mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-14 06:27:45 +08:00
1989 lines
208 KiB
HTML
1989 lines
208 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.models.eagle.model — TensorRT-LLM</title>
|
||
|
||
|
||
|
||
<script data-cfasync="false">
|
||
document.documentElement.dataset.mode = localStorage.getItem("mode") || "";
|
||
document.documentElement.dataset.theme = localStorage.getItem("theme") || "";
|
||
</script>
|
||
<!--
|
||
this give us a css class that will be invisible only if js is disabled
|
||
-->
|
||
<noscript>
|
||
<style>
|
||
.pst-js-only { display: none !important; }
|
||
|
||
</style>
|
||
</noscript>
|
||
|
||
<!-- Loaded before other Sphinx assets -->
|
||
<link href="../../../../_static/styles/theme.css?digest=8878045cc6db502f8baf" rel="stylesheet" />
|
||
<link href="../../../../_static/styles/pydata-sphinx-theme.css?digest=8878045cc6db502f8baf" rel="stylesheet" />
|
||
|
||
<link rel="stylesheet" type="text/css" href="../../../../_static/pygments.css?v=8f2a1f02" />
|
||
<link rel="stylesheet" type="text/css" href="../../../../_static/styles/nvidia-sphinx-theme.css?v=df3ac72c" />
|
||
<link rel="stylesheet" type="text/css" href="../../../../_static/copybutton.css?v=76b2166b" />
|
||
<link rel="stylesheet" type="text/css" href="../../../../_static/autodoc_pydantic.css" />
|
||
<link rel="stylesheet" type="text/css" href="../../../../_static/togglebutton.css?v=13237357" />
|
||
<link rel="stylesheet" type="text/css" href="../../../../_static/custom.css?v=95073da6" />
|
||
|
||
<!-- So that users can add custom icons -->
|
||
<script src="../../../../_static/scripts/fontawesome.js?digest=8878045cc6db502f8baf"></script>
|
||
<!-- Pre-loaded scripts that we'll load fully later -->
|
||
<link rel="preload" as="script" href="../../../../_static/scripts/bootstrap.js?digest=8878045cc6db502f8baf" />
|
||
<link rel="preload" as="script" href="../../../../_static/scripts/pydata-sphinx-theme.js?digest=8878045cc6db502f8baf" />
|
||
|
||
<script src="../../../../_static/documentation_options.js?v=5929fcd5"></script>
|
||
<script src="../../../../_static/doctools.js?v=9a2dae69"></script>
|
||
<script src="../../../../_static/sphinx_highlight.js?v=dc90522c"></script>
|
||
<script src="../../../../_static/clipboard.min.js?v=a7894cd8"></script>
|
||
<script src="../../../../_static/copybutton.js?v=65e89d2a"></script>
|
||
<script>let toggleHintShow = 'Click to show';</script>
|
||
<script>let toggleHintHide = 'Click to hide';</script>
|
||
<script>let toggleOpenOnPrint = 'true';</script>
|
||
<script src="../../../../_static/togglebutton.js?v=4a39c7ea"></script>
|
||
<script>var togglebuttonSelector = '.toggle, .admonition.dropdown';</script>
|
||
<script>var togglebuttonSelector = '.toggle, .admonition.dropdown';</script>
|
||
<script>DOCUMENTATION_OPTIONS.pagename = '_modules/tensorrt_llm/models/eagle/model';</script>
|
||
<script>
|
||
DOCUMENTATION_OPTIONS.theme_version = '0.16.1';
|
||
DOCUMENTATION_OPTIONS.theme_switcher_json_url = './_static/switcher.json';
|
||
DOCUMENTATION_OPTIONS.theme_switcher_version_match = '1.1.0rc3';
|
||
DOCUMENTATION_OPTIONS.show_version_warning_banner =
|
||
false;
|
||
</script>
|
||
<link rel="icon" href="../../../../_static/favicon.png"/>
|
||
<link rel="index" title="Index" href="../../../../genindex.html" />
|
||
<link rel="search" title="Search" href="../../../../search.html" />
|
||
|
||
<meta name="viewport" content="width=device-width, initial-scale=1"/>
|
||
<meta name="docsearch:language" content="en"/>
|
||
<meta name="docsearch:version" content="1.1.0rc3" />
|
||
|
||
|
||
</head>
|
||
|
||
|
||
<body data-bs-spy="scroll" data-bs-target=".bd-toc-nav" data-offset="180" data-bs-root-margin="0px 0px -60%" data-default-mode="">
|
||
|
||
|
||
|
||
<div id="pst-skip-link" class="skip-link d-print-none"><a href="#main-content">Skip to main content</a></div>
|
||
|
||
<div id="pst-scroll-pixel-helper"></div>
|
||
|
||
<button type="button" class="btn rounded-pill" id="pst-back-to-top">
|
||
<i class="fa-solid fa-arrow-up"></i>Back to top</button>
|
||
|
||
|
||
<dialog id="pst-search-dialog">
|
||
|
||
<form class="bd-search d-flex align-items-center"
|
||
action="../../../../search.html"
|
||
method="get">
|
||
<i class="fa-solid fa-magnifying-glass"></i>
|
||
<input type="search"
|
||
class="form-control"
|
||
name="q"
|
||
placeholder="Search the docs ..."
|
||
aria-label="Search the docs ..."
|
||
autocomplete="off"
|
||
autocorrect="off"
|
||
autocapitalize="off"
|
||
spellcheck="false"/>
|
||
<span class="search-button__kbd-shortcut"><kbd class="kbd-shortcut__modifier">Ctrl</kbd>+<kbd>K</kbd></span>
|
||
</form>
|
||
</dialog>
|
||
|
||
<div class="pst-async-banner-revealer d-none">
|
||
<aside id="bd-header-version-warning" class="d-none d-print-none" aria-label="Version warning"></aside>
|
||
</div>
|
||
|
||
|
||
<header class="bd-header navbar navbar-expand-lg bd-navbar d-print-none">
|
||
<div class="bd-header__inner bd-page-width">
|
||
<button class="pst-navbar-icon sidebar-toggle primary-toggle" aria-label="Site navigation">
|
||
<span class="fa-solid fa-bars"></span>
|
||
</button>
|
||
|
||
|
||
<div class="col-lg-3 navbar-header-items__start">
|
||
|
||
<div class="navbar-item">
|
||
|
||
|
||
|
||
|
||
|
||
<a class="navbar-brand logo" href="../../../../index.html">
|
||
|
||
|
||
|
||
|
||
|
||
|
||
|
||
|
||
|
||
|
||
<img src="../../../../_static/nvidia-logo-horiz-rgb-blk-for-screen.svg" class="logo__image only-light" alt="TensorRT-LLM - Home"/>
|
||
<img src="../../../../_static/nvidia-logo-horiz-rgb-wht-for-screen.svg" class="logo__image only-dark pst-js-only" alt="TensorRT-LLM - Home"/>
|
||
|
||
|
||
<p class="title logo__title">TensorRT-LLM</p>
|
||
|
||
</a></div>
|
||
|
||
</div>
|
||
|
||
<div class="col-lg-9 navbar-header-items">
|
||
|
||
<div class="me-auto navbar-header-items__center">
|
||
|
||
<div class="navbar-item">
|
||
|
||
|
||
<div class="version-switcher__container dropdown pst-js-only">
|
||
<button id="pst-version-switcher-button-2"
|
||
type="button"
|
||
class="version-switcher__button btn btn-sm dropdown-toggle"
|
||
data-bs-toggle="dropdown"
|
||
aria-haspopup="listbox"
|
||
aria-controls="pst-version-switcher-list-2"
|
||
aria-label="Version switcher list"
|
||
>
|
||
Choose version <!-- this text may get changed later by javascript -->
|
||
<span class="caret"></span>
|
||
</button>
|
||
<div id="pst-version-switcher-list-2"
|
||
class="version-switcher__menu dropdown-menu list-group-flush py-0"
|
||
role="listbox" aria-labelledby="pst-version-switcher-button-2">
|
||
<!-- dropdown will be populated by javascript on page load -->
|
||
</div>
|
||
</div></div>
|
||
|
||
</div>
|
||
|
||
|
||
<div class="navbar-header-items__end">
|
||
|
||
<div class="navbar-item navbar-persistent--container">
|
||
|
||
|
||
<button class="btn search-button-field search-button__button pst-js-only" title="Search" aria-label="Search" data-bs-placement="bottom" data-bs-toggle="tooltip">
|
||
<i class="fa-solid fa-magnifying-glass"></i>
|
||
<span class="search-button__default-text">Search</span>
|
||
<span class="search-button__kbd-shortcut"><kbd class="kbd-shortcut__modifier">Ctrl</kbd>+<kbd class="kbd-shortcut__modifier">K</kbd></span>
|
||
</button>
|
||
</div>
|
||
|
||
|
||
<div class="navbar-item">
|
||
|
||
<button class="btn btn-sm nav-link pst-navbar-icon theme-switch-button pst-js-only" aria-label="Color mode" data-bs-title="Color mode" data-bs-placement="bottom" data-bs-toggle="tooltip">
|
||
<i class="theme-switch fa-solid fa-sun fa-lg" data-mode="light" title="Light"></i>
|
||
<i class="theme-switch fa-solid fa-moon fa-lg" data-mode="dark" title="Dark"></i>
|
||
<i class="theme-switch fa-solid fa-circle-half-stroke fa-lg" data-mode="auto" title="System Settings"></i>
|
||
</button></div>
|
||
|
||
</div>
|
||
|
||
</div>
|
||
|
||
|
||
<div class="navbar-persistent--mobile">
|
||
|
||
<button class="btn search-button-field search-button__button pst-js-only" title="Search" aria-label="Search" data-bs-placement="bottom" data-bs-toggle="tooltip">
|
||
<i class="fa-solid fa-magnifying-glass"></i>
|
||
<span class="search-button__default-text">Search</span>
|
||
<span class="search-button__kbd-shortcut"><kbd class="kbd-shortcut__modifier">Ctrl</kbd>+<kbd class="kbd-shortcut__modifier">K</kbd></span>
|
||
</button>
|
||
</div>
|
||
|
||
|
||
|
||
</div>
|
||
|
||
</header>
|
||
|
||
|
||
<div class="bd-container">
|
||
<div class="bd-container__inner bd-page-width">
|
||
|
||
|
||
|
||
<dialog id="pst-primary-sidebar-modal"></dialog>
|
||
<div id="pst-primary-sidebar" class="bd-sidebar-primary bd-sidebar">
|
||
|
||
|
||
|
||
|
||
|
||
|
||
|
||
|
||
<a class="navbar-brand logo" href="../../../../index.html">
|
||
|
||
|
||
|
||
|
||
|
||
|
||
|
||
|
||
|
||
|
||
<img src="../../../../_static/nvidia-logo-horiz-rgb-blk-for-screen.svg" class="logo__image only-light" alt="TensorRT-LLM - Home"/>
|
||
<img src="../../../../_static/nvidia-logo-horiz-rgb-wht-for-screen.svg" class="logo__image only-dark pst-js-only" alt="TensorRT-LLM - Home"/>
|
||
|
||
|
||
<p class="title logo__title">TensorRT-LLM</p>
|
||
|
||
</a>
|
||
|
||
|
||
|
||
<div class="sidebar-header-items sidebar-primary__section">
|
||
|
||
|
||
<div class="sidebar-header-items__center">
|
||
|
||
|
||
|
||
<div class="navbar-item">
|
||
|
||
|
||
<div class="version-switcher__container dropdown pst-js-only">
|
||
<button id="pst-version-switcher-button-3"
|
||
type="button"
|
||
class="version-switcher__button btn btn-sm dropdown-toggle"
|
||
data-bs-toggle="dropdown"
|
||
aria-haspopup="listbox"
|
||
aria-controls="pst-version-switcher-list-3"
|
||
aria-label="Version switcher list"
|
||
>
|
||
Choose version <!-- this text may get changed later by javascript -->
|
||
<span class="caret"></span>
|
||
</button>
|
||
<div id="pst-version-switcher-list-3"
|
||
class="version-switcher__menu dropdown-menu list-group-flush py-0"
|
||
role="listbox" aria-labelledby="pst-version-switcher-button-3">
|
||
<!-- dropdown will be populated by javascript on page load -->
|
||
</div>
|
||
</div></div>
|
||
|
||
|
||
</div>
|
||
|
||
|
||
|
||
<div class="sidebar-header-items__end">
|
||
|
||
<div class="navbar-item">
|
||
|
||
<button class="btn btn-sm nav-link pst-navbar-icon theme-switch-button pst-js-only" aria-label="Color mode" data-bs-title="Color mode" data-bs-placement="bottom" data-bs-toggle="tooltip">
|
||
<i class="theme-switch fa-solid fa-sun fa-lg" data-mode="light" title="Light"></i>
|
||
<i class="theme-switch fa-solid fa-moon fa-lg" data-mode="dark" title="Dark"></i>
|
||
<i class="theme-switch fa-solid fa-circle-half-stroke fa-lg" data-mode="auto" title="System Settings"></i>
|
||
</button></div>
|
||
|
||
</div>
|
||
|
||
</div>
|
||
|
||
<div class="sidebar-primary-items__start sidebar-primary__section">
|
||
<div class="sidebar-primary-item">
|
||
|
||
|
||
|
||
<nav class="bd-docs-nav bd-links"
|
||
aria-label="Table of Contents">
|
||
<p class="bd-links__title" role="heading" aria-level="1">Table of Contents</p>
|
||
<div class="bd-toc-item navbar-nav"><p aria-level="2" class="caption" role="heading"><span class="caption-text">Getting Started</span></p>
|
||
<ul class="nav bd-sidenav">
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../overview.html">Overview</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../quick-start-guide.html">Quick Start Guide</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../key-features.html">Key Features</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../torch.html">PyTorch Backend</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../release-notes.html">Release Notes</a></li>
|
||
</ul>
|
||
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Installation</span></p>
|
||
<ul class="nav bd-sidenav">
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../installation/containers.html">Pre-built release container images on NGC</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../installation/linux.html">Installing on Linux via <code class="docutils literal notranslate"><span class="pre">pip</span></code></a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../installation/build-from-source-linux.html">Building from Source Code on Linux</a></li>
|
||
</ul>
|
||
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Deployment Guide</span></p>
|
||
<ul class="nav bd-sidenav">
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../deployment-guide/quick-start-recipe-for-llama4-scout-on-trtllm.html">Quick Start Recipe for Llama4 Scout 17B on TensorRT-LLM - Blackwell & Hopper Hardware</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../deployment-guide/quick-start-recipe-for-deepseek-r1-on-trtllm.html">Quick Start Recipe for DeepSeek R1 on TensorRT-LLM - Blackwell & Hopper Hardware</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../deployment-guide/quick-start-recipe-for-llama3.3-70b-on-trtllm.html">Quick Start Recipe for Llama3.3 70B on TensorRT-LLM - Blackwell & Hopper Hardware</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../deployment-guide/quick-start-recipe-for-gpt-oss-on-trtllm.html">Quick Start Recipe for GPT-OSS on TensorRT-LLM - Blackwell Hardware</a></li>
|
||
</ul>
|
||
<p aria-level="2" class="caption" role="heading"><span class="caption-text">LLM API</span></p>
|
||
<ul class="nav bd-sidenav">
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../llm-api/index.html">LLM API Introduction</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../llm-api/reference.html">API Reference</a></li>
|
||
</ul>
|
||
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Examples</span></p>
|
||
<ul class="nav bd-sidenav">
|
||
<li class="toctree-l1 has-children"><a class="reference internal" href="../../../../examples/index.html">LLM Examples Introduction</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul class="simple">
|
||
</ul>
|
||
</details></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../examples/customization.html">LLM Common Customizations</a></li>
|
||
<li class="toctree-l1 has-children"><a class="reference internal" href="../../../../examples/llm_api_examples.html">LLM Examples</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/llm_inference.html">Generate text</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/llm_inference_async.html">Generate text asynchronously</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/llm_inference_async_streaming.html">Generate text in streaming</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/llm_inference_distributed.html">Distributed LLM Generation</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/llm_guided_decoding.html">Generate text with guided decoding</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/llm_logits_processor.html">Control generated text using logits processor</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/llm_multilora.html">Generate text with multiple LoRA adapters</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/llm_speculative_decoding.html">Speculative Decoding</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/llm_kv_cache_connector.html">KV Cache Connector</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/llm_runtime.html">Runtime Configuration Examples</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/llm_sampling.html">Sampling Techniques Showcase</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/llm_mgmn_llm_distributed.html">Run LLM-API with pytorch backend on Slurm</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/llm_mgmn_trtllm_bench.html">Run trtllm-bench with pytorch backend on Slurm</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/llm_mgmn_trtllm_serve.html">Run trtllm-serve with pytorch backend on Slurm</a></li>
|
||
</ul>
|
||
</details></li>
|
||
<li class="toctree-l1 has-children"><a class="reference internal" href="../../../../examples/trtllm_serve_examples.html">Online Serving Examples</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/curl_chat_client.html">Curl Chat Client</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/curl_chat_client_for_multimodal.html">Curl Chat Client For Multimodal</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/curl_completion_client.html">Curl Completion Client</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/deepseek_r1_reasoning_parser.html">Deepseek R1 Reasoning Parser</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/genai_perf_client.html">Genai Perf Client</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/genai_perf_client_for_multimodal.html">Genai Perf Client For Multimodal</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/openai_chat_client.html">OpenAI Chat Client</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/openai_chat_client_for_multimodal.html">OpenAI Chat Client for Multimodal</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/openai_completion_client.html">OpenAI Completion Client</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/openai_completion_client_for_lora.html">Openai Completion Client For Lora</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../examples/openai_completion_client_json_schema.html">OpenAI Completion Client with JSON Schema</a></li>
|
||
</ul>
|
||
</details></li>
|
||
</ul>
|
||
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Model Definition API</span></p>
|
||
<ul class="nav bd-sidenav">
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../python-api/tensorrt_llm.layers.html">Layers</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../python-api/tensorrt_llm.functional.html">Functionals</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../python-api/tensorrt_llm.models.html">Models</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../python-api/tensorrt_llm.plugin.html">Plugin</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../python-api/tensorrt_llm.quantization.html">Quantization</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../python-api/tensorrt_llm.runtime.html">Runtime</a></li>
|
||
</ul>
|
||
<p aria-level="2" class="caption" role="heading"><span class="caption-text">C++ API</span></p>
|
||
<ul class="nav bd-sidenav">
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../_cpp_gen/executor.html">Executor</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../_cpp_gen/runtime.html">Runtime</a></li>
|
||
</ul>
|
||
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Command-Line Reference</span></p>
|
||
<ul class="nav bd-sidenav">
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../commands/trtllm-bench.html">trtllm-bench</a></li>
|
||
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../commands/trtllm-build.html">trtllm-build</a></li>
|
||
<li class="toctree-l1 has-children"><a class="reference internal" href="../../../../commands/trtllm-serve/index.html">trtllm-serve</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../commands/trtllm-serve/trtllm-serve.html">trtllm-serve</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../commands/trtllm-serve/run-benchmark-with-trtllm-serve.html">Run benchmarking with <code class="docutils literal notranslate"><span class="pre">trtllm-serve</span></code></a></li>
|
||
</ul>
|
||
</details></li>
|
||
</ul>
|
||
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Architecture</span></p>
|
||
<ul class="nav bd-sidenav">
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../architecture/overview.html">TensorRT-LLM Architecture</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../architecture/core-concepts.html">Model Definition</a></li>
|
||
|
||
|
||
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../architecture/checkpoint.html">TensorRT-LLM Checkpoint</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../architecture/workflow.html">TensorRT-LLM Build Workflow</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../architecture/add-model.html">Adding a Model</a></li>
|
||
</ul>
|
||
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Advanced</span></p>
|
||
<ul class="nav bd-sidenav">
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/gpt-attention.html">Multi-Head, Multi-Query, and Group-Query Attention</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/gpt-runtime.html">C++ GPT Runtime</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/executor.html">Executor API</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/graph-rewriting.html">Graph Rewriting Module</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/lora.html">Run gpt-2b + LoRA using Executor / cpp runtime</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/expert-parallelism.html">Expert Parallelism in TensorRT-LLM</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/kv-cache-management.html">KV Cache Management: Pools, Blocks, and Events</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/kv-cache-reuse.html">KV cache reuse</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/speculative-decoding.html">Speculative Sampling</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/disaggregated-service.html">Disaggregated-Service (Prototype)</a></li>
