_overlap_scheduler</span><span class="p">,</span>
|
|
<span class="n">moe_max_num_tokens</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">moe_max_num_tokens</span><span class="p">,</span>
|
|
<span class="n">moe_load_balancer</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">moe_load_balancer</span><span class="p">,</span>
|
|
<span class="n">attn_backend</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">attn_backend</span><span class="p">,</span>
|
|
<span class="n">moe_backend</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">moe_backend</span><span class="p">,</span>
|
|
<span class="n">mixed_sampler</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">mixed_sampler</span><span class="p">,</span>
|
|
<span class="n">enable_trtllm_sampler</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">enable_trtllm_sampler</span><span class="p">,</span>
|
|
<span class="n">kv_cache_dtype</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">kv_cache_dtype</span><span class="p">,</span>
|
|
<span class="n">use_kv_cache</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">use_kv_cache</span><span class="p">,</span>
|
|
<span class="n">enable_iter_perf_stats</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">enable_iter_perf_stats</span><span class="p">,</span>
|
|
<span class="n">enable_iter_req_stats</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">enable_iter_req_stats</span><span class="p">,</span>
|
|
<span class="n">print_iter_log</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">print_iter_log</span><span class="p">,</span>
|
|
<span class="n">torch_compile_enabled</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">torch_compile_enabled</span><span class="p">,</span>
|
|
<span class="n">torch_compile_fullgraph</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">torch_compile_fullgraph</span><span class="p">,</span>
|
|
<span class="n">torch_compile_inductor_enabled</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">torch_compile_inductor_enabled</span><span class="p">,</span>
|
|
<span class="n">torch_compile_piecewise_cuda_graph</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span>
|
|
<span class="n">torch_compile_piecewise_cuda_graph</span><span class="p">,</span>
|
|
<span class="n">torch_compile_enable_userbuffers</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span>
|
|
<span class="n">torch_compile_enable_userbuffers</span><span class="p">,</span>
|
|
<span class="n">autotuner_enabled</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">autotuner_enabled</span><span class="p">,</span>
|
|
<span class="n">enable_layerwise_nvtx_marker</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">enable_layerwise_nvtx_marker</span><span class="p">,</span>
|
|
<span class="n">load_format</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">load_format</span><span class="p">,</span>
|
|
<span class="n">enable_min_latency</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">enable_min_latency</span><span class="p">)</span></div>
|
|
|
|
|
|
<div class="viewcode-block" id="TorchLlmArgs.validate_cuda_graph_max_batch_size">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.TorchLlmArgs.validate_cuda_graph_max_batch_size">[docs]</a>
|
|
<span class="nd">@field_validator</span><span class="p">(</span><span class="s1">'cuda_graph_max_batch_size'</span><span class="p">)</span>
|
|
<span class="nd">@classmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">validate_cuda_graph_max_batch_size</span><span class="p">(</span><span class="bp">cls</span><span class="p">,</span> <span class="n">v</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">"""Validate cuda_graph_max_batch_size is non-negative."""</span>
|
|
<span class="k">if</span> <span class="n">v</span> <span class="o"><</span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s2">"cuda_graph_max_batch_size must be non-negative"</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="n">v</span></div>
|
|
|
|
|
|
<span class="nd">@staticmethod</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">_generate_cuda_graph_batch_sizes</span><span class="p">(</span><span class="n">max_batch_size</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span>
|
|
<span class="n">padding_enabled</span><span class="p">:</span> <span class="nb">bool</span><span class="p">)</span> <span class="o">-></span> <span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]:</span>
|
|
<span class="w"> </span><span class="sd">"""Generate a list of batch sizes for CUDA graphs.</span>
|
|
|
|
<span class="sd"> Args:</span>
|
|
<span class="sd"> max_batch_size: Maximum batch size to generate up to</span>
|
|
<span class="sd"> padding_enabled: Whether padding is enabled, which affects the batch size distribution</span>
|
|
|
|
<span class="sd"> Returns:</span>
|
|
<span class="sd"> List of batch sizes to create CUDA graphs for</span>
|
|
<span class="sd"> """</span>
|
|
<span class="k">if</span> <span class="n">padding_enabled</span><span class="p">:</span>
|
|
