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

2362 lines
279 KiB
HTML
Raw Blame History

<!DOCTYPE html>
<html lang="en" data-content_root="../../" >
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>tensorrt_llm.executor &#8212; tensorrt_llm</title>
<script data-cfasync="false">
document.documentElement.dataset.mode = localStorage.getItem("mode") || "";
document.documentElement.dataset.theme = localStorage.getItem("theme") || "";
</script>
<!--
this give us a css class that will be invisible only if js is disabled
-->
<noscript>
<style>
.pst-js-only { display: none !important; }
</style>
</noscript>
<!-- Loaded before other Sphinx assets -->
<link href="../../_static/styles/theme.css?digest=8878045cc6db502f8baf" rel="stylesheet" />
<link href="../../_static/styles/pydata-sphinx-theme.css?digest=8878045cc6db502f8baf" rel="stylesheet" />
<link rel="stylesheet" type="text/css" href="../../_static/pygments.css?v=a746c00c" />
<link rel="stylesheet" type="text/css" href="../../_static/styles/nvidia-sphinx-theme.css?v=df3ac72c" />
<link rel="stylesheet" type="text/css" href="../../_static/copybutton.css?v=76b2166b" />
<!-- So that users can add custom icons -->
<script src="../../_static/scripts/fontawesome.js?digest=8878045cc6db502f8baf"></script>
<!-- Pre-loaded scripts that we'll load fully later -->
<link rel="preload" as="script" href="../../_static/scripts/bootstrap.js?digest=8878045cc6db502f8baf" />
<link rel="preload" as="script" href="../../_static/scripts/pydata-sphinx-theme.js?digest=8878045cc6db502f8baf" />
<script src="../../_static/documentation_options.js?v=5929fcd5"></script>
<script src="../../_static/doctools.js?v=888ff710"></script>
<script src="../../_static/sphinx_highlight.js?v=dc90522c"></script>
<script src="../../_static/clipboard.min.js?v=a7894cd8"></script>
<script src="../../_static/copybutton.js?v=65e89d2a"></script>
<script>DOCUMENTATION_OPTIONS.pagename = '_modules/tensorrt_llm/executor';</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="" />
</head>
<body data-bs-spy="scroll" data-bs-target=".bd-toc-nav" data-offset="180" data-bs-root-margin="0px 0px -60%" data-default-mode="">
<div id="pst-skip-link" class="skip-link d-print-none"><a href="#main-content">Skip to main content</a></div>
<div id="pst-scroll-pixel-helper"></div>
<button type="button" class="btn rounded-pill" id="pst-back-to-top">
<i class="fa-solid fa-arrow-up"></i>Back to top</button>
<dialog id="pst-search-dialog">
<form class="bd-search d-flex align-items-center"
action="../../search.html"
method="get">
<i class="fa-solid fa-magnifying-glass"></i>
<input type="search"
class="form-control"
name="q"
placeholder="Search the docs ..."
aria-label="Search the docs ..."
autocomplete="off"
autocorrect="off"
autocapitalize="off"
spellcheck="false"/>
<span class="search-button__kbd-shortcut"><kbd class="kbd-shortcut__modifier">Ctrl</kbd>+<kbd>K</kbd></span>
</form>
</dialog>
<div class="pst-async-banner-revealer d-none">
<aside id="bd-header-version-warning" class="d-none d-print-none" aria-label="Version warning"></aside>
</div>
<header class="bd-header navbar navbar-expand-lg bd-navbar d-print-none">
<div class="bd-header__inner bd-page-width">
<button class="pst-navbar-icon sidebar-toggle primary-toggle" aria-label="Site navigation">
<span class="fa-solid fa-bars"></span>
</button>
<div class="col-lg-3 navbar-header-items__start">
<div class="navbar-item">
<a class="navbar-brand logo" href="../../index.html">
<img src="../../_static/nvidia-logo-horiz-rgb-blk-for-screen.svg" class="logo__image only-light" alt="tensorrt_llm - Home"/>
<img src="../../_static/nvidia-logo-horiz-rgb-wht-for-screen.svg" class="logo__image only-dark pst-js-only" alt="tensorrt_llm - Home"/>
<p class="title logo__title">tensorrt_llm</p>
</a></div>
</div>
<div class="col-lg-9 navbar-header-items">
<div class="navbar-header-items__end">
<div class="navbar-item navbar-persistent--container">
<button class="btn search-button-field search-button__button pst-js-only" title="Search" aria-label="Search" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="fa-solid fa-magnifying-glass"></i>
<span class="search-button__default-text">Search</span>
<span class="search-button__kbd-shortcut"><kbd class="kbd-shortcut__modifier">Ctrl</kbd>+<kbd class="kbd-shortcut__modifier">K</kbd></span>
</button>
</div>
<div class="navbar-item">
<button class="btn btn-sm nav-link pst-navbar-icon theme-switch-button pst-js-only" aria-label="Color mode" data-bs-title="Color mode" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="theme-switch fa-solid fa-sun fa-lg" data-mode="light" title="Light"></i>
<i class="theme-switch fa-solid fa-moon fa-lg" data-mode="dark" title="Dark"></i>
<i class="theme-switch fa-solid fa-circle-half-stroke fa-lg" data-mode="auto" title="System Settings"></i>
</button></div>
</div>
</div>
<div class="navbar-persistent--mobile">
<button class="btn search-button-field search-button__button pst-js-only" title="Search" aria-label="Search" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="fa-solid fa-magnifying-glass"></i>
<span class="search-button__default-text">Search</span>
<span class="search-button__kbd-shortcut"><kbd class="kbd-shortcut__modifier">Ctrl</kbd>+<kbd class="kbd-shortcut__modifier">K</kbd></span>
</button>
</div>
</div>
</header>
<div class="bd-container">
<div class="bd-container__inner bd-page-width">
<dialog id="pst-primary-sidebar-modal"></dialog>
<div id="pst-primary-sidebar" class="bd-sidebar-primary bd-sidebar">
<a class="navbar-brand logo" href="../../index.html">
<img src="../../_static/nvidia-logo-horiz-rgb-blk-for-screen.svg" class="logo__image only-light" alt="tensorrt_llm - Home"/>
<img src="../../_static/nvidia-logo-horiz-rgb-wht-for-screen.svg" class="logo__image only-dark pst-js-only" alt="tensorrt_llm - Home"/>
<p class="title logo__title">tensorrt_llm</p>
</a>
<div class="sidebar-header-items sidebar-primary__section">
<div class="sidebar-header-items__end">
<div class="navbar-item">