|
||
</ul>
|
||
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Performance</span></p>
|
||
<ul class="nav bd-sidenav">
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../performance/perf-overview.html">Overview</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../performance/perf-benchmarking.html">Benchmarking</a></li>
|
||
<li class="toctree-l1 has-children"><a class="reference internal" href="../../../../performance/performance-tuning-guide/index.html">Performance Tuning Guide</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../performance/performance-tuning-guide/benchmarking-default-performance.html">Benchmarking Default Performance</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../performance/performance-tuning-guide/useful-build-time-flags.html">Useful Build-Time Flags</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../performance/performance-tuning-guide/tuning-max-batch-size-and-max-num-tokens.html">Tuning Max Batch Size and Max Num Tokens</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../performance/performance-tuning-guide/deciding-model-sharding-strategy.html">Deciding Model Sharding Strategy</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../performance/performance-tuning-guide/fp8-quantization.html">FP8 Quantization</a></li>
|
||
<li class="toctree-l2"><a class="reference internal" href="../../../../performance/performance-tuning-guide/useful-runtime-flags.html">Useful Runtime Options</a></li>
|
||
</ul>
|
||
</details></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../performance/perf-analysis.html">Performance Analysis</a></li>
|
||
</ul>
|
||
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Reference</span></p>
|
||
<ul class="nav bd-sidenav">
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../reference/troubleshooting.html">Troubleshooting</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../reference/support-matrix.html">Support Matrix</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../reference/precision.html">Numerical Precision</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../reference/memory.html">Memory Usage of TensorRT-LLM</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../reference/ci-overview.html">Continuous Integration Overview</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../reference/dev-containers.html">Using Dev Containers</a></li>
|
||
</ul>
|
||
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Blogs</span></p>
|
||
<ul class="nav bd-sidenav">
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/H100vsA100.html">H100 has 4.6x A100 Performance in TensorRT-LLM, achieving 10,000 tok/s at 100ms to first token</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/H200launch.html">H200 achieves nearly 12,000 tokens/sec on Llama2-13B with TensorRT-LLM</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/Falcon180B-H200.html">Falcon-180B on a single H200 GPU with INT4 AWQ, and 6.7x faster Llama-70B over A100</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/quantization-in-TRT-LLM.html">Speed up inference with SOTA quantization techniques in TRT-LLM</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/XQA-kernel.html">New XQA-kernel provides 2.4x more Llama-70B throughput within the same latency budget</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/tech_blog/blog10_ADP_Balance_Strategy.html">ADP Balance Strategy</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/tech_blog/blog1_Pushing_Latency_Boundaries_Optimizing_DeepSeek-R1_Performance_on_NVIDIA_B200_GPUs.html">Pushing Latency Boundaries: Optimizing DeepSeek-R1 Performance on NVIDIA B200 GPUs</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/tech_blog/blog2_DeepSeek_R1_MTP_Implementation_and_Optimization.html">DeepSeek R1 MTP Implementation and Optimization</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/tech_blog/blog3_Optimizing_DeepSeek_R1_Throughput_on_NVIDIA_Blackwell_GPUs.html">Optimizing DeepSeek R1 Throughput on NVIDIA Blackwell GPUs: A Deep Dive for Developers</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/tech_blog/blog4_Scaling_Expert_Parallelism_in_TensorRT-LLM.html">Scaling Expert Parallelism in TensorRT-LLM (Part 1: Design and Implementation of Large-scale EP)</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/tech_blog/blog5_Disaggregated_Serving_in_TensorRT-LLM.html">Disaggregated Serving in TensorRT-LLM</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/tech_blog/blog6_Llama4_maverick_eagle_guide.html">How to launch Llama4 Maverick + Eagle3 TensorRT-LLM server</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/tech_blog/blog7_NGram_performance_Analysis_And_Auto_Enablement.html">N-Gram Speculative Decoding in TensorRT‑LLM</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/tech_blog/blog8_Scaling_Expert_Parallelism_in_TensorRT-LLM_part2.html">Scaling Expert Parallelism in TensorRT-LLM (Part 2: Performance Status and Optimization)</a></li>
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/tech_blog/blog9_Deploying_GPT_OSS_on_TRTLLM.html">Running a High Performance GPT-OSS-120B Inference Server with TensorRT-LLM</a></li>
|
||
</ul>
|
||
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Use TensorRT Engine</span></p>
|
||
<ul class="nav bd-sidenav">
|
||
<li class="toctree-l1"><a class="reference internal" href="../../../../legacy/tensorrt_quickstart.html">LLM API with TensorRT Engine</a></li>
|
||
</ul>
|
||
</div>
|
||
</nav></div>
|
||
</div>
|
||
|
||
|
||
<div class="sidebar-primary-items__end sidebar-primary__section">
|
||
</div>
|
||
|
||
|
||
|
||
</div>
|
||
|
||
<main id="main-content" class="bd-main" role="main">
|
||
|
||
|
||
<div class="bd-content">
|
||
<div class="bd-article-container">
|
||
|
||
<div class="bd-header-article d-print-none">
|
||
<div class="header-article-items header-article__inner">
|
||
|
||
<div class="header-article-items__start">
|
||
|
||
<div class="header-article-item">
|
||
|
||
<nav aria-label="Breadcrumb" class="d-print-none">
|
||
<ul class="bd-breadcrumbs">
|
||
|
||
<li class="breadcrumb-item breadcrumb-home">
|
||
<a href="../../../../index.html" class="nav-link" aria-label="Home">
|
||
<i class="fa-solid fa-home"></i>
|
||
</a>
|
||
</li>
|
||
|
||
<li class="breadcrumb-item"><a href="../../../index.html" class="nav-link">Module code</a></li>
|
||
|
||
<li class="breadcrumb-item active" aria-current="page"><span class="ellipsis">tensorrt_llm.models.eagle.model</span></li>
|
||
</ul>
|
||
</nav>
|
||
</div>
|
||
|
||
</div>
|
||
|
||
|
||
</div>
|
||
</div>
|
||
|
||
|
||
|
||
|
||
<div id="searchbox"></div>
|
||
<article class="bd-article">
|
||
|
||
<h1>Source code for tensorrt_llm.models.eagle.model</h1><div class="highlight"><pre>
|
||
<span></span><span class="c1"># SPDX-FileCopyrightText: Copyright (c) 2022-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.</span>
|
||
<span class="c1"># SPDX-License-Identifier: Apache-2.0</span>
|
||
<span class="c1">#</span>
|
||
<span class="c1"># Licensed under the Apache License, Version 2.0 (the "License");</span>
|
||
<span class="c1"># you may not use this file except in compliance with the License.</span>
|
||
<span class="c1"># You may obtain a copy of the License at</span>
|
||
<span class="c1">#</span>
|
||
<span class="c1"># http://www.apache.org/licenses/LICENSE-2.0</span>
|
||
<span class="c1">#</span>
|
||
<span class="c1"># Unless required by applicable law or agreed to in writing, software</span>
|
||
<span class="c1"># distributed under the License is distributed on an "AS IS" BASIS,</span>
|
||
<span class="c1"># WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.</span>
|
||
<span class="c1"># See the License for the specific language governing permissions and</span>
|
||
<span class="c1"># limitations under the License.</span>
|
||
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">collections</span><span class="w"> </span><span class="kn">import</span> <span class="n">OrderedDict</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">typing</span><span class="w"> </span><span class="kn">import</span> <span class="n">Optional</span><span class="p">,</span> <span class="n">Union</span>
|
||
|
||
<span class="kn">import</span><span class="w"> </span><span class="nn">numpy</span><span class="w"> </span><span class="k">as</span><span class="w"> </span><span class="nn">np</span>
|
||
<span class="kn">import</span><span class="w"> </span><span class="nn">tensorrt</span><span class="w"> </span><span class="k">as</span><span class="w"> </span><span class="nn">trt</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">tqdm</span><span class="w"> </span><span class="kn">import</span> <span class="n">tqdm</span>
|
||
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">tensorrt_llm.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">tensorrt_llm.models.generation_mixin</span><span class="w"> </span><span class="kn">import</span> <span class="n">GenerationMixin</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">tensorrt_llm.models.llama.model</span><span class="w"> </span><span class="kn">import</span> <span class="n">LLaMAForCausalLM</span><span class="p">,</span> <span class="n">LLaMAModel</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">tensorrt_llm.models.model_weights_loader</span><span class="w"> </span><span class="kn">import</span> <span class="n">ModelWeightsLoader</span>
|
||
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">..._common</span><span class="w"> </span><span class="kn">import</span> <span class="n">default_net</span><span class="p">,</span> <span class="n">default_trtnet</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">pad_vocab_size</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">...bindings</span><span class="w"> </span><span class="kn">import</span> <span class="n">KVCacheType</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">...functional</span><span class="w"> </span><span class="kn">import</span> <span class="p">(</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">_create_tensor</span><span class="p">,</span> <span class="n">cast</span><span class="p">,</span> <span class="n">concat</span><span class="p">,</span>
|
||
<span class="n">gather_last_token_logits</span><span class="p">,</span> <span class="n">index_select</span><span class="p">,</span> <span class="n">shape</span><span class="p">)</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">...layers</span><span class="w"> </span><span class="kn">import</span> <span class="n">AttentionParams</span><span class="p">,</span> <span class="n">ColumnLinear</span><span class="p">,</span> <span class="n">SpecDecodingParams</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">...module</span><span class="w"> </span><span class="kn">import</span> <span class="n">Module</span><span class="p">,</span> <span class="n">ModuleList</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">...plugin</span><span class="w"> </span><span class="kn">import</span> <span class="n">TRT_LLM_PLUGIN_NAMESPACE</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">..modeling_utils</span><span class="w"> </span><span class="kn">import</span> <span class="n">QuantConfig</span>
|
||
<span class="kn">from</span><span class="w"> </span><span class="nn">.config</span><span class="w"> </span><span class="kn">import</span> <span class="n">EagleConfig</span>
|
||
|
||
|
||
<span class="k">class</span><span class="w"> </span><span class="nc">TreeParams</span><span class="p">(</span><span class="nb">object</span><span class="p">):</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">paths</span><span class="p">:</span> <span class="n">Tensor</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">paths</span> <span class="o">=</span> <span class="n">paths</span> <span class="c1"># on GPU</span>
|
||
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">eagle_sample_and_accept_draft_plugin</span><span class="p">(</span><span class="n">lm_logits</span><span class="p">:</span> <span class="n">Tensor</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">draft_tokens</span><span class="p">:</span> <span class="n">Tensor</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">draft_lens</span><span class="p">:</span> <span class="n">Tensor</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">eagle_temperature</span><span class="p">:</span> <span class="n">Tensor</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">rand_data_validation</span><span class="p">:</span> <span class="n">Tensor</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">posterior_alpha</span><span class="p">:</span> <span class="n">Tensor</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">posterior_threshold</span><span class="p">:</span> <span class="n">Tensor</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">tree_params</span><span class="p">:</span> <span class="n">TreeParams</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">greedy_sampling</span><span class="p">:</span> <span class="n">Tensor</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">use_dynamic_tree</span><span class="p">:</span> <span class="n">Tensor</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span>
|
||
<span class="w"> </span><span class="sd">'''</span>
|
||
<span class="sd"> Takes input logits and samples golden token + predictions from draft tokens.</span>
|
||
<span class="sd"> Runs acceptance algorithm to accept draft tokens.</span>
|
||
<span class="sd"> When greedy_sampling is True, all decoding is done using Top1 and token equality is used</span>
|
||
<span class="sd"> for acceptance. Otherwise, typical acceptance and multinomial samplings are used.</span>
|
||
|
||
<span class="sd"> Visit tests/model/eagle/test_sample_accept_draft_tokens.py for input/output examples.</span>
|
||
|
||
<span class="sd"> Parameters:</span>
|
||
<span class="sd"> lm_logits : Tensor</span>
|
||
<span class="sd"> [num_tokens, vocab_size]</span>
|
||
<span class="sd"> Logits produced by the base model.</span>
|
||
|
||
<span class="sd"> draft_tokens : Tensor</span>
|
||
<span class="sd"> [batch_size, max_decoding_draft_tokens]</span>
|
||
<span class="sd"> Input draft tokens. Only the first draft_lens[bi] tokens are relevant for bi'th row.</span>
|
||
|
||
<span class="sd"> draft_lens : Tensor</span>
|
||
<span class="sd"> [batch_size]</span>
|
||
<span class="sd"> Lengths of the draft_tokens. 0 for context request. Actual draft length for generation requests.</span>
|
||
|
||
<span class="sd"> eagle_temperature : Tensor</span>
|
||
<span class="sd"> [batch_size]</span>
|
||
<span class="sd"> Temperature of the decoding.</span>
|
||
|
||
<span class="sd"> rand_data_validation : Tensor</span>
|
||
<span class="sd"> [batch_size, max_decoding_tokens]</span>
|
||
<span class="sd"> Random data for multinomial sampling.</span>
|
||
|
||
<span class="sd"> posterior_alpha : Tensor</span>
|
||
<span class="sd"> [batch_size]</span>
|
||
<span class="sd"> Delta in typical acceptance in https://arxiv.org/pdf/2401.10774.</span>
|
||
|
||
<span class="sd"> posterior_threshold : Tensor</span>
|
||
<span class="sd"> [batch_size]</span>
|
||
<span class="sd"> Minimum probability threshold.</span>
|
||
<span class="sd"> Epsilon in typical acceptance in https://arxiv.org/pdf/2401.10774.</span>
|
||
|
||
<span class="sd"> tree_params : TreeParams</span>
|
||
<span class="sd"> Tree params of the input draft tokens.</span>
|
||
|
||
<span class="sd"> greedy_sampling : Tensor</span>
|
||
<span class="sd"> Whether to do greedy or non-greedy sampling.</span>
|
||
|
||
<span class="sd"> use_dynamic_tree: Tensor</span>
|
||
<span class="sd"> Whether to use dynamic tree (i.e., Eagle-2)</span>
|
||
|
||
|
||
<span class="sd"> Return:</span>
|
||
<span class="sd"> accepted_tokens : Tensor</span>
|
||
<span class="sd"> [batch_size, max_path_len]</span>
|
||
<span class="sd"> Accepted token ids. Only the first num_accepted_tokens[bi] tokens are relevant for bi'th row,</span>
|
||
|
||
<span class="sd"> num_accepted_tokens : Tensor</span>
|
||
<span class="sd"> [batch_size]</span>
|
||
<span class="sd"> Number of accepted tokens per request. Each entry is >= 1.</span>
|
||
|
||
<span class="sd"> accepted_paths : Tensor</span>
|
||
<span class="sd"> [batch_size]</span>
|
||
<span class="sd"> Indices of the accepted path per request of the input paths in tree_params.paths.</span>
|
||
|
||
<span class="sd"> next_draft_tokens : Tensor</span>
|
||
<span class="sd"> [batch_size, max_decoding_draft_tokens]</span>
|
||
<span class="sd"> Empty tensor used to allocate space for the next draft tokens.</span>
|
||
|
||
<span class="sd"> next_draft_lens : Tensor</span>
|
||
<span class="sd"> [batch_size]</span>
|
||
<span class="sd"> Empty tensor used to allocate space for lens of the next draft tokens.</span>
|
||
|
||
<span class="sd"> next_draft_paths : Tensor</span>
|
||
<span class="sd"> [batch_size, max_decoding_len, max_path_len]</span>
|
||
<span class="sd"> For EAGLE-1 just a copy of input path.</span>
|
||
|
||
<span class="sd"> hidden_size_batch_level_starts : Tensor</span>
|
||
<span class="sd"> [max_draft_path_len * batch_size + 1]</span>
|
||
<span class="sd"> Empty tensor used to allocate space for eagle_prepare_drafter_inputs_plugin.</span>
|
||
<span class="sd"> '''</span>
|
||
|
||
<span class="n">plg_creator</span> <span class="o">=</span> <span class="n">trt</span><span class="o">.</span><span class="n">get_plugin_registry</span><span class="p">()</span><span class="o">.</span><span class="n">get_plugin_creator</span><span class="p">(</span>
|
||
<span class="s1">'EagleSampleAndAcceptDraftTokens'</span><span class="p">,</span> <span class="s1">'1'</span><span class="p">,</span> <span class="n">TRT_LLM_PLUGIN_NAMESPACE</span><span class="p">)</span>
|
||
<span class="k">assert</span> <span class="n">plg_creator</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
|
||
|
||
<span class="n">pf_type</span> <span class="o">=</span> <span class="n">trt</span><span class="o">.</span><span class="n">PluginField</span><span class="p">(</span><span class="s2">"type_id"</span><span class="p">,</span>
|
||
<span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="nb">int</span><span class="p">(</span><span class="n">lm_logits</span><span class="o">.</span><span class="n">dtype</span><span class="p">)],</span> <span class="n">np</span><span class="o">.</span><span class="n">int32</span><span class="p">),</span>
|
||
<span class="n">trt</span><span class="o">.</span><span class="n">PluginFieldType</span><span class="o">.</span><span class="n">INT32</span><span class="p">)</span>
|
||
|
||
<span class="n">pfc</span> <span class="o">=</span> <span class="n">trt</span><span class="o">.</span><span class="n">PluginFieldCollection</span><span class="p">([</span><span class="n">pf_type</span><span class="p">])</span>
|
||
<span class="n">plugin</span> <span class="o">=</span> <span class="n">plg_creator</span><span class="o">.</span><span class="n">create_plugin</span><span class="p">(</span><span class="s2">"eagle_sample_and_accept_draft_plugin"</span><span class="p">,</span>
|
||
<span class="n">pfc</span><span class="p">)</span>
|
||
|
||
<span class="n">plug_inputs</span> <span class="o">=</span> <span class="p">[</span>
|
||
<span class="n">lm_logits</span><span class="p">,</span> <span class="n">draft_tokens</span><span class="p">,</span> <span class="n">draft_lens</span><span class="p">,</span> <span class="n">eagle_temperature</span><span class="p">,</span>
|
||
<span class="n">rand_data_validation</span><span class="p">,</span> <span class="n">posterior_alpha</span><span class="p">,</span> <span class="n">posterior_threshold</span><span class="p">,</span>
|
||
<span class="n">tree_params</span><span class="o">.</span><span class="n">paths</span><span class="p">,</span> <span class="n">greedy_sampling</span><span class="p">,</span> <span class="n">use_dynamic_tree</span>
|
||
<span class="p">]</span>
|
||
|
||
<span class="n">plug_inputs</span> <span class="o">=</span> <span class="p">[</span><span class="n">i</span><span class="o">.</span><span class="n">trt_tensor</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="n">plug_inputs</span><span class="p">]</span>
|
||
<span class="n">layer</span> <span class="o">=</span> <span class="n">default_trtnet</span><span class="p">()</span><span class="o">.</span><span class="n">add_plugin_v2</span><span class="p">(</span><span class="n">plug_inputs</span><span class="p">,</span> <span class="n">plugin</span><span class="p">)</span>
|
||
|
||