<span class="n">batch_sizes</span> <span class="o">=</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">4</span><span class="p">]</span> <span class="o">+</span> <span class="p">[</span><span class="n">i</span> <span class="o">*</span> <span class="mi">8</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">17</span><span class="p">)]</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="n">batch_sizes</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="nb">range</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">32</span><span class="p">))</span> <span class="o">+</span> <span class="p">[</span><span class="mi">32</span><span class="p">,</span> <span class="mi">64</span><span class="p">,</span> <span class="mi">128</span><span class="p">]</span>
|
|
|
|
<span class="c1"># Add powers of 2 up to max_batch_size</span>
|
|
<span class="n">batch_sizes</span> <span class="o">+=</span> <span class="p">[</span>
|
|
<span class="mi">2</span><span class="o">**</span><span class="n">i</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">8</span><span class="p">,</span> <span class="n">math</span><span class="o">.</span><span class="n">floor</span><span class="p">(</span><span class="n">math</span><span class="o">.</span><span class="n">log</span><span class="p">(</span><span class="n">max_batch_size</span><span class="p">,</span> <span class="mi">2</span><span class="p">)))</span>
|
|
<span class="p">]</span>
|
|
|
|
<span class="c1"># Filter and sort batch sizes</span>
|
|
<span class="n">batch_sizes</span> <span class="o">=</span> <span class="nb">sorted</span><span class="p">(</span>
|
|
<span class="p">[</span><span class="n">size</span> <span class="k">for</span> <span class="n">size</span> <span class="ow">in</span> <span class="n">batch_sizes</span> <span class="k">if</span> <span class="n">size</span> <span class="o"><=</span> <span class="n">max_batch_size</span><span class="p">])</span>
|
|
|
|
<span class="c1"># Add max_batch_size if not already included</span>
|
|
<span class="k">if</span> <span class="n">max_batch_size</span> <span class="o">!=</span> <span class="n">batch_sizes</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">]:</span>
|
|
<span class="n">batch_sizes</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">max_batch_size</span><span class="p">)</span>
|
|
|
|
<span class="k">return</span> <span class="n">batch_sizes</span>
|
|
|
|
<div class="viewcode-block" id="TorchLlmArgs.validate_cuda_graph_config">
|
|
<a class="viewcode-back" href="../../../llm-api/reference.html#tensorrt_llm.llmapi.TorchLlmArgs.validate_cuda_graph_config">[docs]</a>
|
|
<span class="nd">@model_validator</span><span class="p">(</span><span class="n">mode</span><span class="o">=</span><span class="s1">'after'</span><span class="p">)</span>
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">validate_cuda_graph_config</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="s1">'TorchLlmArgs'</span><span class="p">:</span>
|
|
<span class="w"> </span><span class="sd">"""Validate CUDA graph configuration.</span>
|
|
|
|
<span class="sd"> Ensures that:</span>
|
|
<span class="sd"> 1. If cuda_graph_batch_sizes is provided, cuda_graph_max_batch_size must be 0</span>
|
|
<span class="sd"> 2. If cuda_graph_batch_sizes is not provided, it is generated based on cuda_graph_max_batch_size</span>
|
|
<span class="sd"> 3. If both are provided, cuda_graph_batch_sizes must match the generated values</span>
|
|
<span class="sd"> """</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">cuda_graph_batch_sizes</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">cuda_graph_batch_sizes</span> <span class="o">=</span> <span class="nb">sorted</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">cuda_graph_batch_sizes</span><span class="p">)</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">cuda_graph_max_batch_size</span> <span class="o">!=</span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">cuda_graph_batch_sizes</span> <span class="o">!=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_generate_cuda_graph_batch_sizes</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">cuda_graph_max_batch_size</span><span class="p">,</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">cuda_graph_padding_enabled</span><span class="p">):</span>
|
|
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
|
|
<span class="s2">"Please don't set both cuda_graph_batch_sizes "</span>
|
|
<span class="s2">"and cuda_graph_max_batch_size.</span><span class="se">\n</span><span class="s2">"</span>
|
|
<span class="sa">f</span><span class="s2">"cuda_graph_batch_sizes: </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">cuda_graph_batch_sizes</span><span class="si">}</span><span class="s2">, "</span>
|
|
<span class="sa">f</span><span class="s2">"cuda_graph_max_batch_size: </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">cuda_graph_max_batch_size</span><span class="si">}</span><span class="s2">"</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">cuda_graph_max_batch_size</span> <span class="o">=</span> <span class="nb">max</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">cuda_graph_batch_sizes</span><span class="p">)</span>
|
|
<span class="k">else</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">cuda_graph_max_batch_size</span> <span class="ow">or</span> <span class="mi">128</span>
|
|
<span class="n">generated_sizes</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_generate_cuda_graph_batch_sizes</span><span class="p">(</span>
|
|