<button class="btn btn-sm nav-link pst-navbar-icon theme-switch-button pst-js-only" aria-label="Color mode" data-bs-title="Color mode" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="theme-switch fa-solid fa-sun fa-lg" data-mode="light" title="Light"></i>
<i class="theme-switch fa-solid fa-moon fa-lg" data-mode="dark" title="Dark"></i>
<i class="theme-switch fa-solid fa-circle-half-stroke fa-lg" data-mode="auto" title="System Settings"></i>
</button></div>
</div>
</div>
<div class="sidebar-primary-items__start sidebar-primary__section">
<div class="sidebar-primary-item">
<nav class="bd-docs-nav bd-links"
aria-label="Table of Contents">
<p class="bd-links__title" role="heading" aria-level="1">Table of Contents</p>
<div class="bd-toc-item navbar-nav"><p aria-level="2" class="caption" role="heading"><span class="caption-text">Getting Started</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../../overview.html">Overview</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../quick-start-guide.html">Quick Start Guide</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../key-features.html">Key Features</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../torch.html">PyTorch Backend</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../release-notes.html">Release Notes</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Installation</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../../installation/linux.html">Installing on Linux</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../installation/build-from-source-linux.html">Building from Source Code on Linux</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../installation/grace-hopper.html">Installing on Grace Hopper</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">LLM API</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../../llm-api/index.html">API Introduction</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../llm-api/reference.html">API Reference</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">LLM API Examples</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1 has-children"><a class="reference internal" href="../../llm-api-examples/index.html">LLM Examples Introduction</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_multilora.html">Generate text with multiple LoRA adapters</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_quantization.html">Generation with Quantization</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_inference_async.html">Generate Text Asynchronously</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_guided_decoding.html">Generate text with guided decoding</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_inference_customize.html">Generate text with customization</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_logits_processor.html">Control generated text using logits post processor</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_inference_async_streaming.html">Generate Text in Streaming</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_inference_distributed.html">Distributed LLM Generation</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_medusa_decoding.html">Generate Text Using Medusa Decoding</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_lookahead_decoding.html">Generate Text Using Lookahead Decoding</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_inference.html">Generate text</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_auto_parallel.html">Automatic Parallelism with LLM</a></li>
</ul>
</details></li>
<li class="toctree-l1"><a class="reference internal" href="../../llm-api-examples/customization.html">Common Customizations</a></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="../../llm-api-examples/llm_api_examples.html">Examples</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_multilora.html">Generate text with multiple LoRA adapters</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_quantization.html">Generation with Quantization</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_inference_async.html">Generate Text Asynchronously</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_guided_decoding.html">Generate text with guided decoding</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_inference_customize.html">Generate text with customization</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_logits_processor.html">Control generated text using logits post processor</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_inference_async_streaming.html">Generate Text in Streaming</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_inference_distributed.html">Distributed LLM Generation</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_medusa_decoding.html">Generate Text Using Medusa Decoding</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_lookahead_decoding.html">Generate Text Using Lookahead Decoding</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_inference.html">Generate text</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../llm-api-examples/llm_auto_parallel.html">Automatic Parallelism with LLM</a></li>
</ul>
</details></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Model Definition API</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../../python-api/tensorrt_llm.layers.html">Layers</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../python-api/tensorrt_llm.functional.html">Functionals</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../python-api/tensorrt_llm.models.html">Models</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../python-api/tensorrt_llm.plugin.html">Plugin</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../python-api/tensorrt_llm.quantization.html">Quantization</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../python-api/tensorrt_llm.runtime.html">Runtime</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">C++ API</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../../_cpp_gen/executor.html">Executor</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../_cpp_gen/runtime.html">Runtime</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Command-Line