<span class="n">accepted_tokens</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">0</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">num_accepted_tokens</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">1</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">accepted_paths</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">2</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">next_draft_tokens</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">3</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">next_draft_lens</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">4</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">next_draft_paths</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">5</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">hidden_size_batch_level_starts</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">6</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="k">return</span> <span class="nb">tuple</span><span class="p">([</span>
|
||
<span class="n">accepted_tokens</span><span class="p">,</span> <span class="n">num_accepted_tokens</span><span class="p">,</span> <span class="n">accepted_paths</span><span class="p">,</span> <span class="n">next_draft_tokens</span><span class="p">,</span>
|
||
<span class="n">next_draft_lens</span><span class="p">,</span> <span class="n">next_draft_paths</span><span class="p">,</span> <span class="n">hidden_size_batch_level_starts</span>
|
||
<span class="p">])</span>
|
||
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">eagle_draft_decoder_plugin</span><span class="p">(</span>
|
||
<span class="n">layer_idx</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">num_eagle_layers</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">top_k_sampling</span><span class="p">:</span> <span class="nb">bool</span><span class="p">,</span>
|
||
<span class="n">logits</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span> <span class="n">num_last_token_indices</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span> <span class="n">input_paths</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span>
|
||
<span class="n">use_dynamic_tree</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span> <span class="n">dynamic_tree_max_topK</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span>
|
||
<span class="n">input_draft_token_ids</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span> <span class="n">input_draft_lens</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span>
|
||
<span class="n">input_prev_scores</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span> <span class="n">input_current_expand_indices</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span>
|
||
<span class="n">input_all_layers_scores</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span>
|
||
<span class="n">input_all_layers_draft_token_ids</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span>
|
||
<span class="n">input_all_layers_draft_token_ids_predecessor</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">):</span>
|
||
<span class="w"> </span><span class="sd">'''</span>
|
||
<span class="sd"> Parameters:</span>
|
||
<span class="sd"> layer_idx : int</span>
|
||
<span class="sd"> The index of the EagleNet.</span>
|
||
|
||
<span class="sd"> num_eagle_layers: int</span>
|
||
<span class="sd"> The total number of eagle layers.</span>
|
||
|
||
<span class="sd"> top_k_sampling: bool</span>
|
||
<span class="sd"> Whether to use top K sampling. Otherwise, use multinomial sampling.</span>
|
||
|
||
<span class="sd"> logits : Tensor</span>
|
||
<span class="sd"> [num_logits, vocab_size]</span>
|
||
<span class="sd"> Input logits.</span>
|
||
|
||
<span class="sd"> num_last_token_indices : Tensor</span>
|
||
<span class="sd"> [1]</span>
|
||
<span class="sd"> Number of valid logits in logits.</span>
|
||
|
||
<span class="sd"> input_paths: Tensor</span>
|
||
<span class="sd"> [batch_size, max_decoding_tokens, max_path_len]</span>
|
||
<span class="sd"> Input paths</span>
|
||
|
||
<span class="sd"> use_dynamic_tree: Tensor</span>
|
||
<span class="sd"> [1]</span>
|
||
<span class="sd"> Whether use dynamic tree (i.e., Eagle-2)</span>
|
||
|
||
<span class="sd"> dynamic_tree_max_topK: Tensor</span>
|
||
<span class="sd"> [1]</span>
|
||
<span class="sd"> Number of draft tokens expand in Eagle-2.</span>
|
||
|
||
<span class="sd"> input_draft_token_ids: Tensor</span>
|
||
<span class="sd"> [batch_size, max_decoding_draft_tokens]</span>
|
||
<span class="sd"> Draft tokens generated by previous EagleNets.</span>
|
||
|
||
<span class="sd"> input_draft_lens: Tensor</span>
|
||
<span class="sd"> [batch_size]</span>
|
||
<span class="sd"> Number of draft tokens for each request.</span>
|
||
|
||
<span class="sd"> input_prev_scores: Tensor</span>
|
||
<span class="sd"> [batch_size, max_decoding_draft_tokens]</span>
|
||
<span class="sd"> Last layer's scores</span>
|
||
|
||
<span class="sd"> input_current_expand_indices: Tensor</span>
|
||
<span class="sd"> [batch_size, max_decoding_draft_tokens]</span>
|
||
<span class="sd"> The indices of the nodes that expand in this layer.</span>
|
||
<span class="sd"> The index is related to the final output tree, which has max_decoding_draft_tokens draft tokens.</span>
|
||
|
||
<span class="sd"> input_all_layers_scores: Tensor</span>
|
||
<span class="sd"> [batch_size, num_eagle_layers, max_decoding_draft_tokens x max_decoding_draft_tokens]</span>
|
||
<span class="sd"> For Eagle-2, record scores from all EagleNets</span>
|
||
|
||
<span class="sd"> input_all_layers_draft_token_ids: Tensor</span>
|
||
<span class="sd"> [batch_size, num_eagle_layers, max_decoding_draft_tokens x max_decoding_draft_tokens]</span>
|
||
<span class="sd"> For Eagle-2, record all draft tokens from all EagleNets</span>
|
||
|
||
<span class="sd"> input_all_layers_draft_token_ids_predecessor: Tensor</span>
|
||
<span class="sd"> [batch_size, num_eagle_layers, max_decoding_draft_tokens x max_decoding_draft_tokens]</span>
|
||
<span class="sd"> For Eagle-2, record all draft tokens' predecessor</span>
|
||
|
||
<span class="sd"> Return:</span>
|
||
<span class="sd"> output_draft_token_ids: Tensor</span>
|
||
<span class="sd"> [batch_size, max_decoding_draft_tokens]</span>
|
||
<span class="sd"> Draft tokens generated by this EagleNets, also include the previous draft tokens.</span>
|
||
|
||
<span class="sd"> output_draft_draft_lens: Tensor</span>
|
||
<span class="sd"> [batch_size]</span>
|
||
<span class="sd"> Number of draft tokens for each request.</span>
|
||
|
||
<span class="sd"> output_paths: Tensor</span>
|
||
<span class="sd"> [batch_size, max_decoding_draft_tokens, max_path_len]</span>
|
||
<span class="sd"> The latest path.</span>
|
||
|
||
<span class="sd"> output_current_scores: Tensor</span>
|
||
<span class="sd"> [batch_size, max_decoding_draft_tokens]</span>
|
||
<span class="sd"> This layer's scores, which will be used in next layer.</span>
|
||
|
||
<span class="sd"> output_next_expand_indices:</span>
|
||
<span class="sd"> [batch_size, max_decoding_draft_tokens]</span>
|
||
<span class="sd"> The indices of the nodes that expand in next layer.</span>
|
||
<span class="sd"> The index is related to the final output tree, which has max_decoding_draft_tokens draft tokens.</span>
|
||
|
||
<span class="sd"> output_all_layers_scores:</span>
|
||
<span class="sd"> [batch_size, num_eagle_layers, max_decoding_draft_tokens x max_decoding_draft_tokens]</span>
|
||
<span class="sd"> For Eagle-2, record scores from all EagleNets</span>
|
||
|
||
<span class="sd"> output_all_layers_draft_token_ids: Tensor</span>
|
||
<span class="sd"> [batch_size, num_eagle_layers, max_decoding_draft_tokens x max_decoding_draft_tokens]</span>
|
||
<span class="sd"> For Eagle-2, record all draft tokens from all EagleNets</span>
|
||
|
||
<span class="sd"> output_all_layers_draft_token_ids_predecessor: Tensor</span>
|
||
<span class="sd"> [batch_size, num_eagle_layers, max_decoding_draft_tokens x max_decoding_draft_tokens]</span>
|
||
<span class="sd"> For Eagle-2, record all draft tokens' predecessor</span>
|
||
|
||
<span class="sd"> '''</span>
|
||
|
||
<span class="n">plg_creator</span> <span class="o">=</span> <span class="n">trt</span><span class="o">.</span><span class="n">get_plugin_registry</span><span class="p">()</span><span class="o">.</span><span class="n">get_plugin_creator</span><span class="p">(</span>
|
||
<span class="s1">'EagleDecodeDraftTokens'</span><span class="p">,</span> <span class="s1">'1'</span><span class="p">,</span> <span class="n">TRT_LLM_PLUGIN_NAMESPACE</span><span class="p">)</span>
|
||
<span class="k">assert</span> <span class="n">plg_creator</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
|
||
|
||
<span class="n">pf_type</span> <span class="o">=</span> <span class="n">trt</span><span class="o">.</span><span class="n">PluginField</span><span class="p">(</span><span class="s2">"type_id"</span><span class="p">,</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="nb">int</span><span class="p">(</span><span class="n">logits</span><span class="o">.</span><span class="n">dtype</span><span class="p">)],</span>
|
||
<span class="n">np</span><span class="o">.</span><span class="n">int32</span><span class="p">),</span>
|
||
<span class="n">trt</span><span class="o">.</span><span class="n">PluginFieldType</span><span class="o">.</span><span class="n">INT32</span><span class="p">)</span>
|
||
|
||
<span class="n">layer_idx_t</span> <span class="o">=</span> <span class="n">trt</span><span class="o">.</span><span class="n">PluginField</span><span class="p">(</span><span class="s2">"layer_idx"</span><span class="p">,</span>
|
||
<span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span><span class="n">layer_idx</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">int32</span><span class="p">),</span>
|
||
<span class="n">trt</span><span class="o">.</span><span class="n">PluginFieldType</span><span class="o">.</span><span class="n">INT32</span><span class="p">)</span>
|
||
|
||
<span class="n">num_eagle_layers_t</span> <span class="o">=</span> <span class="n">trt</span><span class="o">.</span><span class="n">PluginField</span><span class="p">(</span>
|
||
<span class="s2">"num_eagle_layers"</span><span class="p">,</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span><span class="n">num_eagle_layers</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">int32</span><span class="p">),</span>
|
||
<span class="n">trt</span><span class="o">.</span><span class="n">PluginFieldType</span><span class="o">.</span><span class="n">INT32</span><span class="p">)</span>
|
||
|
||
<span class="n">top_k_sampling_t</span> <span class="o">=</span> <span class="mi">1</span> <span class="k">if</span> <span class="n">top_k_sampling</span> <span class="k">else</span> <span class="mi">0</span>
|
||
<span class="n">top_k_sampling_t</span> <span class="o">=</span> <span class="n">trt</span><span class="o">.</span><span class="n">PluginField</span><span class="p">(</span>
|
||
<span class="s2">"top_k_sampling"</span><span class="p">,</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span><span class="n">top_k_sampling_t</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">int32</span><span class="p">),</span>
|
||
<span class="n">trt</span><span class="o">.</span><span class="n">PluginFieldType</span><span class="o">.</span><span class="n">INT32</span><span class="p">)</span>
|
||
|
||
<span class="n">pfc</span> <span class="o">=</span> <span class="n">trt</span><span class="o">.</span><span class="n">PluginFieldCollection</span><span class="p">(</span>
|
||
<span class="p">[</span><span class="n">pf_type</span><span class="p">,</span> <span class="n">layer_idx_t</span><span class="p">,</span> <span class="n">num_eagle_layers_t</span><span class="p">,</span> <span class="n">top_k_sampling_t</span><span class="p">])</span>
|
||
<span class="n">plugin</span> <span class="o">=</span> <span class="n">plg_creator</span><span class="o">.</span><span class="n">create_plugin</span><span class="p">(</span><span class="s2">"eagle_draft_decoder_plugin"</span><span class="p">,</span> <span class="n">pfc</span><span class="p">)</span>
|
||
|
||
<span class="n">plug_inputs</span> <span class="o">=</span> <span class="p">[</span>
|
||
<span class="n">logits</span><span class="p">,</span> <span class="n">input_paths</span><span class="p">,</span> <span class="n">num_last_token_indices</span><span class="p">,</span> <span class="n">use_dynamic_tree</span><span class="p">,</span>
|
||
<span class="n">dynamic_tree_max_topK</span><span class="p">,</span> <span class="n">input_draft_token_ids</span><span class="p">,</span> <span class="n">input_draft_lens</span><span class="p">,</span>
|
||
<span class="n">input_prev_scores</span><span class="p">,</span> <span class="n">input_current_expand_indices</span><span class="p">,</span>
|
||
<span class="n">input_all_layers_scores</span><span class="p">,</span> <span class="n">input_all_layers_draft_token_ids</span><span class="p">,</span>
|
||
<span class="n">input_all_layers_draft_token_ids_predecessor</span>
|
||
<span class="p">]</span>
|
||
|
||
<span class="n">plug_inputs</span> <span class="o">=</span> <span class="p">[</span><span class="n">i</span><span class="o">.</span><span class="n">trt_tensor</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="n">plug_inputs</span><span class="p">]</span>
|
||
<span class="n">layer</span> <span class="o">=</span> <span class="n">default_trtnet</span><span class="p">()</span><span class="o">.</span><span class="n">add_plugin_v2</span><span class="p">(</span><span class="n">plug_inputs</span><span class="p">,</span> <span class="n">plugin</span><span class="p">)</span>
|
||
|
||
<span class="n">output_draft_token_ids</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">0</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">output_draft_lens</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">1</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">output_paths</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">2</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">output_current_scores</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">3</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">output_next_expand_indices</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">4</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">output_all_layers_scores</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">5</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">output_all_layers_draft_token_ids</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">6</span><span class="p">),</span>
|
||
<span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">output_all_layers_draft_token_ids_predecessor</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span>
|
||
<span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">7</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="k">return</span> <span class="nb">tuple</span><span class="p">([</span>
|
||
<span class="n">output_draft_token_ids</span><span class="p">,</span> <span class="n">output_draft_lens</span><span class="p">,</span> <span class="n">output_paths</span><span class="p">,</span>
|
||
<span class="n">output_current_scores</span><span class="p">,</span> <span class="n">output_next_expand_indices</span><span class="p">,</span>
|
||
<span class="n">output_all_layers_scores</span><span class="p">,</span> <span class="n">output_all_layers_draft_token_ids</span><span class="p">,</span>
|
||
<span class="n">output_all_layers_draft_token_ids_predecessor</span>
|
||
<span class="p">])</span>
|
||
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">eagle_prepare_drafter_inputs_plugin</span><span class="p">(</span>
|
||
<span class="n">layer_idx</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">num_layers</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">max_non_leaves_per_layer</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span>
|
||
<span class="n">attention_params</span><span class="p">:</span> <span class="n">AttentionParams</span><span class="p">,</span> <span class="n">input_ids</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span>
|
||
<span class="n">chunked_context_next_tokens</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span> <span class="n">accepted_token_ids</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span>
|
||
<span class="n">accepted_lens</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span> <span class="n">accepted_path_ids</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span>
|
||
<span class="n">next_draft_tokens</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span> <span class="n">next_draft_lens</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span>
|
||
<span class="n">next_draft_paths</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span> <span class="n">prev_draft_lens</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span>
|
||
<span class="n">prev_draft_paths</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span> <span class="n">hidden_size_batch_level_starts</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span>
|
||
<span class="n">input_gen_tokens</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span>
|
||
<span class="n">input_spec_decoding_generation_lengths</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">):</span>
|
||
<span class="w"> </span><span class="sd">'''</span>
|
||
<span class="sd"> Prepares inputs for the EagleNet inference.</span>
|
||
|
||
<span class="sd"> Visit tests/model/eagle/test_prepare_drafter_inputs.py for input/output examples.</span>
|
||
|
||
<span class="sd"> Parameters:</span>
|
||
<span class="sd"> layer_idx : int</span>
|
||
<span class="sd"> Index of the EagleNet. 0 means context phase EagleNet or EagleNet0,</span>
|
||
<span class="sd"> > 0 means EagleNetX or generation phase of EagleNet</span>
|
||
|
||
<span class="sd"> num_layers : int</span>
|
||
<span class="sd"> Number of Eagle layers.</span>
|
||
|
||
<span class="sd"> max_non_leaves_per_layer : int</span>
|
||
<span class="sd"> Number of nodes that can be non leaf in the tree at each level of the tree.</span>
|
||
|
||
<span class="sd"> attention_params : AttentionParams</span>
|
||
|
||
<span class="sd"> input_ids : Tensor</span>
|
||
<span class="sd"> [num_tokens]</span>
|
||
<span class="sd"> Tokens ids, inputs to the base model.</span>
|
||
|
||
<span class="sd"> chunked_context_next_tokens : Tensor</span>
|
||
<span class="sd"> [batch_size]</span>
|
||
<span class="sd"> The first token of the next chunk in chunked context.</span>
|
||
<span class="sd"> -1 if current chunk is the last chunk or requests is in the gen phase.</span>
|
||
|
||
<span class="sd"> accepted_token_ids : Tensor</span>
|
||
<span class="sd"> [batch_size, max_path_len]</span>
|
||
<span class="sd"> Accepted tokens ids.</span>
|
||
|
||
<span class="sd"> accepted_lens : Tensor</span>
|
||
<span class="sd"> [batch_size]</span>
|
||
<span class="sd"> Number of accepted tokens.</span>
|
||
|
||
<span class="sd"> accepted_path_ids : Tensor</span>
|
||
<span class="sd"> [batch_size]</span>
|
||
<span class="sd"> Idx of the accepted path in prev_draft_paths.</span>
|
||
|
||
<span class="sd"> next_draft_tokens : Tensor</span>
|
||
<span class="sd"> [batch_size, max_decoding_draft_tokens]</span>
|
||
<span class="sd"> Tokens ids of the draft tokens being drafted by EagleNet</span>
|
||
|
||
<span class="sd"> next_draft_lens : Tensor</span>
|
||
<span class="sd"> [batch_size]</span>
|
||
<span class="sd"> Number of drafted tokens in next_draft_tokens</span>
|
||
|
||
<span class="sd"> next_draft_paths : Tensor</span>
|
||
<span class="sd"> [batch_size, max_decoding_tokens, max_path_len]</span>
|
||
<span class="sd"> Paths of the draft tokens for the next iteration. In EAGLE-1 is the same as prev_draft_paths</span>
|
||
|
||
<span class="sd"> prev_draft_lens : Tensor</span>
|
||
<span class="sd"> [batch_size]</span>
|
||
<span class="sd"> Number of draft tokens, inputs to the base model.</span>
|
||
<span class="sd"> 0 for ctx requests and actual draft len for gen requests.</span>
|
||
|
||
<span class="sd"> prev_draft_paths : Tensor</span>
|
||
<span class="sd"> [batch_size, max_decoding_tokens, max_path_len]</span>
|
||
<span class="sd"> Paths of the draft tokens inputs to the base model.</span>
|
||
|
||
<span class="sd"> hidden_size_batch_level_starts : Tensor</span>
|
||
<span class="sd"> [max_draft_path_len * batch_size + 1]</span>
|
||
<span class="sd"> Exclusive sum of the starts of the segments of the hidden states in the concatenated array.</span>
|
||
<span class="sd"> Hidden states shape is (flattened and w/o padding)</span>
|
||
<span class="sd"> [max_draft_path_len, batch_size, num_output_tokens_i_j], where num_output_tokens_i_j</span>
|
||
<span class="sd"> depends on the path of request j at level i.</span>
|
||
|
||
<span class="sd"> input_gen_tokens : Tensor</span>
|
||
<span class="sd"> [num_gen_tokens]</span>
|
||
<span class="sd"> Only needed to infer number of generation tokens from its shape. The content is irrelevant</span>
|
||
|
||
<span class="sd"> input_spec_decoding_generation_lengths : Tensor</span>
|
||
<span class="sd"> [num_gen_requests]</span>
|
||
<span class="sd"> Number of tokens for the base model. Only used to infer num_gen_requests from its shape, the content is irrelevant.</span>