<span class="n">max_batch_size</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">cuda_graph_padding_enabled</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">cuda_graph_batch_sizes</span> <span class="o">=</span> <span class="n">generated_sizes</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">cuda_graph_max_batch_size</span> <span class="o">=</span> <span class="n">max_batch_size</span>
|
|
|
|
<span class="k">return</span> <span class="bp">self</span></div>
|
|
</div>
|
|
|
|
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">update_llm_args_with_extra_dict</span><span class="p">(</span>
|
|
<span class="n">llm_args</span><span class="p">:</span> <span class="n">Dict</span><span class="p">,</span>
|
|
<span class="n">llm_args_dict</span><span class="p">:</span> <span class="n">Dict</span><span class="p">,</span>
|
|
<span class="n">extra_llm_api_options</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">Dict</span><span class="p">:</span>
|
|
|
|
<span class="kn">from</span><span class="w"> </span><span class="nn">.._torch.pyexecutor.config</span><span class="w"> </span><span class="kn">import</span> <span class="n">PyTorchConfig</span>
|
|
<span class="n">field_mapping</span> <span class="o">=</span> <span class="p">{</span>
|
|
<span class="s2">"quant_config"</span><span class="p">:</span> <span class="n">QuantConfig</span><span class="p">,</span>
|
|
<span class="s2">"calib_config"</span><span class="p">:</span> <span class="n">CalibConfig</span><span class="p">,</span>
|
|
<span class="s2">"build_config"</span><span class="p">:</span> <span class="n">BuildConfig</span><span class="p">,</span>
|
|
<span class="s2">"kv_cache_config"</span><span class="p">:</span> <span class="n">KvCacheConfig</span><span class="p">,</span>
|
|
<span class="s2">"decoding_config"</span><span class="p">:</span> <span class="n">DecodingConfig</span><span class="p">,</span>
|
|
<span class="s2">"enable_build_cache"</span><span class="p">:</span> <span class="n">BuildCacheConfig</span><span class="p">,</span>
|
|
<span class="s2">"peft_cache_config"</span><span class="p">:</span> <span class="n">PeftCacheConfig</span><span class="p">,</span>
|
|
<span class="s2">"scheduler_config"</span><span class="p">:</span> <span class="n">SchedulerConfig</span><span class="p">,</span>
|
|
<span class="s2">"speculative_config"</span><span class="p">:</span> <span class="n">DecodingBaseConfig</span><span class="p">,</span>
|
|
<span class="s2">"batching_type"</span><span class="p">:</span> <span class="n">BatchingType</span><span class="p">,</span>
|
|
<span class="s2">"extended_runtime_perf_knob_config"</span><span class="p">:</span> <span class="n">ExtendedRuntimePerfKnobConfig</span><span class="p">,</span>
|
|
<span class="s2">"pytorch_backend_config"</span><span class="p">:</span> <span class="n">PyTorchConfig</span><span class="p">,</span>
|
|
<span class="s2">"cache_transceiver_config"</span><span class="p">:</span> <span class="n">CacheTransceiverConfig</span><span class="p">,</span>
|
|
<span class="p">}</span>
|
|
<span class="k">for</span> <span class="n">field</span><span class="p">,</span> <span class="n">field_type</span> <span class="ow">in</span> <span class="n">field_mapping</span><span class="o">.</span><span class="n">items</span><span class="p">():</span>
|
|
<span class="k">if</span> <span class="n">field</span> <span class="ow">in</span> <span class="n">llm_args_dict</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="n">field</span> <span class="o">==</span> <span class="s2">"speculative_config"</span><span class="p">:</span>
|
|
<span class="n">llm_args_dict</span><span class="p">[</span><span class="n">field</span><span class="p">]</span> <span class="o">=</span> <span class="n">field_type</span><span class="o">.</span><span class="n">from_dict</span><span class="p">(</span>
|
|
<span class="n">llm_args_dict</span><span class="p">[</span><span class="n">field</span><span class="p">])</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="n">llm_args_dict</span><span class="p">[</span><span class="n">field</span><span class="p">]</span> <span class="o">=</span> <span class="n">field_type</span><span class="p">(</span><span class="o">**</span><span class="n">llm_args_dict</span><span class="p">[</span><span class="n">field</span><span class="p">])</span>
|
|
<span class="n">extra_llm_str</span> <span class="o">=</span> <span class="sa">f</span><span class="s2">"because it's specified in </span><span class="si">{</span><span class="n">extra_llm_api_options</span><span class="si">}</span><span class="s2">"</span> <span class="k">if</span> <span class="n">extra_llm_api_options</span> <span class="k">else</span> <span class="s2">""</span>
|
|
<span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span><span class="sa">f</span><span class="s2">"Overriding </span><span class="si">{</span><span class="n">field</span><span class="si">}</span><span class="s2"> </span><span class="si">{</span><span class="n">extra_llm_str</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span>
|
|
|
|
<span class="n">llm_args</span> <span class="o">=</span> <span class="n">llm_args</span> <span class="o">|</span> <span class="n">llm_args_dict</span>
|
|
<span class="k">return</span> <span class="n">llm_args</span>
|
|