|
||
|
||
<span class="sd"> Return:</span>
|
||
<span class="sd"> sequence_length : Tensor</span>
|
||
<span class="sd"> [batch_size]</span>
|
||
<span class="sd"> Sequence length of the next EagleNet iteration.</span>
|
||
<span class="sd"> For EagleNet0 equals to the (prompt_len + num_generated_tokens + accepted_lens).</span>
|
||
<span class="sd"> For EagleNetX (X > 0) (seq_len_eagle_net_0 + spec_decoding_generation_lengths).</span>
|
||
|
||
<span class="sd"> context_length : Tensor</span>
|
||
<span class="sd"> [batch_size]</span>
|
||
<span class="sd"> Context length of the next EagleNet iteration.</span>
|
||
<span class="sd"> For EagleNet0 it is either the actual context length of the request (for ctx requests)</span>
|
||
<span class="sd"> or the number of accepted tokens in this iteration. EagleNet0's attn does chunked context attn.</span>
|
||
<span class="sd"> For EagleNetX (X > 0), context length equals to the sequence length of the EagleNet0.</span>
|
||
|
||
<span class="sd"> spec_decoding_generation_lengths : Tensor</span>
|
||
<span class="sd"> [batch_size]</span>
|
||
<span class="sd"> Only relevant for EagleNetX (X > 0).</span>
|
||
<span class="sd"> Number of draft tokens.</span>
|
||
|
||
<span class="sd"> spec_decoding_position_offsets : Tensor</span>
|
||
<span class="sd"> [batch_size, max_decoding_tokens]</span>
|
||
<span class="sd"> Only relevant for EagleNetX (X > 0).</span>
|
||
<span class="sd"> Position offsets of the selected tokens from output_ids.</span>
|
||
|
||
<span class="sd"> spec_decoding_packed_mask : Tensor</span>
|
||
<span class="sd"> [batch_size, max_decoding_tokens, ceil(max_decoding_tokens / 32)]</span>
|
||
<span class="sd"> Only relevant for EagleNetX (X > 0).</span>
|
||
<span class="sd"> uint32_t packed masks.</span>
|
||
|
||
<span class="sd"> output_ids : Tensor</span>
|
||
<span class="sd"> [batch_size * max_non_leaves_per_layer * layer_idx] for layer_idx > 0</span>
|
||
<span class="sd"> [num_tokens - num_gen_tokens + num_gen_requests * (num_layers + 1)] for layer_idx == 0</span>
|
||
<span class="sd"> Token ids selected for the EagleNet iteration.</span>
|
||
<span class="sd"> Tensor's actual size is larger than num_output_tokens.</span>
|
||
|
||
<span class="sd"> position_ids : Tensor</span>
|
||
<span class="sd"> [batch_size] for layer_idx > 0</span>
|
||
<span class="sd"> [num_tokens - num_gen_tokens + num_gen_requests * (num_layers + 1)] for layer_idx == 0</span>
|
||
<span class="sd"> Position ids of the tokens selected for the EagleNet iteration.</span>
|
||
<span class="sd"> Tensor's actual size is larger than num_output_tokens.</span>
|
||
|
||
<span class="sd"> hidden_states_indices : Tensor</span>
|
||
<span class="sd"> [batch_size * max_non_leaves_per_layer * layer_idx] for layer_idx > 0</span>
|
||
<span class="sd"> [num_tokens - num_gen_tokens + num_gen_requests * (num_layers + 1)] for layer_idx == 0</span>
|
||
<span class="sd"> Indices of the hidden states to be selected from aggregated hidden states for the next iteration.</span>
|
||
<span class="sd"> Tensor's actual size is larger than num_output_tokens.</span>
|
||
|
||
<span class="sd"> last_token_indices : Tensor</span>
|
||
<span class="sd"> [batch_size * max_non_leaves_per_layer]</span>
|
||
<span class="sd"> Indices of the hidden states to be converted to logits after the next EagleNet iteration.</span>
|
||
<span class="sd"> Tensor's actual size is larger than num_output_tokens.</span>
|
||
|
||
<span class="sd"> num_last_token_indices : Tensor</span>
|
||
<span class="sd"> []</span>
|
||
<span class="sd"> Number of logits selected after the next EagleNet iteration.</span>
|
||
<span class="sd"> Tensors containing size of the outputs of V3 plugins. 0-D tensor.</span>
|
||
|
||
<span class="sd"> out_hidden_size_batch_level_starts : Tensor</span>
|
||
<span class="sd"> [max_draft_path_len * batch_size + 1]</span>
|
||
<span class="sd"> Same as hidden_size_batch_level_starts, but with updated path lens for the next level.</span>
|
||
<span class="sd"> '''</span>
|
||
|
||
<span class="n">plg_creator</span> <span class="o">=</span> <span class="n">trt</span><span class="o">.</span><span class="n">get_plugin_registry</span><span class="p">()</span><span class="o">.</span><span class="n">get_creator</span><span class="p">(</span>
|
||
<span class="s1">'EaglePrepareDrafterInputs'</span><span class="p">,</span> <span class="s1">'1'</span><span class="p">,</span> <span class="n">TRT_LLM_PLUGIN_NAMESPACE</span><span class="p">)</span>
|
||
<span class="k">assert</span> <span class="n">plg_creator</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
|
||
|
||
<span class="n">layer_idx</span> <span class="o">=</span> <span class="n">trt</span><span class="o">.</span><span class="n">PluginField</span><span class="p">(</span><span class="s2">"layer_idx"</span><span class="p">,</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span><span class="n">layer_idx</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">int32</span><span class="p">),</span>
|
||
<span class="n">trt</span><span class="o">.</span><span class="n">PluginFieldType</span><span class="o">.</span><span class="n">INT32</span><span class="p">)</span>
|
||
|
||
<span class="n">num_layers</span> <span class="o">=</span> <span class="n">trt</span><span class="o">.</span><span class="n">PluginField</span><span class="p">(</span><span class="s2">"num_layers"</span><span class="p">,</span>
|
||
<span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span><span class="n">num_layers</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">int32</span><span class="p">),</span>
|
||
<span class="n">trt</span><span class="o">.</span><span class="n">PluginFieldType</span><span class="o">.</span><span class="n">INT32</span><span class="p">)</span>
|
||
|
||
<span class="n">max_non_leaves_per_layer</span> <span class="o">=</span> <span class="n">trt</span><span class="o">.</span><span class="n">PluginField</span><span class="p">(</span>
|
||
<span class="s2">"max_non_leaves_per_layer"</span><span class="p">,</span>
|
||
<span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span><span class="n">max_non_leaves_per_layer</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">int32</span><span class="p">),</span>
|
||
<span class="n">trt</span><span class="o">.</span><span class="n">PluginFieldType</span><span class="o">.</span><span class="n">INT32</span><span class="p">)</span>
|
||
|
||
<span class="n">pfc</span> <span class="o">=</span> <span class="n">trt</span><span class="o">.</span><span class="n">PluginFieldCollection</span><span class="p">(</span>
|
||
<span class="p">[</span><span class="n">layer_idx</span><span class="p">,</span> <span class="n">num_layers</span><span class="p">,</span> <span class="n">max_non_leaves_per_layer</span><span class="p">])</span>
|
||
<span class="n">plugin</span> <span class="o">=</span> <span class="n">plg_creator</span><span class="o">.</span><span class="n">create_plugin</span><span class="p">(</span><span class="s2">"eagle_prepare_drafter_inputs_plugin"</span><span class="p">,</span>
|
||
<span class="n">pfc</span><span class="p">,</span> <span class="n">trt</span><span class="o">.</span><span class="n">TensorRTPhase</span><span class="o">.</span><span class="n">BUILD</span><span class="p">)</span>
|
||
|
||
<span class="n">plug_inputs</span> <span class="o">=</span> <span class="p">[</span>
|
||
<span class="n">attention_params</span><span class="o">.</span><span class="n">sequence_length</span><span class="p">,</span> <span class="n">attention_params</span><span class="o">.</span><span class="n">context_lengths</span><span class="p">,</span>
|
||
<span class="n">input_ids</span><span class="p">,</span> <span class="n">chunked_context_next_tokens</span><span class="p">,</span> <span class="n">accepted_token_ids</span><span class="p">,</span>
|
||
<span class="n">accepted_lens</span><span class="p">,</span> <span class="n">accepted_path_ids</span><span class="p">,</span> <span class="n">next_draft_tokens</span><span class="p">,</span> <span class="n">next_draft_lens</span><span class="p">,</span>
|
||
<span class="n">next_draft_paths</span><span class="p">,</span> <span class="n">prev_draft_lens</span><span class="p">,</span> <span class="n">prev_draft_paths</span><span class="p">,</span>
|
||
<span class="n">hidden_size_batch_level_starts</span><span class="p">,</span> <span class="n">input_gen_tokens</span><span class="p">,</span>
|
||
<span class="n">input_spec_decoding_generation_lengths</span>
|
||
<span class="p">]</span>
|
||
|
||
<span class="n">plug_inputs</span> <span class="o">=</span> <span class="p">[</span><span class="n">i</span><span class="o">.</span><span class="n">trt_tensor</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="n">plug_inputs</span><span class="p">]</span>
|
||
<span class="n">shape_inputs</span> <span class="o">=</span> <span class="p">[]</span>
|
||
<span class="n">layer</span> <span class="o">=</span> <span class="n">default_trtnet</span><span class="p">()</span><span class="o">.</span><span class="n">add_plugin_v3</span><span class="p">(</span><span class="n">plug_inputs</span><span class="p">,</span> <span class="n">shape_inputs</span><span class="p">,</span> <span class="n">plugin</span><span class="p">)</span>
|
||
|
||
<span class="n">sequence_length</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">0</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">context_length</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">1</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">spec_decoding_generation_lengths</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">2</span><span class="p">),</span>
|
||
<span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">spec_decoding_position_offsets</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">3</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">spec_decoding_packed_mask</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">4</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">output_ids</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">5</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">position_ids</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">6</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">hidden_states_indices</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">7</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">last_token_indices</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">8</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">num_last_token_indices</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">9</span><span class="p">),</span> <span class="n">layer</span><span class="p">)</span>
|
||
<span class="n">out_hidden_size_batch_level_starts</span> <span class="o">=</span> <span class="n">_create_tensor</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">get_output</span><span class="p">(</span><span class="mi">10</span><span class="p">),</span>
|
||
<span class="n">layer</span><span class="p">)</span>
|
||
<span class="k">return</span> <span class="nb">tuple</span><span class="p">([</span>
|
||
<span class="n">sequence_length</span><span class="p">,</span> <span class="n">context_length</span><span class="p">,</span> <span class="n">spec_decoding_generation_lengths</span><span class="p">,</span>
|
||
<span class="n">spec_decoding_position_offsets</span><span class="p">,</span> <span class="n">spec_decoding_packed_mask</span><span class="p">,</span> <span class="n">output_ids</span><span class="p">,</span>
|
||
<span class="n">position_ids</span><span class="p">,</span> <span class="n">hidden_states_indices</span><span class="p">,</span> <span class="n">last_token_indices</span><span class="p">,</span>
|
||
<span class="n">num_last_token_indices</span><span class="p">,</span> <span class="n">out_hidden_size_batch_level_starts</span>
|
||
<span class="p">])</span>
|
||
|
||
|
||
<span class="k">class</span><span class="w"> </span><span class="nc">EagleNet</span><span class="p">(</span><span class="n">Module</span><span class="p">):</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">config</span><span class="p">,</span> <span class="n">logits_dtype</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="bp">self</span><span class="o">.</span><span class="n">drafter</span> <span class="o">=</span> <span class="n">LLaMAModel</span><span class="p">(</span><span class="n">config</span><span class="p">)</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">config</span> <span class="o">=</span> <span class="n">config</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">logits_dtype</span> <span class="o">=</span> <span class="n">logits_dtype</span>
|
||
|
||
<span class="n">vocab_size_padded</span> <span class="o">=</span> <span class="n">pad_vocab_size</span><span class="p">(</span><span class="n">config</span><span class="o">.</span><span class="n">vocab_size</span><span class="p">,</span>
|
||
<span class="n">config</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="k">if</span> <span class="n">config</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">is_last_pp_rank</span><span class="p">():</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">lm_head</span> <span class="o">=</span> <span class="n">ColumnLinear</span><span class="p">(</span><span class="n">config</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
||
<span class="n">vocab_size_padded</span><span class="p">,</span>
|
||
<span class="n">bias</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">config</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
||
<span class="n">tp_group</span><span class="o">=</span><span class="n">config</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
|
||
<span class="n">tp_size</span><span class="o">=</span><span class="n">config</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">gather_output</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">lm_head</span> <span class="o">=</span> <span class="kc">None</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span>
|
||
<span class="n">input_ids</span><span class="p">,</span>
|
||
<span class="n">position_ids</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">hidden_states</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">last_token_indices</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">spec_decoding_params</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">kv_cache_params</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
||
<span class="n">attention_params</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
|
||
<span class="n">hidden_states</span><span class="p">,</span> <span class="n">cache</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">drafter</span><span class="p">(</span>
|
||
<span class="n">input_ids</span><span class="p">,</span>
|
||
<span class="n">position_ids</span><span class="o">=</span><span class="n">position_ids</span><span class="p">,</span>
|
||
<span class="n">use_cache</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
||
<span class="n">spec_decoding_params</span><span class="o">=</span><span class="n">spec_decoding_params</span><span class="p">,</span>
|
||
<span class="n">kv_cache_params</span><span class="o">=</span><span class="n">kv_cache_params</span><span class="p">,</span>
|
||
<span class="n">attention_params</span><span class="o">=</span><span class="n">attention_params</span><span class="p">,</span>
|
||
<span class="n">hidden_states_for_embed</span><span class="o">=</span><span class="n">hidden_states</span><span class="p">)</span>
|
||
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">is_last_pp_rank</span><span class="p">():</span>
|
||
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">gather_last_token_logits</span><span class="p">(</span>
|
||
<span class="n">hidden_states</span><span class="p">,</span> <span class="n">last_token_indices</span><span class="p">,</span>
|
||
<span class="n">default_net</span><span class="p">()</span><span class="o">.</span><span class="n">plugin_config</span><span class="o">.</span><span class="n">remove_input_padding</span><span class="p">)</span>
|
||
<span class="k">return</span> <span class="n">cast</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">lm_head</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">),</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">logits_dtype</span><span class="p">),</span> <span class="n">hidden_states</span><span class="p">,</span> <span class="n">cache</span>
|
||
|
||
<span class="k">return</span> <span class="kc">None</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">,</span> <span class="n">cache</span>
|
||
|
||
|
||
<div class="viewcode-block" id="EagleForCausalLM">
|
||
<a class="viewcode-back" href="../../../../python-api/tensorrt_llm.models.html#tensorrt_llm.models.EagleForCausalLM">[docs]</a>
|
||
<span class="k">class</span><span class="w"> </span><span class="nc">EagleForCausalLM</span><span class="p">(</span><span class="n">LLaMAForCausalLM</span><span class="p">):</span>
|
||
<span class="n">config_class</span> <span class="o">=</span> <span class="n">EagleConfig</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">config</span><span class="p">:</span> <span class="n">EagleConfig</span><span class="p">):</span>
|
||
|
||
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">(</span><span class="n">config</span><span class="p">)</span>
|
||
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">num_eagle_layers</span> <span class="o">=</span> <span class="n">config</span><span class="o">.</span><span class="n">num_eagle_layers</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">max_non_leaves_per_layer</span> <span class="o">=</span> <span class="n">config</span><span class="o">.</span><span class="n">max_non_leaves_per_layer</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span> <span class="o">=</span> <span class="n">config</span><span class="o">.</span><span class="n">hidden_size</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">vocab_size</span> <span class="o">=</span> <span class="n">config</span><span class="o">.</span><span class="n">vocab_size</span>
|
||
<span class="n">vocab_size_padded</span> <span class="o">=</span> <span class="n">pad_vocab_size</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">vocab_size</span><span class="p">,</span>
|
||
<span class="n">config</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">eagle_net_config</span> <span class="o">=</span> <span class="n">config</span><span class="o">.</span><span class="n">eagle_net_config</span>
|
||
<span class="n">eagle_net_config</span><span class="o">.</span><span class="n">mapping</span> <span class="o">=</span> <span class="n">Mapping</span><span class="p">(</span><span class="n">world_size</span><span class="o">=</span><span class="n">config</span><span class="o">.</span><span class="n">mapping</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">config</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">rank</span><span class="p">,</span>
|
||
<span class="n">cp_size</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span>
|
||
<span class="n">tp_size</span><span class="o">=</span><span class="n">config</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">world_size</span><span class="p">,</span>
|
||
<span class="n">pp_size</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
|
||
|
||
<span class="n">eagle_net_config</span><span class="o">.</span><span class="n">fc_after_embed</span> <span class="o">=</span> <span class="kc">True</span>
|
||
<span class="n">eagle_net_config</span><span class="o">.</span><span class="n">use_input_layernorm_in_first_layer</span> <span class="o">=</span> <span class="kc">False</span>
|
||
<span class="n">eagle_net_config</span><span class="o">.</span><span class="n">use_last_layernorm</span> <span class="o">=</span> <span class="kc">False</span>
|
||
<span class="n">eagle_net_config</span><span class="o">.</span><span class="n">layer_idx_offset</span> <span class="o">=</span> <span class="n">config</span><span class="o">.</span><span class="n">num_hidden_layers</span>
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">is_last_pp_rank</span><span class="p">():</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">eagle_nets</span> <span class="o">=</span> <span class="n">ModuleList</span><span class="p">([</span>
|
||
<span class="n">EagleNet</span><span class="p">(</span><span class="n">config</span><span class="o">=</span><span class="n">eagle_net_config</span><span class="p">,</span>
|
||
<span class="n">logits_dtype</span><span class="o">=</span><span class="n">config</span><span class="o">.</span><span class="n">logits_dtype</span><span class="p">)</span>
|
||
<span class="k">for</span> <span class="n">_</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">num_eagle_layers</span><span class="p">)</span>
|
||
<span class="p">])</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">max_draft_len</span> <span class="o">=</span> <span class="n">config</span><span class="o">.</span><span class="n">max_draft_len</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">_prepare_drafter_inputs</span><span class="p">(</span>
|
||
<span class="bp">self</span><span class="p">,</span> <span class="n">layer_idx</span><span class="p">,</span> <span class="n">input_ids</span><span class="p">,</span> <span class="n">chunked_context_next_tokens</span><span class="p">,</span>
|
||
<span class="n">accepted_token_ids</span><span class="p">,</span> <span class="n">accepted_lens</span><span class="p">,</span> <span class="n">accepted_path_ids</span><span class="p">,</span>
|
||
<span class="n">next_draft_tokens</span><span class="p">,</span> <span class="n">next_draft_lens</span><span class="p">,</span> <span class="n">next_draft_paths</span><span class="p">,</span>
|
||
<span class="n">prev_draft_lens</span><span class="p">,</span> <span class="n">prev_draft_paths</span><span class="p">,</span> <span class="n">input_attention_params</span><span class="p">,</span>
|
||
<span class="n">input_kv_cache_params</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">,</span>
|
||
<span class="n">host_ctx_eagle_net_request_types</span><span class="p">,</span>
|
||
<span class="n">host_ctx_eagle_net_context_lengths</span><span class="p">,</span>
|
||
<span class="n">host_ctx_eagle_net_past_key_value_lengths</span><span class="p">,</span>
|
||
<span class="n">host_gen_eagle_net_request_types</span><span class="p">,</span>
|
||
<span class="n">host_gen_eagle_net_context_lengths</span><span class="p">,</span>
|
||
<span class="n">host_gen_eagle_net_past_key_value_lengths</span><span class="p">,</span>
|
||
<span class="n">hidden_size_batch_level_starts</span><span class="p">,</span> <span class="n">input_gen_tokens</span><span class="p">,</span>
|
||
<span class="n">input_spec_decoding_generation_lengths</span><span class="p">,</span> <span class="n">spec_decoding_use</span><span class="p">):</span>
|
||
|
||
<span class="n">drafter_inputs</span> <span class="o">=</span> <span class="n">eagle_prepare_drafter_inputs_plugin</span><span class="p">(</span>
|
||
<span class="n">layer_idx</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">num_eagle_layers</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_non_leaves_per_layer</span><span class="p">,</span>
|
||
<span class="n">input_attention_params</span><span class="p">,</span> <span class="n">input_ids</span><span class="p">,</span> <span class="n">chunked_context_next_tokens</span><span class="p">,</span>
|
||
<span class="n">accepted_token_ids</span><span class="p">,</span> <span class="n">accepted_lens</span><span class="p">,</span> <span class="n">accepted_path_ids</span><span class="p">,</span>
|
||
<span class="n">next_draft_tokens</span><span class="p">,</span> <span class="n">next_draft_lens</span><span class="p">,</span> <span class="n">next_draft_paths</span><span class="p">,</span>
|
||
<span class="n">prev_draft_lens</span><span class="p">,</span> <span class="n">prev_draft_paths</span><span class="p">,</span> <span class="n">hidden_size_batch_level_starts</span><span class="p">,</span>
|
||
<span class="n">input_gen_tokens</span><span class="p">,</span> <span class="n">input_spec_decoding_generation_lengths</span><span class="p">)</span>
|
||
|
||
<span class="n">sequence_length</span><span class="p">,</span> <span class="n">context_lengths</span><span class="p">,</span> \
|
||
<span class="n">spec_decoding_generation_lengths</span><span class="p">,</span> <span class="n">spec_decoding_position_offsets</span><span class="p">,</span> \
|
||
<span class="n">spec_decoding_packed_mask</span><span class="p">,</span> <span class="n">output_ids</span><span class="p">,</span> <span class="n">position_ids</span><span class="p">,</span> <span class="n">hidden_states_indices</span><span class="p">,</span> \
|
||
<span class="n">last_token_indices</span><span class="p">,</span> <span class="n">num_last_token_indices</span><span class="p">,</span> <span class="n">out_hidden_size_batch_level_starts</span> \
|
||
<span class="o">=</span> <span class="n">drafter_inputs</span>
|
||
|
||
<span class="n">attention_params</span> <span class="o">=</span> <span class="n">input_attention_params</span>
|
||
<span class="n">kv_cache_params</span> <span class="o">=</span> <span class="n">input_kv_cache_params</span>
|
||
<span class="n">attention_params</span><span class="o">.</span><span class="n">sequence_length</span> <span class="o">=</span> <span class="n">sequence_length</span>
|
||
<span class="n">attention_params</span><span class="o">.</span><span class="n">context_lengths</span> <span class="o">=</span> <span class="n">context_lengths</span>
|
||
|
||
<span class="k">if</span> <span class="n">layer_idx</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
|
||
<span class="n">attention_params</span><span class="o">.</span><span class="n">host_request_types</span> <span class="o">=</span> <span class="n">host_ctx_eagle_net_request_types</span>
|
||
<span class="n">attention_params</span><span class="o">.</span><span class="n">host_context_lengths</span> <span class="o">=</span> <span class="n">host_ctx_eagle_net_context_lengths</span>
|
||
<span class="n">kv_cache_params</span><span class="o">.</span><span class="n">host_past_key_value_lengths</span> <span class="o">=</span> <span class="n">host_ctx_eagle_net_past_key_value_lengths</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="n">attention_params</span><span class="o">.</span><span class="n">host_request_types</span> <span class="o">=</span> <span class="n">host_gen_eagle_net_request_types</span>
|
||
<span class="n">attention_params</span><span class="o">.</span><span class="n">host_context_lengths</span> <span class="o">=</span> <span class="n">host_gen_eagle_net_context_lengths</span>
|
||
<span class="n">kv_cache_params</span><span class="o">.</span><span class="n">host_past_key_value_lengths</span> <span class="o">=</span> <span class="n">host_gen_eagle_net_past_key_value_lengths</span>
|
||
|
||
<span class="n">spec_decoding_params</span> <span class="o">=</span> <span class="kc">None</span>
|
||
<span class="k">if</span> <span class="n">layer_idx</span> <span class="o">></span> <span class="mi">0</span><span class="p">:</span>
|
||
<span class="n">spec_decoding_params</span> <span class="o">=</span> <span class="n">SpecDecodingParams</span><span class="p">(</span>
|
||
<span class="kc">True</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_draft_len</span><span class="p">,</span> <span class="n">spec_decoding_generation_lengths</span><span class="p">,</span>
|
||
<span class="n">spec_decoding_position_offsets</span><span class="p">,</span> <span class="n">spec_decoding_packed_mask</span><span class="p">,</span>
|
||
<span class="n">spec_decoding_use</span><span class="p">)</span>
|
||
|
||
<span class="c1"># Get hidden states for accepted ids</span>
|
||
<span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_slice_hidden_states</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">,</span>
|
||
<span class="n">hidden_states_indices</span><span class="p">)</span>
|
||
|
||
<span class="n">eagle_net_inputs</span> <span class="o">=</span> <span class="p">{}</span>
|
||
<span class="n">eagle_net_inputs</span><span class="p">[</span><span class="s2">"input_ids"</span><span class="p">]</span> <span class="o">=</span> <span class="n">output_ids</span>
|
||
<span class="n">eagle_net_inputs</span><span class="p">[</span><span class="s2">"position_ids"</span><span class="p">]</span> <span class="o">=</span> <span class="n">position_ids</span>
|
||
<span class="n">eagle_net_inputs</span><span class="p">[</span><span class="s2">"last_token_indices"</span><span class="p">]</span> <span class="o">=</span> <span class="n">last_token_indices</span>
|
||
<span class="n">eagle_net_inputs</span><span class="p">[</span><span class="s2">"attention_params"</span><span class="p">]</span> <span class="o">=</span> <span class="n">attention_params</span>
|
||
<span class="n">eagle_net_inputs</span><span class="p">[</span><span class="s2">"kv_cache_params"</span><span class="p">]</span> <span class="o">=</span> <span class="n">kv_cache_params</span>
|
||
<span class="n">eagle_net_inputs</span><span class="p">[</span><span class="s2">"spec_decoding_params"</span><span class="p">]</span> <span class="o">=</span> <span class="n">spec_decoding_params</span>
|
||
<span class="n">eagle_net_inputs</span><span class="p">[</span><span class="s2">"hidden_states"</span><span class="p">]</span> <span class="o">=</span> <span class="n">hidden_states</span>
|
||
<span class="k">return</span> <span class="n">eagle_net_inputs</span><span class="p">,</span> <span class="n">out_hidden_size_batch_level_starts</span><span class="p">,</span> <span class="n">num_last_token_indices</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">_slice_hidden_states</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">,</span> <span class="n">indices</span><span class="p">):</span>
|
||
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">index_select</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">,</span> <span class="mi">0</span><span class="p">,</span> <span class="n">indices</span><span class="p">)</span>
|
||
|
||
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">hidden_states</span><span class="o">.</span><span class="n">view</span><span class="p">(</span><span class="n">concat</span><span class="p">(</span>
|
||
<span class="p">[</span><span class="n">shape</span><span class="p">(</span><span class="n">indices</span><span class="p">,</span> <span class="mi">0</span><span class="p">),</span> <span class="n">shape</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">,</span> <span class="mi">1</span><span class="p">)]),</span>
|
||
<span class="n">zero_is_placeholder</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
|
||
<span class="k">return</span> <span class="n">hidden_states</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">_eagle_fwd_helper</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">lm_logits</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">,</span> <span class="o">*</span><span class="n">args</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">):</span>
|
||
<span class="w"> </span><span class="sd">'''</span>
|
||
<span class="sd"> EAGLE inference can be viewed as</span>
|
||
<span class="sd"> TRT_Engine(Target -> Draft0 -> Draft1 -> .. -> DraftK-1) -> Runtime -> TRT_Engine(..) -> ..</span>
|
||
<span class="sd"> Target is Base model and Draft is EagleNet.</span>
|
||
|
||
<span class="sd"> Each EagleNet call can be viewed as call to Draft LLM in TensorRT-LLM.</span>
|
||
<span class="sd"> We have to</span>
|
||
<span class="sd"> 1. prepare input tensors before the EagleNet call (like in the the runtime),</span>
|
||
<span class="sd"> 2. call EagleNet,</span>
|
||
<span class="sd"> 3. decode draft tokens after the EagleNet.</span>
|
||
<span class="sd"> The only difference with normal execution of the Draft model is that in EAGLE,</span>
|
||
<span class="sd"> all these 3 things happen inside of the TensorRT engine execution.</span>
|
||
<span class="sd"> We do 1 and 3 inside of the plugins.</span>
|
||
<span class="sd"> For 1. We call eagle_prepare_drafter_inputs_plugin and for 3. eagle_draft_decoder_plugin.</span>
|
||
|
||
<span class="sd"> The first call to the EagleNet (Draft0 == EagleNet0) is the context phase.</span>
|
||
<span class="sd"> For context request we populate the KV cache of the EagleNet.</span>
|
||
<span class="sd"> For generation request that have accepted tokens we emulate KV cache reuse by doing chunked attention,</span>
|
||
<span class="sd"> where chunk is the newly accepted tokens -- all previous tokens are already in the KV cache.</span>
|
||
|
||
<span class="sd"> The following calls to the EagleNet (EagleNetX (X > 0)) are generation phase.</span>
|
||
<span class="sd"> For each EagleNetX we select tokens based on the current path which are going to be used for the generation.</span>
|
||
|
||
<span class="sd"> Let's consider an example: prompt ABCD. EAGLE-1, i.e tree is fixed for the iteration.</span>
|
||
<span class="sd"> Tree:</span>
|
||
<span class="sd"> ┌───┐</span>
|
||
<span class="sd"> │ 0 │</span>
|
||
<span class="sd"> └─┬─┘</span>
|
||
<span class="sd"> ┌─────┴─────┐</span>
|
||
<span class="sd"> ┌─┴─┐ ┌─┴─┐ ┌─┴─┐</span>
|
||
<span class="sd"> │ 1 │ │ 2 │ │ 3 │</span>
|
||
<span class="sd"> └─┬─┘ └─┬─┘ └───┘</span>
|
||
<span class="sd"> ┌─┴─┐ ┌─┴─┐</span>
|
||
<span class="sd"> │ 4 │ │ 5 │</span>
|
||
<span class="sd"> └─┬─┘ └─┬─┘</span>
|
||
<span class="sd"> ┌─┴─┐ ┌─┴─┐</span>
|
||
<span class="sd"> │ 6 │ │ 7 │</span>
|
||
<span class="sd"> └───┘ └───┘</span>
|
||
|
||
<span class="sd"> First iteration of the TRT engine. Request is context request:</span>
|
||
<span class="sd"> 1. Base model is called for [ABCD] tokens produces token E.</span>
|
||
<span class="sd"> 2. Draft0 is called for tokens [BCDE] and produces</span>
|
||
<span class="sd"> three possibilities F, G and H for positions 1, 2 and 3 respectively.</span>
|
||
<span class="sd"> 3. Since H (position 3) is a leaf, it is not chosen as the input to Draft1 inference.</span>
|
||
<span class="sd"> 4. Draft1 is called for tokens [FG] with appropriate mask of:</span>
|
||
<span class="sd"> |F|G</span>
|
||
<span class="sd"> F|1|0</span>
|
||
<span class="sd"> G|0|1</span>
|
||
<span class="sd"> It produces tokens I and J for positions 4 and 5.</span>
|
||
<span class="sd"> 6. Draft2 is called for inputs [FGIJ] with mask of</span>
|
||
<span class="sd"> |F|G|I|J</span>
|
||
<span class="sd"> F|1|0|0|0</span>
|
||
<span class="sd"> G|0|1|0|0</span>
|
||
<span class="sd"> I|1|0|1|0</span>
|
||
<span class="sd"> J|0|1|0|1</span>
|
||
<span class="sd"> Note that we could've stored FG in KV cache and provide only IJ tokens here</span>
|
||
<span class="sd"> with mask for past KV cache, but it is not supported in TensorRT-LLM attention at the moment.</span>
|
||
|
||
<span class="sd"> Draft2 produces tokens K and L at positions 6 and 7.</span>
|
||
<span class="sd"> 7. Resulting outputs are:</span>
|
||
<span class="sd"> 7.1 accepted_ids [E]</span>
|
||
<span class="sd"> 7.2 next_draft_tokens [FGHIJKL]</span>
|
||
|
||
<span class="sd"> Second iteration of the TRT engine. Request is the generation request.</span>
|
||
<span class="sd"> 1. Base model is called for [EFGHIJKL]. Let's assume that it accepts [FIKM], i.e. the left-most path in the tree.</span>
|
||
<span class="sd"> 2. Draft0 is called as context phase for [FIKM] -- to append to kv cache of the existing [BCDE].</span>
|
||
<span class="sd"> It produces tokens N, O and P for positions 1, 2 and 3.</span>
|
||
<span class="sd"> 3. Draft1 is called as generation phase for [NO] tokens.</span>
|
||
<span class="sd"> etc.</span>
|
||
|
||
<span class="sd"> '''</span>
|
||
<span class="n">input_tree_params</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s2">"tree_params"</span><span class="p">]</span>
|
||
|
||
<span class="n">draft_tokens</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s1">'draft_tokens'</span><span class="p">]</span>
|
||
<span class="n">draft_lens</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s1">'draft_lens'</span><span class="p">]</span>
|
||
<span class="n">eagle_temperature</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s1">'eagle_temperature'</span><span class="p">]</span>
|
||
<span class="n">rand_data_validation</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s1">'rand_data_validation'</span><span class="p">]</span>
|
||
<span class="n">posterior_alpha</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s1">'posterior_alpha'</span><span class="p">]</span>
|
||
<span class="n">posterior_threshold</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s1">'posterior_threshold'</span><span class="p">]</span>
|
||
<span class="n">input_ids</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s1">'input_ids'</span><span class="p">]</span>
|
||
<span class="n">chunked_context_next_tokens</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s1">'chunked_context_next_tokens'</span><span class="p">]</span>
|
||
<span class="n">host_ctx_eagle_net_request_types</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span>
|
||
<span class="s1">'host_ctx_eagle_net_request_types'</span><span class="p">]</span>
|
||
<span class="n">host_ctx_eagle_net_context_lengths</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span>
|
||
<span class="s1">'host_ctx_eagle_net_context_lengths'</span><span class="p">]</span>
|
||
<span class="n">host_ctx_eagle_net_past_key_value_lengths</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span>
|
||
<span class="s1">'host_ctx_eagle_net_past_key_value_lengths'</span><span class="p">]</span>
|
||
<span class="n">host_gen_eagle_net_request_types</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span>
|
||
<span class="s1">'host_gen_eagle_net_request_types'</span><span class="p">]</span>
|
||
<span class="n">host_gen_eagle_net_context_lengths</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span>
|
||
<span class="s1">'host_gen_eagle_net_context_lengths'</span><span class="p">]</span>
|
||
<span class="n">host_gen_eagle_net_past_key_value_lengths</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span>
|
||
<span class="s1">'host_gen_eagle_net_past_key_value_lengths'</span><span class="p">]</span>
|
||
<span class="n">input_gen_tokens</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s2">"input_gen_tokens"</span><span class="p">]</span>
|
||
<span class="n">greedy_sampling</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s2">"greedy_sampling"</span><span class="p">]</span>
|
||
|
||
<span class="c1"># Eagle-2</span>
|
||
<span class="n">use_dynamic_tree</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s1">'use_dynamic_tree'</span><span class="p">]</span>
|
||
<span class="n">dynamic_tree_max_topK</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s1">'dynamic_tree_max_topK'</span><span class="p">]</span>
|
||
<span class="n">prev_scores</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s1">'prev_scores'</span><span class="p">]</span>
|
||
<span class="n">current_expand_indices</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s1">'current_expand_indices'</span><span class="p">]</span>
|
||
<span class="n">all_layers_scores</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s1">'all_layers_scores'</span><span class="p">]</span>
|
||
<span class="n">all_layers_draft_token_ids</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s1">'all_layers_draft_token_ids'</span><span class="p">]</span>
|
||
<span class="n">all_layers_draft_token_ids_predecessor</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span>
|
||
<span class="s1">'all_layers_draft_token_ids_predecessor'</span><span class="p">]</span>
|
||
|
||
<span class="c1"># Sample target tokens and accept them</span>
|
||
<span class="c1"># next_draft_tokens, next_draft_lens, hidden_size_batch_level_starts are outputted here just to</span>
|
||
<span class="c1"># reserve the tensor with max size, which eagle_draft_decoder_plugin and</span>
|
||
<span class="c1"># eagle_prepare_drafter_inputs_plugin are going to directly write to</span>
|
||
<span class="n">output</span> <span class="o">=</span> <span class="n">eagle_sample_and_accept_draft_plugin</span><span class="p">(</span>
|
||
<span class="n">lm_logits</span><span class="p">,</span> <span class="n">draft_tokens</span><span class="p">,</span> <span class="n">draft_lens</span><span class="p">,</span> <span class="n">eagle_temperature</span><span class="p">,</span>
|
||
<span class="n">rand_data_validation</span><span class="p">,</span> <span class="n">posterior_alpha</span><span class="p">,</span> <span class="n">posterior_threshold</span><span class="p">,</span>
|
||
<span class="n">input_tree_params</span><span class="p">,</span> <span class="n">greedy_sampling</span><span class="p">,</span> <span class="n">use_dynamic_tree</span><span class="p">)</span>
|
||
<span class="n">accepted_tokens</span><span class="p">,</span> <span class="n">num_accepted_tokens</span><span class="p">,</span> <span class="n">accepted_paths</span><span class="p">,</span> <span class="n">next_draft_tokens</span><span class="p">,</span> \
|
||
<span class="n">next_draft_lens</span><span class="p">,</span> <span class="n">next_draft_paths</span><span class="p">,</span> <span class="n">hidden_size_batch_level_starts</span> <span class="o">=</span> <span class="n">output</span>
|
||
|
||
<span class="n">attention_params</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s2">"attention_params"</span><span class="p">]</span>
|
||
<span class="n">kv_cache_params</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s2">"kv_cache_params"</span><span class="p">]</span>
|
||
<span class="n">spec_decoding_params</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s2">"spec_decoding_params"</span><span class="p">]</span>