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">update_llm_args_with_extra_options</span><span class="p">(</span><span class="n">llm_args</span><span class="p">:</span> <span class="n">Dict</span><span class="p">,</span>
|
|
<span class="n">extra_llm_api_options</span><span class="p">:</span> <span class="nb">str</span><span class="p">)</span> <span class="o">-></span> <span class="n">Dict</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="n">extra_llm_api_options</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="k">with</span> <span class="nb">open</span><span class="p">(</span><span class="n">extra_llm_api_options</span><span class="p">,</span> <span class="s1">'r'</span><span class="p">)</span> <span class="k">as</span> <span class="n">f</span><span class="p">:</span>
|
|
<span class="n">llm_args_dict</span> <span class="o">=</span> <span class="n">yaml</span><span class="o">.</span><span class="n">safe_load</span><span class="p">(</span><span class="n">f</span><span class="p">)</span>
|
|
<span class="n">llm_args</span> <span class="o">=</span> <span class="n">update_llm_args_with_extra_dict</span><span class="p">(</span><span class="n">llm_args</span><span class="p">,</span> <span class="n">llm_args_dict</span><span class="p">,</span>
|
|
<span class="n">extra_llm_api_options</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="n">llm_args</span>
|
|
|
|
|
|
<span class="k">def</span><span class="w"> </span><span class="nf">get_model_format</span><span class="p">(</span><span class="n">model_dir</span><span class="p">:</span> <span class="nb">str</span><span class="p">)</span> <span class="o">-></span> <span class="n">_ModelFormatKind</span><span class="p">:</span>
|
|
<span class="w"> </span><span class="sd">''' Get the format of the model. '''</span>
|
|
<span class="k">if</span> <span class="ow">not</span> <span class="p">(</span><span class="n">Path</span><span class="p">(</span><span class="n">model_dir</span><span class="p">)</span> <span class="o">/</span> <span class="s1">'config.json'</span><span class="p">)</span><span class="o">.</span><span class="n">exists</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">"Failed to infer model format because no config.json exists in </span><span class="si">{</span><span class="n">model_dir</span><span class="si">}</span><span class="s2">"</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="k">with</span> <span class="nb">open</span><span class="p">(</span><span class="n">Path</span><span class="p">(</span><span class="n">model_dir</span><span class="p">)</span> <span class="o">/</span> <span class="s1">'config.json'</span><span class="p">)</span> <span class="k">as</span> <span class="n">f</span><span class="p">:</span>
|
|
<span class="n">config</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="k">try</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="s1">'pretrained_config'</span> <span class="ow">in</span> <span class="n">config</span> <span class="ow">and</span> <span class="s1">'build_config'</span> <span class="ow">in</span> <span class="n">config</span><span class="p">:</span>
|
|
<span class="n">model_format</span> <span class="o">=</span> <span class="n">_ModelFormatKind</span><span class="o">.</span><span class="n">TLLM_ENGINE</span>
|
|
<span class="n">EngineConfig</span><span class="o">.</span><span class="n">from_json_file</span><span class="p">(</span><span class="n">Path</span><span class="p">(</span><span class="n">model_dir</span><span class="p">)</span> <span class="o">/</span> <span class="s1">'config.json'</span><span class="p">)</span>
|
|
<span class="k">elif</span> <span class="s1">'architecture'</span> <span class="ow">in</span> <span class="n">config</span> <span class="ow">and</span> <span class="s1">'dtype'</span> <span class="ow">in</span> <span class="n">config</span><span class="p">:</span>
|
|
<span class="n">model_format</span> <span class="o">=</span> <span class="n">_ModelFormatKind</span><span class="o">.</span><span class="n">TLLM_CKPT</span>
|
|
<span class="n">PretrainedConfig</span><span class="o">.</span><span class="n">from_checkpoint</span><span class="p">(</span><span class="n">model_dir</span><span class="p">)</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="n">model_format</span> <span class="o">=</span> <span class="n">_ModelFormatKind</span><span class="o">.</span><span class="n">HF</span>
|
|
<span class="n">AutoConfig</span><span class="o">.</span><span class="n">from_hugging_face</span><span class="p">(</span><span class="n">model_dir</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="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span>
|
|
<span class="sa">f</span><span class="s2">"Inferred model format </span><span class="si">{</span><span class="n">model_format</span><span class="si">}</span><span class="s2">, but failed to load config.json: </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">else</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="n">model_format</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 June 03, 2025.</p>
|
|
|
|
<p>This page is generated by TensorRT-LLM commit <a href="https://github.com/NVIDIA/TensorRT-LLM/tree/9ae2ce6">9ae2ce6</a>.</p>
|
|
|
|
</div></div>
|
|
|
|
</div>
|
|
|
|
|
|
|
|
</div>
|
|
|
|
</footer>
|
|
</body>
|
|
</html> |