|
||
|
||
<span class="n">input_hidden_states</span> <span class="o">=</span> <span class="n">hidden_states</span>
|
||
|
||
<span class="c1"># Run EAGLE nets</span>
|
||
<span class="k">for</span> <span class="n">li</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">num_eagle_layers</span><span class="p">):</span>
|
||
<span class="c1"># Prepare EAGLE Net inputs.</span>
|
||
<span class="n">eagle_net_inputs</span><span class="p">,</span> <span class="n">hidden_size_batch_level_starts</span><span class="p">,</span> <span class="n">num_last_token_indices</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_prepare_drafter_inputs</span><span class="p">(</span>
|
||
<span class="n">layer_idx</span><span class="o">=</span><span class="n">li</span><span class="p">,</span>
|
||
<span class="n">input_ids</span><span class="o">=</span><span class="n">input_ids</span><span class="p">,</span>
|
||
<span class="n">chunked_context_next_tokens</span><span class="o">=</span><span class="n">chunked_context_next_tokens</span><span class="p">,</span>
|
||
<span class="n">accepted_token_ids</span><span class="o">=</span><span class="n">accepted_tokens</span><span class="p">,</span>
|
||
<span class="n">accepted_lens</span><span class="o">=</span><span class="n">num_accepted_tokens</span><span class="p">,</span>
|
||
<span class="n">accepted_path_ids</span><span class="o">=</span><span class="n">accepted_paths</span><span class="p">,</span>
|
||
<span class="n">next_draft_tokens</span><span class="o">=</span><span class="n">next_draft_tokens</span><span class="p">,</span>
|
||
<span class="n">next_draft_lens</span><span class="o">=</span><span class="n">next_draft_lens</span><span class="p">,</span>
|
||
<span class="n">next_draft_paths</span><span class="o">=</span><span class="n">next_draft_paths</span><span class="p">,</span>
|
||
<span class="n">prev_draft_lens</span><span class="o">=</span><span class="n">draft_lens</span><span class="p">,</span>
|
||
<span class="n">prev_draft_paths</span><span class="o">=</span><span class="n">input_tree_params</span><span class="o">.</span><span class="n">paths</span><span class="p">,</span>
|
||
<span class="n">input_attention_params</span><span class="o">=</span><span class="n">attention_params</span><span class="p">,</span>
|
||
<span class="n">input_kv_cache_params</span><span class="o">=</span><span class="n">kv_cache_params</span><span class="p">,</span>
|
||
<span class="n">hidden_states</span><span class="o">=</span><span class="n">input_hidden_states</span><span class="p">,</span>
|
||
<span class="n">host_ctx_eagle_net_request_types</span><span class="o">=</span>
|
||
<span class="n">host_ctx_eagle_net_request_types</span><span class="p">,</span>
|
||
<span class="n">host_ctx_eagle_net_context_lengths</span><span class="o">=</span>
|
||
<span class="n">host_ctx_eagle_net_context_lengths</span><span class="p">,</span>
|
||
<span class="n">host_ctx_eagle_net_past_key_value_lengths</span><span class="o">=</span>
|
||
<span class="n">host_ctx_eagle_net_past_key_value_lengths</span><span class="p">,</span>
|
||
<span class="n">host_gen_eagle_net_request_types</span><span class="o">=</span>
|
||
<span class="n">host_gen_eagle_net_request_types</span><span class="p">,</span>
|
||
<span class="n">host_gen_eagle_net_context_lengths</span><span class="o">=</span>
|
||
<span class="n">host_gen_eagle_net_context_lengths</span><span class="p">,</span>
|
||
<span class="n">host_gen_eagle_net_past_key_value_lengths</span><span class="o">=</span>
|
||
<span class="n">host_gen_eagle_net_past_key_value_lengths</span><span class="p">,</span>
|
||
<span class="n">hidden_size_batch_level_starts</span><span class="o">=</span><span class="n">hidden_size_batch_level_starts</span><span class="p">,</span>
|
||
<span class="n">input_gen_tokens</span><span class="o">=</span><span class="n">input_gen_tokens</span><span class="p">,</span>
|
||
<span class="n">input_spec_decoding_generation_lengths</span><span class="o">=</span><span class="n">spec_decoding_params</span><span class="o">.</span>
|
||
<span class="n">spec_decoding_generation_lengths</span><span class="p">,</span>
|
||
<span class="n">spec_decoding_use</span><span class="o">=</span><span class="n">spec_decoding_params</span><span class="o">.</span><span class="n">spec_decoding_use</span><span class="p">)</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">single_eagle_net_iter</span><span class="p">(</span><span class="n">next_draft_tokens</span><span class="p">,</span> <span class="n">next_draft_lens</span><span class="p">,</span>
|
||
<span class="n">next_draft_paths</span><span class="p">,</span> <span class="n">prev_scores</span><span class="p">,</span>
|
||
<span class="n">current_expand_indices</span><span class="p">,</span> <span class="n">all_layers_scores</span><span class="p">,</span>
|
||
<span class="n">all_layers_draft_token_ids</span><span class="p">,</span>
|
||
<span class="n">all_layers_draft_token_ids_predecessor</span><span class="p">):</span>
|
||
<span class="c1"># Run EAGLE Net</span>
|
||
<span class="c1"># NOTE: handle base net kv cache and eagle net kv cache are in the same tensor.</span>
|
||
<span class="c1"># EagleNet's kv cache is located starting at numBaseNetHiddenLayers in the kv tensor.</span>
|
||
<span class="n">logits</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">,</span> <span class="n">_</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">eagle_nets</span><span class="p">[</span><span class="n">li</span><span class="p">](</span>
|
||
<span class="o">**</span><span class="n">eagle_net_inputs</span><span class="p">)</span>
|
||
|
||
<span class="c1"># FIXME We need to take top_k_sampling as an input</span>
|
||
<span class="n">top_k_sampling</span> <span class="o">=</span> <span class="kc">True</span>
|
||
|
||
<span class="c1"># Decode draft tokens</span>
|
||
<span class="n">next_draft_tokens</span><span class="p">,</span> <span class="n">next_draft_lens</span><span class="p">,</span> <span class="n">next_draft_paths</span><span class="p">,</span> <span class="n">prev_scores</span><span class="p">,</span> <span class="n">current_expand_indices</span><span class="p">,</span> <span class="n">all_layers_scores</span><span class="p">,</span> <span class="n">all_layers_draft_token_ids</span><span class="p">,</span> <span class="n">all_layers_draft_token_ids_predecessor</span> <span class="o">=</span> <span class="n">eagle_draft_decoder_plugin</span><span class="p">(</span>
|
||
<span class="n">li</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">num_eagle_layers</span><span class="p">,</span> <span class="n">top_k_sampling</span><span class="p">,</span> <span class="n">logits</span><span class="p">,</span>
|
||
<span class="n">num_last_token_indices</span><span class="p">,</span> <span class="n">next_draft_paths</span><span class="p">,</span> <span class="n">use_dynamic_tree</span><span class="p">,</span>
|
||
<span class="n">dynamic_tree_max_topK</span><span class="p">,</span> <span class="n">next_draft_tokens</span><span class="p">,</span> <span class="n">next_draft_lens</span><span class="p">,</span>
|
||
<span class="n">prev_scores</span><span class="p">,</span> <span class="n">current_expand_indices</span><span class="p">,</span> <span class="n">all_layers_scores</span><span class="p">,</span>
|
||
<span class="n">all_layers_draft_token_ids</span><span class="p">,</span>
|
||
<span class="n">all_layers_draft_token_ids_predecessor</span><span class="p">)</span>
|
||
|
||
<span class="k">return</span> <span class="n">next_draft_tokens</span><span class="p">,</span> <span class="n">next_draft_lens</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">,</span> <span class="n">next_draft_paths</span><span class="p">,</span> <span class="n">prev_scores</span><span class="p">,</span> <span class="n">current_expand_indices</span><span class="p">,</span> <span class="n">all_layers_scores</span><span class="p">,</span> <span class="n">all_layers_draft_token_ids</span><span class="p">,</span> <span class="n">all_layers_draft_token_ids_predecessor</span>
|
||
|
||
|
||
<span class="n">next_draft_tokens</span><span class="p">,</span> <span class="n">next_draft_lens</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">,</span> <span class="n">next_draft_paths</span><span class="p">,</span> <span class="n">prev_scores</span><span class="p">,</span> \
|
||
<span class="n">current_expand_indices</span><span class="p">,</span> <span class="n">all_layers_scores</span><span class="p">,</span> <span class="n">all_layers_draft_token_ids</span><span class="p">,</span> <span class="n">all_layers_draft_token_ids_predecessor</span> \
|
||
<span class="o">=</span> <span class="n">single_eagle_net_iter</span><span class="p">(</span><span class="n">next_draft_tokens</span><span class="p">,</span> <span class="n">next_draft_lens</span><span class="p">,</span> <span class="n">next_draft_paths</span><span class="p">,</span> \
|
||
<span class="n">prev_scores</span><span class="p">,</span> <span class="n">current_expand_indices</span><span class="p">,</span> <span class="n">all_layers_scores</span><span class="p">,</span> \
|
||
<span class="n">all_layers_draft_token_ids</span><span class="p">,</span> <span class="n">all_layers_draft_token_ids_predecessor</span><span class="p">)</span>
|
||
|
||
<span class="c1"># Update params</span>
|
||
<span class="k">if</span> <span class="n">li</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
|
||
<span class="n">eagle_net_0_sequence_length</span> <span class="o">=</span> <span class="n">eagle_net_inputs</span><span class="p">[</span>
|
||
<span class="s2">"attention_params"</span><span class="p">]</span><span class="o">.</span><span class="n">sequence_length</span>
|
||
<span class="n">input_hidden_states</span> <span class="o">=</span> <span class="n">hidden_states</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="n">input_hidden_states</span> <span class="o">=</span> <span class="n">concat</span><span class="p">(</span>
|
||
<span class="p">[</span><span class="n">input_hidden_states</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">])</span>
|
||
|
||
<span class="n">kv_cache_params</span> <span class="o">=</span> <span class="n">eagle_net_inputs</span><span class="p">[</span><span class="s2">"kv_cache_params"</span><span class="p">]</span>
|
||
<span class="n">attention_params</span> <span class="o">=</span> <span class="n">eagle_net_inputs</span><span class="p">[</span><span class="s2">"attention_params"</span><span class="p">]</span>
|
||
<span class="n">attention_params</span><span class="o">.</span><span class="n">context_lengths</span> <span class="o">=</span> <span class="n">eagle_net_0_sequence_length</span>
|
||
<span class="n">attention_params</span><span class="o">.</span><span class="n">sequence_length</span> <span class="o">=</span> <span class="n">eagle_net_0_sequence_length</span>
|
||
|
||
<span class="c1"># Mark tensors as output</span>
|
||
<span class="n">accepted_tokens</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span><span class="s1">'accepted_tokens'</span><span class="p">)</span>
|
||
<span class="n">num_accepted_tokens</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span><span class="s1">'num_accepted_tokens'</span><span class="p">)</span>
|
||
<span class="n">accepted_paths</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span><span class="s1">'accepted_paths'</span><span class="p">)</span>
|
||
<span class="n">next_draft_tokens</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span><span class="s1">'next_draft_tokens'</span><span class="p">)</span>
|
||
<span class="n">next_draft_lens</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span><span class="s1">'next_draft_lens'</span><span class="p">)</span>
|
||
<span class="n">next_draft_paths</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span><span class="s1">'next_draft_paths'</span><span class="p">)</span>
|
||
|
||
<span class="k">return</span> <span class="n">next_draft_tokens</span>
|
||
|
||
<div class="viewcode-block" id="EagleForCausalLM.forward">
|
||
<a class="viewcode-back" href="../../../../python-api/tensorrt_llm.models.html#tensorrt_llm.models.EagleForCausalLM.forward">[docs]</a>
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="o">*</span><span class="n">args</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">):</span>
|
||
<span class="n">extra_args</span> <span class="o">=</span> <span class="p">[</span>
|
||
<span class="s2">"draft_tokens"</span><span class="p">,</span> <span class="s2">"draft_lens"</span><span class="p">,</span> <span class="s2">"eagle_temperature"</span><span class="p">,</span>
|
||
<span class="s2">"rand_data_validation"</span><span class="p">,</span> <span class="s2">"tree_params"</span><span class="p">,</span>
|
||
<span class="s2">"host_ctx_eagle_net_request_types"</span><span class="p">,</span>
|
||
<span class="s2">"host_ctx_eagle_net_context_lengths"</span><span class="p">,</span>
|
||
<span class="s2">"host_ctx_eagle_net_past_key_value_lengths"</span><span class="p">,</span>
|
||
<span class="s2">"host_gen_eagle_net_request_types"</span><span class="p">,</span>
|
||
<span class="s2">"host_gen_eagle_net_context_lengths"</span><span class="p">,</span>
|
||
<span class="s2">"host_gen_eagle_net_past_key_value_lengths"</span><span class="p">,</span> <span class="s2">"input_gen_tokens"</span><span class="p">,</span>
|
||
<span class="s2">"chunked_context_next_tokens"</span><span class="p">,</span> <span class="s2">"posterior_alpha"</span><span class="p">,</span>
|
||
<span class="s2">"posterior_threshold"</span><span class="p">,</span> <span class="s2">"greedy_sampling"</span><span class="p">,</span> <span class="s2">"use_dynamic_tree"</span><span class="p">,</span>
|
||
<span class="s2">"dynamic_tree_max_topK"</span><span class="p">,</span> <span class="s2">"prev_scores"</span><span class="p">,</span> <span class="s2">"current_expand_indices"</span><span class="p">,</span>
|
||
<span class="s2">"all_layers_scores"</span><span class="p">,</span> <span class="s2">"all_layers_draft_token_ids"</span><span class="p">,</span>
|
||
<span class="s2">"all_layers_draft_token_ids_predecessor"</span>
|
||
<span class="p">]</span>
|
||
|
||
<span class="n">base_kwargs</span> <span class="o">=</span> <span class="p">{</span><span class="n">k</span><span class="p">:</span> <span class="n">v</span> <span class="k">for</span> <span class="n">k</span><span class="p">,</span> <span class="n">v</span> <span class="ow">in</span> <span class="n">kwargs</span><span class="o">.</span><span class="n">items</span><span class="p">()</span> <span class="k">if</span> <span class="n">k</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">extra_args</span><span class="p">}</span>
|
||
|
||
<span class="c1"># Base model forward</span>
|
||
<span class="n">hidden_states</span> <span class="o">=</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="n">forward</span><span class="p">(</span><span class="o">*</span><span class="n">args</span><span class="p">,</span> <span class="o">**</span><span class="n">base_kwargs</span><span class="p">)</span>
|
||
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">is_last_pp_rank</span><span class="p">():</span>
|
||
<span class="n">extra_args</span> <span class="o">=</span> <span class="p">[</span><span class="s2">"hidden_states"</span><span class="p">]</span>
|
||
<span class="n">kwargs</span> <span class="o">=</span> <span class="p">{</span><span class="n">k</span><span class="p">:</span> <span class="n">v</span> <span class="k">for</span> <span class="n">k</span><span class="p">,</span> <span class="n">v</span> <span class="ow">in</span> <span class="n">kwargs</span><span class="o">.</span><span class="n">items</span><span class="p">()</span> <span class="k">if</span> <span class="n">k</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">extra_args</span><span class="p">}</span>
|
||
|
||
<span class="k">assert</span> <span class="n">kwargs</span><span class="p">[</span><span class="s1">'use_cache'</span><span class="p">]</span> <span class="ow">and</span> <span class="n">default_net</span><span class="p">(</span>
|
||
<span class="p">)</span><span class="o">.</span><span class="n">plugin_config</span><span class="o">.</span><span class="n">paged_kv_cache</span>
|
||
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">is_last_pp_rank</span><span class="p">():</span>
|
||
<span class="n">lm_logits</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">,</span> <span class="n">all_hidden_states</span> <span class="o">=</span> <span class="n">hidden_states</span>
|
||
<span class="n">lm_logits</span> <span class="o">=</span> <span class="n">cast</span><span class="p">(</span><span class="n">lm_logits</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">logits_dtype</span><span class="p">)</span>
|
||
<span class="c1"># Call eagle logic to accept prev draft tokens and predict next draft tokens</span>
|
||
<span class="n">next_draft_tokens</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_eagle_fwd_helper</span><span class="p">(</span><span class="n">lm_logits</span><span class="p">,</span>
|
||
<span class="n">all_hidden_states</span><span class="p">,</span> <span class="o">*</span><span class="n">args</span><span class="p">,</span>
|
||
<span class="o">**</span><span class="n">kwargs</span><span class="p">)</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="n">hidden_states</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span><span class="s1">'hidden_states_output'</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span>
|
||
|
||
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">is_last_pp_rank</span><span class="p">():</span>
|
||
<span class="k">return</span> <span class="n">next_draft_tokens</span>
|
||
<span class="k">return</span> <span class="n">hidden_states</span></div>
|
||
|
||
|
||
<div class="viewcode-block" id="EagleForCausalLM.prepare_inputs">
|
||
<a class="viewcode-back" href="../../../../python-api/tensorrt_llm.models.html#tensorrt_llm.models.EagleForCausalLM.prepare_inputs">[docs]</a>
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">prepare_inputs</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="o">*</span><span class="n">args</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">):</span>
|
||
<span class="w"> </span><span class="sd">"""</span>
|
||
<span class="sd"> Inputs needed:</span>
|
||
<span class="sd"> device_request_types: [bs]</span>
|
||
<span class="sd"> draft_tokens: [bs, max_draft_len]</span>
|
||
<span class="sd"> draft_lens: [bs]</span>
|
||
<span class="sd"> spec_decoding_generation_lengths: [bs]</span>
|
||
<span class="sd"> spec_decoding_position_offsets: [bs, max_gen_tokens]</span>
|
||
<span class="sd"> spec_decoding_packed_mask: [bs, max_draft_len, packed_length] **</span>
|
||
<span class="sd"> eagle_temperature: [bs]</span>
|
||
<span class="sd"> rand_data_validation: [bs, max_draft_len]</span>
|
||
|
||
<span class="sd"> ** The mask is tricky since the boolean mask will need to be</span>
|
||
<span class="sd"> packed in runtime. So, the last dim will be:</span>
|
||
<span class="sd"> packed_length = ceil((max_draft_len+1)/32)</span>
|
||
<span class="sd"> """</span>
|
||
<span class="n">default_range</span> <span class="o">=</span> <span class="n">GenerationMixin</span><span class="o">.</span><span class="n">default_range</span>
|
||
<span class="n">remove_input_padding</span> <span class="o">=</span> <span class="n">default_net</span><span class="p">()</span><span class="o">.</span><span class="n">plugin_config</span><span class="o">.</span><span class="n">remove_input_padding</span>
|
||
<span class="n">use_gpt_attention_plugin</span> <span class="o">=</span> <span class="n">default_net</span><span class="p">(</span>
|
||
<span class="p">)</span><span class="o">.</span><span class="n">plugin_config</span><span class="o">.</span><span class="n">gpt_attention_plugin</span>
|
||
<span class="n">use_gemm_plugin</span> <span class="o">=</span> <span class="n">default_net</span><span class="p">()</span><span class="o">.</span><span class="n">plugin_config</span><span class="o">.</span><span class="n">gemm_plugin</span>
|
||
<span class="n">paged_kv_cache</span> <span class="o">=</span> <span class="n">default_net</span><span class="p">()</span><span class="o">.</span><span class="n">plugin_config</span><span class="o">.</span><span class="n">paged_kv_cache</span>
|
||
<span class="n">multiple_profiles</span> <span class="o">=</span> <span class="n">default_net</span><span class="p">()</span><span class="o">.</span><span class="n">plugin_config</span><span class="o">.</span><span class="n">multiple_profiles</span>
|
||
<span class="n">max_batch_size</span> <span class="o">=</span> <span class="n">kwargs</span><span class="p">[</span><span class="s1">'max_batch_size'</span><span class="p">]</span>
|
||
<span class="k">assert</span> <span class="n">max_batch_size</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
|
||
<span class="n">gt_range</span> <span class="o">=</span> <span class="n">default_range</span><span class="p">(</span><span class="n">max_batch_size</span> <span class="o">*</span> <span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">max_draft_len</span> <span class="o">+</span> <span class="mi">1</span><span class="p">),</span>
|
||
<span class="n">min_range</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span>
|
||
<span class="n">opt_offset</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
|
||
|
||
<span class="n">kwargs</span><span class="p">[</span><span class="s1">'speculative_decoding_draft_tokens_external'</span><span class="p">]</span> <span class="o">=</span> <span class="kc">False</span>
|
||
<span class="n">kwargs</span><span class="p">[</span><span class="s1">'max_draft_len'</span><span class="p">]</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">max_draft_len</span>
|
||
<span class="n">kwargs</span><span class="p">[</span><span class="s1">'spec_decoding_is_generation_length_variable'</span><span class="p">]</span> <span class="o">=</span> <span class="kc">True</span>
|
||
<span class="n">kwargs</span><span class="p">[</span>
|
||
<span class="s1">'num_hidden_layers'</span><span class="p">]</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">num_hidden_layers</span> <span class="o">+</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">eagle_net_config</span><span class="o">.</span><span class="n">num_hidden_layers</span>
|
||
|
||
<span class="c1"># Call base class prepare inputs</span>
|
||
<span class="n">inputs</span> <span class="o">=</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="n">prepare_inputs</span><span class="p">(</span><span class="o">*</span><span class="n">args</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">)</span>
|
||
|
||
<span class="k">assert</span> <span class="n">inputs</span><span class="p">[</span><span class="s1">'spec_decoding_params'</span><span class="p">]</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
|
||
|
||
<span class="n">kv_cache_type</span> <span class="o">=</span> <span class="n">KVCacheType</span><span class="o">.</span><span class="n">PAGED</span> <span class="k">if</span> <span class="n">paged_kv_cache</span> <span class="k">else</span> <span class="n">KVCacheType</span><span class="o">.</span><span class="n">CONTINUOUS</span>
|
||
<span class="n">enable_ctx_gen_opt_profiles</span> <span class="o">=</span> <span class="n">GenerationMixin</span><span class="o">.</span><span class="n">has_ctx_gen_opt_profiles</span><span class="p">(</span>
|
||
<span class="n">use_gpt_attention_plugin</span><span class="o">=</span><span class="n">use_gpt_attention_plugin</span><span class="p">,</span>
|
||
<span class="n">use_gemm_plugin</span><span class="o">=</span><span class="n">use_gemm_plugin</span><span class="p">,</span>
|
||
<span class="n">remove_input_padding</span><span class="o">=</span><span class="n">remove_input_padding</span><span class="p">,</span>
|
||
<span class="n">kv_cache_type</span><span class="o">=</span><span class="n">kv_cache_type</span><span class="p">)</span>
|
||
|
||
<span class="n">num_profiles</span><span class="p">,</span> <span class="n">ranges</span> <span class="o">=</span> <span class="n">GenerationMixin</span><span class="o">.</span><span class="n">get_profiles_ranges</span><span class="p">(</span>
|
||
<span class="n">max_batch_size</span><span class="o">=</span><span class="n">max_batch_size</span><span class="p">,</span>
|
||
<span class="n">max_beam_width</span><span class="o">=</span><span class="n">kwargs</span><span class="p">[</span><span class="s1">'max_beam_width'</span><span class="p">],</span>
|
||
<span class="n">max_input_len</span><span class="o">=</span><span class="n">kwargs</span><span class="p">[</span><span class="s1">'max_input_len'</span><span class="p">],</span>
|
||
<span class="n">max_num_tokens</span><span class="o">=</span><span class="n">kwargs</span><span class="p">[</span><span class="s1">'max_num_tokens'</span><span class="p">],</span>
|
||
<span class="n">max_draft_len</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">max_draft_len</span><span class="p">,</span>
|
||
<span class="n">opt_batch_size</span><span class="o">=</span><span class="kc">None</span>
|
||
<span class="k">if</span> <span class="s1">'opt_batch_size'</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">kwargs</span> <span class="k">else</span> <span class="n">kwargs</span><span class="p">[</span><span class="s1">'opt_batch_size'</span><span class="p">],</span>
|
||
<span class="n">opt_num_tokens</span><span class="o">=</span><span class="kc">None</span>
|
||
<span class="k">if</span> <span class="s1">'opt_num_tokens'</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">kwargs</span> <span class="k">else</span> <span class="n">kwargs</span><span class="p">[</span><span class="s1">'opt_num_tokens'</span><span class="p">],</span>
|
||
<span class="n">enable_ctx_gen_opt_profiles</span><span class="o">=</span><span class="n">enable_ctx_gen_opt_profiles</span><span class="p">,</span>
|
||
<span class="n">multiple_profiles</span><span class="o">=</span><span class="n">multiple_profiles</span><span class="p">,</span>
|
||
<span class="n">kv_cache_type</span><span class="o">=</span><span class="n">kv_cache_type</span><span class="p">)</span>
|
||
|
||
<span class="n">bb_range</span> <span class="o">=</span> <span class="n">ranges</span><span class="p">[</span><span class="s1">'bb_range'</span><span class="p">]</span>
|
||
|
||
<span class="n">draft_len_range</span> <span class="o">=</span> <span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">max_draft_len</span> <span class="k">for</span> <span class="n">_</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">bb_range</span><span class="p">))]</span>
|
||
<span class="n">decoding_len_range</span> <span class="o">=</span> <span class="p">[(</span><span class="bp">self</span><span class="o">.</span><span class="n">max_draft_len</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)</span>
|
||
<span class="k">for</span> <span class="n">_</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">bb_range</span><span class="p">))]</span>
|
||
<span class="n">path_len_range</span> <span class="o">=</span> <span class="p">[(</span><span class="bp">self</span><span class="o">.</span><span class="n">num_eagle_layers</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)</span>
|
||
<span class="k">for</span> <span class="n">_</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">bb_range</span><span class="p">))]</span>
|
||
<span class="n">gen_tokens_range</span> <span class="o">=</span> <span class="p">[</span><span class="n">gt_range</span> <span class="k">for</span> <span class="n">_</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">bb_range</span><span class="p">))]</span>
|
||
<span class="n">num_eagle_layers_range</span> <span class="o">=</span> <span class="p">[</span>
|
||
<span class="bp">self</span><span class="o">.</span><span class="n">num_eagle_layers</span> <span class="k">for</span> <span class="n">_</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">bb_range</span><span class="p">))</span>
|
||
<span class="p">]</span>
|
||
<span class="n">draft_len_square_range</span> <span class="o">=</span> <span class="p">[(</span><span class="bp">self</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">max_draft_len</span><span class="p">)</span>
|
||
<span class="k">for</span> <span class="n">_</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">bb_range</span><span class="p">))]</span>
|
||
|
||
<span class="n">draft_tokens</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">'draft_tokens'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</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">max_draft_len</span><span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s1">'batch_size'</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">(</span><span class="s1">'draft_len'</span><span class="p">,</span> <span class="n">draft_len_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
<span class="n">draft_lens</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">'draft_lens'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s1">'batch_size'</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
<span class="n">eagle_temperature</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">'eagle_temperature'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">float32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s2">"batch_size"</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
<span class="n">rand_data_validation</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">'rand_data_validation'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">float32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</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">max_draft_len</span> <span class="o">+</span> <span class="mi">1</span><span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s1">'batch_size'</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">(</span><span class="s1">'decoding_len'</span><span class="p">,</span> <span class="n">decoding_len_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
<span class="n">posterior_alpha</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">'posterior_alpha'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">float32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s2">"batch_size"</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
<span class="n">posterior_threshold</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">'posterior_threshold'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">float32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s2">"batch_size"</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
<span class="n">draft_paths</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
|
||
<span class="n">name</span><span class="o">=</span><span class="s1">'draft_paths'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</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">max_draft_len</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">num_eagle_layers</span> <span class="o">+</span> <span class="mi">1</span><span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s1">'batch_size'</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">(</span><span class="s1">'decoding_len'</span><span class="p">,</span> <span class="n">decoding_len_range</span><span class="p">),</span>
|
||
<span class="p">(</span><span class="s1">'path_len'</span><span class="p">,</span> <span class="n">path_len_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
|
||
<span class="n">host_ctx_eagle_net_request_types</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
|
||
<span class="n">name</span><span class="o">=</span><span class="s1">'host_ctx_eagle_net_request_types'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s1">'batch_size'</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
<span class="n">host_ctx_eagle_net_context_lengths</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
|
||
<span class="n">name</span><span class="o">=</span><span class="s1">'host_ctx_eagle_net_context_lengths'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s1">'batch_size'</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
<span class="n">host_ctx_eagle_net_past_key_value_lengths</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
|
||
<span class="n">name</span><span class="o">=</span><span class="s1">'host_ctx_eagle_net_past_key_value_lengths'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s1">'batch_size'</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
<span class="n">host_gen_eagle_net_request_types</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
|
||
<span class="n">name</span><span class="o">=</span><span class="s1">'host_gen_eagle_net_request_types'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s1">'batch_size'</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
<span class="n">host_gen_eagle_net_context_lengths</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
|
||
<span class="n">name</span><span class="o">=</span><span class="s1">'host_gen_eagle_net_context_lengths'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s1">'batch_size'</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
<span class="n">host_gen_eagle_net_past_key_value_lengths</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
|
||
<span class="n">name</span><span class="o">=</span><span class="s1">'host_gen_eagle_net_past_key_value_lengths'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s1">'batch_size'</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
|
||
<span class="n">input_gen_tokens</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">'input_gen_tokens'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s1">'gen_tokens'</span><span class="p">,</span> <span class="n">gen_tokens_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
<span class="n">chunked_context_next_tokens</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">'chunked_context_next_tokens'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s1">'batch_size'</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
<span class="n">greedy_sampling</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">'greedy_sampling'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">1</span><span class="p">])</span>
|
||
|
||
<span class="n">use_dynamic_tree</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">'use_dynamic_tree'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">1</span><span class="p">])</span>
|
||
|
||
<span class="n">dynamic_tree_max_topK</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">'dynamic_tree_max_topK'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">1</span><span class="p">])</span>
|
||
|
||
<span class="n">prev_scores</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">'prev_scores'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">float32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</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">max_draft_len</span><span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s1">'batch_size'</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">(</span><span class="s1">'draft_len'</span><span class="p">,</span> <span class="n">draft_len_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
|
||
<span class="n">current_expand_indices</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">'current_expand_indices'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</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">max_draft_len</span><span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s1">'batch_size'</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">(</span><span class="s1">'draft_len'</span><span class="p">,</span> <span class="n">draft_len_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
|
||
<span class="n">all_layers_scores</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">'all_layers_scores'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">float32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</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">num_eagle_layers</span><span class="p">,</span>
|
||
<span class="bp">self</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">max_draft_len</span>
|
||
<span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s1">'batch_size'</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">(</span><span class="s1">'num_eagle_layers'</span><span class="p">,</span>
|
||
<span class="n">num_eagle_layers_range</span><span class="p">),</span>
|
||
<span class="p">(</span><span class="s1">'draft_len_square'</span><span class="p">,</span>
|
||
<span class="n">draft_len_square_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
|
||
<span class="n">all_layers_draft_token_ids</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
|
||
<span class="n">name</span><span class="o">=</span><span class="s1">'all_layers_draft_token_ids'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</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">num_eagle_layers</span><span class="p">,</span>
|
||
<span class="bp">self</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">max_draft_len</span>
|
||
<span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s1">'batch_size'</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">(</span><span class="s1">'num_eagle_layers'</span><span class="p">,</span> <span class="n">num_eagle_layers_range</span><span class="p">),</span>
|
||
<span class="p">(</span><span class="s1">'draft_len_square'</span><span class="p">,</span> <span class="n">draft_len_square_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
|
||
<span class="n">all_layers_draft_token_ids_predecessor</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
|
||
<span class="n">name</span><span class="o">=</span><span class="s1">'all_layers_draft_token_ids_predecessor'</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
|
||
<span class="n">shape</span><span class="o">=</span><span class="p">[</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">num_eagle_layers</span><span class="p">,</span>
|
||
<span class="bp">self</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">max_draft_len</span>
|
||
<span class="p">],</span>
|
||
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
|
||
<span class="p">(</span><span class="s1">'batch_size'</span><span class="p">,</span> <span class="n">bb_range</span><span class="p">),</span>
|
||
<span class="p">(</span><span class="s1">'num_eagle_layers'</span><span class="p">,</span> <span class="n">num_eagle_layers_range</span><span class="p">),</span>
|
||
<span class="p">(</span><span class="s1">'draft_len_square'</span><span class="p">,</span> <span class="n">draft_len_square_range</span><span class="p">),</span>
|
||
<span class="p">]))</span>
|
||
|
||
<span class="n">tree_params</span> <span class="o">=</span> <span class="n">TreeParams</span><span class="p">(</span><span class="n">paths</span><span class="o">=</span><span class="n">draft_paths</span><span class="p">)</span>
|
||
|
||
<span class="n">inputs</span><span class="p">[</span><span class="s1">'draft_tokens'</span><span class="p">]</span> <span class="o">=</span> <span class="n">draft_tokens</span>
|
||
<span class="n">inputs</span><span class="p">[</span><span class="s1">'draft_lens'</span><span class="p">]</span> <span class="o">=</span> <span class="n">draft_lens</span>
|
||
<span class="n">inputs</span><span class="p">[</span><span class="s1">'eagle_temperature'</span><span class="p">]</span> <span class="o">=</span> <span class="n">eagle_temperature</span>
|
||
<span class="n">inputs</span><span class="p">[</span><span class="s1">'posterior_alpha'</span><span class="p">]</span> <span class="o">=</span> <span class="n">posterior_alpha</span>
|
||
<span class="n">inputs</span><span class="p">[</span><span class="s1">'posterior_threshold'</span><span class="p">]</span> <span class="o">=</span> <span class="n">posterior_threshold</span>
|
||
<span class="n">inputs</span><span class="p">[</span><span class="s1">'rand_data_validation'</span><span class="p">]</span> <span class="o">=</span> <span class="n">rand_data_validation</span>
|
||
<span class="n">inputs</span><span class="p">[</span><span class="s1">'tree_params'</span><span class="p">]</span> <span class="o">=</span> <span class="n">tree_params</span>
|
||
<span class="n">inputs</span><span class="p">[</span>
|
||
<span class="s1">'host_ctx_eagle_net_request_types'</span><span class="p">]</span> <span class="o">=</span> <span class="n">host_ctx_eagle_net_request_types</span>
|
||
<span class="n">inputs</span><span class="p">[</span>
|
||
<span class="s1">'host_ctx_eagle_net_context_lengths'</span><span class="p">]</span> <span class="o">=</span> <span class="n">host_ctx_eagle_net_context_lengths</span>
|
||
<span class="n">inputs</span><span class="p">[</span>
|
||
<span class="s1">'host_ctx_eagle_net_past_key_value_lengths'</span><span class="p">]</span> <span class="o">=</span> <span class="n">host_ctx_eagle_net_past_key_value_lengths</span>
|
||
<span class="n">inputs</span><span class="p">[</span>
|
||
<span class="s1">'host_gen_eagle_net_request_types'</span><span class="p">]</span> <span class="o">=</span> <span class="n">host_gen_eagle_net_request_types</span>
|
||
<span class="n">inputs</span><span class="p">[</span>
|
||
<span class="s1">'host_gen_eagle_net_context_lengths'</span><span class="p">]</span> <span class="o">=</span> <span class="n">host_gen_eagle_net_context_lengths</span>
|
||
<span class="n">inputs</span><span class="p">[</span>
|
||
<span class="s1">'host_gen_eagle_net_past_key_value_lengths'</span><span class="p">]</span> <span class="o">=</span> <span class="n">host_gen_eagle_net_past_key_value_lengths</span>
|
||
<span class="n">inputs</span><span class="p">[</span><span class="s1">'input_gen_tokens'</span><span class="p">]</span> <span class="o">=</span> <span class="n">input_gen_tokens</span>
|
||
<span class="n">inputs</span><span class="p">[</span><span class="s1">'chunked_context_next_tokens'</span><span class="p">]</span> <span class="o">=</span> <span class="n">chunked_context_next_tokens</span>
|
||
<span class="n">inputs</span><span class="p">[</span><span class="s1">'greedy_sampling'</span><span class="p">]</span> <span class="o">=</span> <span class="n">greedy_sampling</span>
|
||
<span class="n">inputs</span><span class="p">[</span><span class="s1">'use_dynamic_tree'</span><span class="p">]</span> <span class="o">=</span> <span class="n">use_dynamic_tree</span>
|
||
<span class="n">inputs</span><span class="p">[</span><span class="s1">'dynamic_tree_max_topK'</span><span class="p">]</span> <span class="o">=</span> <span class="n">dynamic_tree_max_topK</span>
|
||
<span class="n">inputs</span><span class="p">[</span><span class="s1">'prev_scores'</span><span class="p">]</span> <span class="o">=</span> <span class="n">prev_scores</span>
|
||
<span class="n">inputs</span><span class="p">[</span><span class="s1">'current_expand_indices'</span><span class="p">]</span> <span class="o">=</span> <span class="n">current_expand_indices</span>
|
||
<span class="n">inputs</span><span class="p">[</span><span class="s1">'all_layers_scores'</span><span class="p">]</span> <span class="o">=</span> <span class="n">all_layers_scores</span>
|
||
<span class="n">inputs</span><span class="p">[</span><span class="s1">'all_layers_draft_token_ids'</span><span class="p">]</span> <span class="o">=</span> <span class="n">all_layers_draft_token_ids</span>
|
||
<span class="n">inputs</span><span class="p">[</span>
|
||
<span class="s1">'all_layers_draft_token_ids_predecessor'</span><span class="p">]</span> <span class="o">=</span> <span class="n">all_layers_draft_token_ids_predecessor</span>
|
||
|
||
<span class="k">return</span> <span class="n">inputs</span></div>
|
||
|
||
|
||
<div class="viewcode-block" id="EagleForCausalLM.from_hugging_face">
|
||
<a class="viewcode-back" href="../../../../python-api/tensorrt_llm.models.html#tensorrt_llm.models.EagleForCausalLM.from_hugging_face">[docs]</a>
|
||
<span class="nd">@classmethod</span>
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">from_hugging_face</span><span class="p">(</span>
|
||
<span class="bp">cls</span><span class="p">,</span>
|
||
<span class="n">hf_model_or_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="s1">'transformers.PreTrainedModel'</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="s1">'auto'</span><span class="p">,</span>
|
||
<span class="n">mapping</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Mapping</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</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="kc">None</span><span class="p">,</span>
|
||
<span class="o">**</span><span class="n">kwargs</span><span class="p">):</span>
|
||
<span class="k">assert</span> <span class="n">hf_model_or_dir</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
|
||
<span class="n">speculative_model_dir</span> <span class="o">=</span> <span class="n">kwargs</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s1">'speculative_model_dir'</span><span class="p">,</span> <span class="kc">None</span><span class="p">)</span>
|
||
<span class="n">tllm_config</span> <span class="o">=</span> <span class="n">EagleConfig</span><span class="o">.</span><span class="n">from_hugging_face</span><span class="p">(</span><span class="n">hf_model_or_dir</span><span class="p">,</span>
|
||
<span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">,</span>
|
||
<span class="n">mapping</span><span class="o">=</span><span class="n">mapping</span><span class="p">,</span>
|
||
<span class="n">quant_config</span><span class="o">=</span><span class="n">quant_config</span><span class="p">,</span>
|
||
<span class="o">**</span><span class="n">kwargs</span><span class="p">)</span>
|
||
|
||
<span class="c1"># for rank in range(mapping.world_size):</span>
|
||
<span class="n">tllm_config</span><span class="o">.</span><span class="n">mapping</span> <span class="o">=</span> <span class="n">Mapping</span><span class="p">(</span><span class="n">world_size</span><span class="o">=</span><span class="n">mapping</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">mapping</span><span class="o">.</span><span class="n">rank</span><span class="p">,</span>
|
||
<span class="n">cp_size</span><span class="o">=</span><span class="mi">1</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">model</span> <span class="o">=</span> <span class="n">EagleForCausalLM</span><span class="p">(</span><span class="n">tllm_config</span><span class="p">)</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">check_and_update</span><span class="p">(</span><span class="n">module</span><span class="p">,</span> <span class="nb">dict</span><span class="p">):</span>
|
||
<span class="k">if</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">module</span><span class="p">,</span> <span class="s1">'tllm_to_externel_key_dict'</span><span class="p">):</span>
|
||
<span class="n">module</span><span class="o">.</span><span class="n">tllm_to_externel_key_dict</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="nb">dict</span><span class="p">)</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="n">module</span><span class="o">.</span><span class="n">tllm_to_externel_key_dict</span> <span class="o">=</span> <span class="nb">dict</span>
|
||
|
||
<span class="k">def</span><span class="w"> </span><span class="nf">copy</span><span class="p">(</span><span class="n">tensors</span><span class="p">):</span>
|
||
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">tensors</span><span class="p">,</span> <span class="nb">list</span><span class="p">):</span>
|
||
<span class="k">if</span> <span class="kc">None</span> <span class="ow">in</span> <span class="n">tensors</span><span class="p">:</span>
|
||
<span class="k">return</span> <span class="n">tensors</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="k">return</span> <span class="p">[</span><span class="n">tensor</span><span class="o">.</span><span class="n">clone</span><span class="p">()</span> <span class="k">for</span> <span class="n">tensor</span> <span class="ow">in</span> <span class="n">tensors</span><span class="p">]</span>
|
||
<span class="k">elif</span> <span class="n">tensors</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="k">return</span> <span class="n">tensors</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="k">return</span> <span class="n">tensors</span><span class="o">.</span><span class="n">clone</span><span class="p">()</span>
|
||
|
||
<span class="n">shared_weight_prefixs</span> <span class="o">=</span> <span class="p">[]</span>
|
||
<span class="n">tllm_weights</span> <span class="o">=</span> <span class="p">{}</span>
|
||
<span class="n">customized_dict</span> <span class="o">=</span> <span class="p">{</span><span class="s2">"drafter"</span><span class="p">:</span> <span class="s2">""</span><span class="p">}</span>
|
||
<span class="k">if</span> <span class="n">speculative_model_dir</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
||
<span class="c1"># Single checkpoint for ModelOpt</span>
|
||
<span class="k">for</span> <span class="n">idx</span><span class="p">,</span> <span class="n">eagle_net</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">model</span><span class="o">.</span><span class="n">eagle_nets</span><span class="p">):</span>
|
||
<span class="n">check_and_update</span><span class="p">(</span><span class="n">eagle_net</span><span class="o">.</span><span class="n">drafter</span><span class="o">.</span><span class="n">fc</span><span class="p">,</span> <span class="p">{</span><span class="s2">"fc"</span><span class="p">:</span> <span class="s2">"fc"</span><span class="p">})</span>
|
||
<span class="n">check_and_update</span><span class="p">(</span><span class="n">eagle_net</span><span class="o">.</span><span class="n">drafter</span><span class="o">.</span><span class="n">vocab_embedding</span><span class="p">,</span>
|
||
<span class="p">{</span><span class="sa">f</span><span class="s2">"eagle_nets.</span><span class="si">{</span><span class="n">idx</span><span class="si">}</span><span class="s2">"</span><span class="p">:</span> <span class="s2">"model"</span><span class="p">})</span>
|
||
<span class="n">check_and_update</span><span class="p">(</span><span class="n">eagle_net</span><span class="o">.</span><span class="n">lm_head</span><span class="p">,</span> <span class="p">{</span><span class="sa">f</span><span class="s2">"eagle_nets.</span><span class="si">{</span><span class="n">idx</span><span class="si">}</span><span class="s2">"</span><span class="p">:</span> <span class="s2">""</span><span class="p">})</span>
|
||
<span class="n">shared_weight_prefixs</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="sa">f</span><span class="s2">"eagle_nets.</span><span class="si">{</span><span class="n">idx</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span>
|
||
<span class="n">customized_dict</span><span class="p">[</span><span class="sa">f</span><span class="s1">'eagle_nets.</span><span class="si">{</span><span class="n">idx</span><span class="si">}</span><span class="s1">'</span><span class="p">]</span> <span class="o">=</span> <span class="s1">'eagle_module'</span>
|
||
<span class="n">loader</span> <span class="o">=</span> <span class="n">ModelWeightsLoader</span><span class="p">(</span><span class="n">speculative_model_dir</span><span class="p">,</span> <span class="n">customized_dict</span><span class="p">)</span>
|
||
<span class="n">loader</span><span class="o">.</span><span class="n">update_key_mapping</span><span class="p">(</span><span class="n">model</span><span class="p">)</span>
|
||
<span class="k">for</span> <span class="n">tllm_key</span><span class="p">,</span> <span class="n">_</span> <span class="ow">in</span> <span class="n">tqdm</span><span class="p">(</span><span class="n">model</span><span class="o">.</span><span class="n">named_parameters</span><span class="p">()):</span>
|
||
<span class="k">if</span> <span class="nb">any</span><span class="p">([</span>
|
||
<span class="n">tllm_key</span><span class="o">.</span><span class="n">startswith</span><span class="p">(</span><span class="n">prefix</span><span class="p">)</span>
|
||
<span class="k">for</span> <span class="n">prefix</span> <span class="ow">in</span> <span class="n">shared_weight_prefixs</span>
|
||
<span class="p">]):</span>
|
||
<span class="n">tllm_weights</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="n">loader</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="n">tllm_key</span><span class="p">,</span> <span class="n">preprocess</span><span class="o">=</span><span class="n">copy</span><span class="p">))</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="n">tllm_weights</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="n">loader</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="n">tllm_key</span><span class="p">))</span>
|
||
<span class="n">loader</span><span class="o">.</span><span class="n">fill</span><span class="p">(</span><span class="n">tllm_weights</span><span class="p">)</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="c1"># Double checkpoint for HF</span>
|
||
<span class="k">for</span> <span class="n">idx</span><span class="p">,</span> <span class="n">eagle_net</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">model</span><span class="o">.</span><span class="n">eagle_nets</span><span class="p">):</span>
|
||
<span class="n">check_and_update</span><span class="p">(</span><span class="n">eagle_net</span><span class="o">.</span><span class="n">drafter</span><span class="o">.</span><span class="n">fc</span><span class="p">,</span> <span class="p">{</span><span class="s2">"fc"</span><span class="p">:</span> <span class="s2">"fc"</span><span class="p">})</span>
|
||
<span class="n">check_and_update</span><span class="p">(</span><span class="n">eagle_net</span><span class="o">.</span><span class="n">drafter</span><span class="o">.</span><span class="n">vocab_embedding</span><span class="p">,</span>
|
||
<span class="p">{</span><span class="sa">f</span><span class="s2">"eagle_nets.</span><span class="si">{</span><span class="n">idx</span><span class="si">}</span><span class="s2">"</span><span class="p">:</span> <span class="s2">""</span><span class="p">})</span>
|
||
<span class="n">check_and_update</span><span class="p">(</span><span class="n">eagle_net</span><span class="o">.</span><span class="n">lm_head</span><span class="p">,</span> <span class="p">{</span><span class="sa">f</span><span class="s2">"eagle_nets.</span><span class="si">{</span><span class="n">idx</span><span class="si">}</span><span class="s2">"</span><span class="p">:</span> <span class="s2">""</span><span class="p">})</span>
|
||
<span class="n">shared_weight_prefixs</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="sa">f</span><span class="s2">"eagle_nets.</span><span class="si">{</span><span class="n">idx</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span>
|
||
<span class="n">customized_dict</span><span class="p">[</span><span class="sa">f</span><span class="s1">'eagle_nets.</span><span class="si">{</span><span class="n">idx</span><span class="si">}</span><span class="s1">'</span><span class="p">]</span> <span class="o">=</span> <span class="s1">''</span>
|
||
|
||
<span class="c1"># Load base model</span>
|
||
<span class="n">base_loader</span> <span class="o">=</span> <span class="n">ModelWeightsLoader</span><span class="p">(</span><span class="n">hf_model_or_dir</span><span class="p">)</span>
|
||
<span class="n">base_loader</span><span class="o">.</span><span class="n">update_key_mapping</span><span class="p">(</span><span class="n">model</span><span class="p">)</span>
|
||
<span class="k">for</span> <span class="n">tllm_key</span><span class="p">,</span> <span class="n">_</span> <span class="ow">in</span> <span class="n">tqdm</span><span class="p">(</span><span class="n">model</span><span class="o">.</span><span class="n">transformer</span><span class="o">.</span><span class="n">named_parameters</span><span class="p">()):</span>
|
||
<span class="n">tllm_weights</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="n">base_loader</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="s2">"transformer."</span> <span class="o">+</span> <span class="n">tllm_key</span><span class="p">))</span>
|
||
<span class="n">tllm_weights</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="n">base_loader</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="s2">"lm_head.weight"</span><span class="p">))</span>
|
||
<span class="c1"># for idx in range(args.num_eagle_layers):</span>
|
||
<span class="k">for</span> <span class="n">idx</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">4</span><span class="p">):</span>
|
||
<span class="n">tllm_weights</span><span class="o">.</span><span class="n">update</span><span class="p">(</span>
|
||
<span class="n">base_loader</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="sa">f</span><span class="s2">"eagle_nets.</span><span class="si">{</span><span class="n">idx</span><span class="si">}</span><span class="s2">.lm_head.weight"</span><span class="p">,</span>
|
||
<span class="n">preprocess</span><span class="o">=</span><span class="n">copy</span><span class="p">))</span>
|
||
|
||
<span class="c1"># Load eagle model</span>
|
||
<span class="n">eagle_loader</span> <span class="o">=</span> <span class="n">ModelWeightsLoader</span><span class="p">(</span><span class="nb">str</span><span class="p">(</span><span class="n">speculative_model_dir</span><span class="p">),</span>
|
||
<span class="n">customized_dict</span><span class="p">)</span>
|
||
<span class="n">eagle_loader</span><span class="o">.</span><span class="n">update_key_mapping</span><span class="p">(</span><span class="n">model</span><span class="p">)</span>
|
||
<span class="k">for</span> <span class="n">tllm_key</span><span class="p">,</span> <span class="n">_</span> <span class="ow">in</span> <span class="n">tqdm</span><span class="p">(</span><span class="n">model</span><span class="o">.</span><span class="n">eagle_nets</span><span class="o">.</span><span class="n">named_parameters</span><span class="p">()):</span>
|
||
<span class="k">if</span> <span class="ow">not</span> <span class="n">tllm_key</span><span class="o">.</span><span class="n">endswith</span><span class="p">(</span><span class="s2">"lm_head.weight"</span><span class="p">):</span>
|
||
<span class="k">if</span> <span class="nb">any</span><span class="p">([</span>
|
||
<span class="n">tllm_key</span><span class="o">.</span><span class="n">startswith</span><span class="p">(</span><span class="n">prefix</span><span class="p">)</span>
|
||
<span class="k">for</span> <span class="n">prefix</span> <span class="ow">in</span> <span class="n">shared_weight_prefixs</span>
|
||
<span class="p">]):</span>
|
||
<span class="n">tllm_weights</span><span class="o">.</span><span class="n">update</span><span class="p">(</span>
|
||
<span class="n">eagle_loader</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="s2">"eagle_nets."</span> <span class="o">+</span> <span class="n">tllm_key</span><span class="p">,</span>
|
||
<span class="n">preprocess</span><span class="o">=</span><span class="n">copy</span><span class="p">))</span>
|
||
<span class="k">else</span><span class="p">:</span>
|
||
<span class="n">tllm_weights</span><span class="o">.</span><span class="n">update</span><span class="p">(</span>
|
||
<span class="n">eagle_loader</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="s2">"eagle_nets."</span> <span class="o">+</span> <span class="n">tllm_key</span><span class="p">))</span>
|
||
<span class="n">base_loader</span><span class="o">.</span><span class="n">fill</span><span class="p">(</span><span class="n">tllm_weights</span><span class="p">)</span>
|
||
|
||
<span class="k">return</span> <span class="n">model</span></div>
|
||
</div>
|
||
|
||
</pre></div>
|
||
|
||
</article>
|
||
|
||
|
||
|
||
|
||
|
||
<footer class="prev-next-footer d-print-none">
|
||
|
||
<div class="prev-next-area">
|
||
</div>
|
||
</footer>
|
||
|
||
</div>
|
||
|
||
|
||
|
||
<div class="bd-sidebar-secondary"></div>
|
||
|
||
|
||
|
||
|
||
|
||
</div>
|
||
<footer class="bd-footer-content">
|
||
|
||
</footer>
|
||
|
||
</main>
|
||
</div>
|
||
</div>
|
||
|
||
<!-- Scripts loaded after <body> so the DOM is not blocked -->
|
||
<script defer src="../../../../_static/scripts/bootstrap.js?digest=8878045cc6db502f8baf"></script>
|
||
<script defer src="../../../../_static/scripts/pydata-sphinx-theme.js?digest=8878045cc6db502f8baf"></script>
|
||
|
||
<footer class="bd-footer">
|
||
<div class="bd-footer__inner bd-page-width">
|
||
|
||
<div class="footer-items__start">
|
||
|
||
<div class="footer-item">
|
||
<a class="footer-brand logo" href="https://www.nvidia.com">
|
||
<img src="../../../../_static/nvidia-logo-horiz-rgb-1c-blk-for-screen.svg" class="logo__image only-light" alt="NVIDIA"/>
|
||
<img src="../../../../_static/nvidia-logo-horiz-rgb-1c-wht-for-screen.svg" class="logo__image only-dark" alt="NVIDIA"/>
|
||
</a></div>
|
||
|
||
<div class="footer-item">
|
||
|
||
<div class="footer-links">
|
||
|
||
|
||
<a class="external" href="https://www.nvidia.com/en-us/about-nvidia/privacy-policy/">Privacy Policy</a>
|
||
|
|
||
|
||
|
||
|
||
<a class="external" href="https://www.nvidia.com/en-us/about-nvidia/privacy-center/">Manage My Privacy</a>
|
||
|
|
||
|
||
|
||
|
||
<a class="external" href="https://www.nvidia.com/en-us/preferences/start/">Do Not Sell or Share My Data</a>
|
||
|
|
||
|
||
|
||
|
||
<a class="external" href="https://www.nvidia.com/en-us/about-nvidia/terms-of-service/">Terms of Service</a>
|
||
|
|
||
|
||
|
||
|
||
<a class="external" href="https://www.nvidia.com/en-us/about-nvidia/accessibility/">Accessibility</a>
|
||
|
|
||
|
||
|
||
|
||
<a class="external" href="https://www.nvidia.com/en-us/about-nvidia/company-policies/">Corporate Policies</a>
|
||
|
|
||
|
||
|
||
|
||
<a class="external" href="https://www.nvidia.com/en-us/product-security/">Product Security</a>
|
||
|
|
||
|
||
|
||
|
||
<a class="external" href="https://www.nvidia.com/en-us/contact/">Contact</a>
|
||
|
||
|
||
|
||
</div>
|
||
</div>
|
||
|
||
<div class="footer-item">
|
||
|
||
|
||
|
||
|
||
<p class="copyright">
|
||
|
||
Copyright © 2025, NVidia.
|
||
<br/>
|
||
|
||
</p>
|
||
</div>
|
||
|
||
<div class="footer-item">
|
||
<div class="extra_footer">
|
||
|
||
<p>Last updated on September 02, 2025.</p>
|
||
|
||
<p>This page is generated by TensorRT-LLM commit <a href="https://github.com/NVIDIA/TensorRT-LLM/tree/e81c50d">e81c50d</a>.</p>
|
||
|
||
</div></div>
|
||
|
||
</div>
|
||
|
||
|
||
|
||
</div>
|
||
|
||
</footer>
|
||
</body>
|
||
</html> |