Reference</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../../commands/trtllm-build.html">trtllm-build</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../commands/trtllm-serve.html">trtllm-serve</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Architecture</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../../architecture/overview.html">TensorRT-LLM Architecture</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../architecture/core-concepts.html">Model Definition</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../architecture/checkpoint.html">TensorRT-LLM Checkpoint</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../architecture/workflow.html">TensorRT-LLM Build Workflow</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../architecture/add-model.html">Adding a Model</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Advanced</span></p>
<ul class="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/inference-request.html">Inference Request</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../advanced/lora.html">Run gpt-2b + LoRA using GptManager / 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-reuse.html">KV cache reuse</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../advanced/speculative-decoding.html">Speculative Sampling</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../advanced/disaggregated-service.html">Disaggregated-Service (experimental)</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Performance</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../../performance/perf-overview.html">Overview</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../performance/perf-benchmarking.html">Benchmarking</a></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="../../performance/performance-tuning-guide/index.html">Performance Tuning Guide</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="../../performance/performance-tuning-guide/benchmarking-default-performance.html">Benchmarking Default Performance</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../performance/performance-tuning-guide/useful-build-time-flags.html">Useful Build-Time Flags</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../performance/performance-tuning-guide/tuning-max-batch-size-and-max-num-tokens.html">Tuning Max Batch Size and Max Num Tokens</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../performance/performance-tuning-guide/deciding-model-sharding-strategy.html">Deciding Model Sharding Strategy</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../performance/performance-tuning-guide/fp8-quantization.html">FP8 Quantization</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../performance/performance-tuning-guide/useful-runtime-flags.html">Useful Runtime Options</a></li>
</ul>
</details></li>
<li class="toctree-l1"><a class="reference internal" href="../../performance/perf-analysis.html">Performance Analysis</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Reference</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../../reference/troubleshooting.html">Troubleshooting</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../reference/support-matrix.html">Support Matrix</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../reference/precision.html">Numerical Precision</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../reference/memory.html">Memory Usage of TensorRT-LLM</a></li>
</ul>
<p aria-level="2" class="caption" role="heading"><span class="caption-text">Blogs</span></p>
<ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="../../blogs/H100vsA100.html">H100 has 4.6x A100 Performance in TensorRT-LLM, achieving 10,000 tok/s at 100ms to first token</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../blogs/H200launch.html">H200 achieves nearly 12,000 tokens/sec on Llama2-13B with TensorRT-LLM</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../blogs/Falcon180B-H200.html">Falcon-180B on a single H200 GPU with INT4 AWQ, and 6.7x faster Llama-70B over A100</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../blogs/quantization-in-TRT-LLM.html">Speed up inference with SOTA quantization techniques in TRT-LLM</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../blogs/XQA-kernel.html">New XQA-kernel provides 2.4x more Llama-70B throughput within the same latency budget</a></li>
</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.executor</span></li>
</ul>
</nav>
</div>
</div>
</div>
</div>
<div id="searchbox"></div>
<article class="bd-article">
<h1>Source code for tensorrt_llm.executor</h1><div class="highlight"><pre>
<span></span><span class="kn">import</span> <span class="nn">asyncio</span>
<span class="kn">import</span> <span class="nn">atexit</span>
<span class="kn">import</span> <span class="nn">concurrent.futures</span>
<span class="kn">import</span> <span class="nn">copy</span>
<span class="kn">import</span> <span class="nn">datetime</span>
<span class="kn">import</span> <span class="nn">faulthandler</span>
<span class="kn">import</span> <span class="nn">io</span>
<span class="kn">import</span> <span class="nn">json</span>
<span class="kn">import</span> <span class="nn">multiprocessing</span>
<span class="kn">import</span> <span class="nn">os</span>
<span class="kn">import</span> <span class="nn">pickle</span> <span class="c1"># nosec B403</span>
<span class="kn">import</span> <span class="nn">platform</span>
<span class="kn">import</span> <span class="nn">signal</span>
<span class="kn">import</span> <span class="nn">time</span>
<span class="kn">import</span> <span class="nn">traceback</span>
<span class="kn">import</span> <span class="nn">weakref</span>
<span class="kn">from</span> <span class="nn">abc</span> <span class="kn">import</span> <span class="n">ABC</span><span class="p">,</span> <span class="n">abstractmethod</span>
<span class="kn">from</span> <span class="nn">concurrent.futures</span> <span class="kn">import</span> <span class="n">ProcessPoolExecutor</span>
<span class="kn">from</span> <span class="nn">dataclasses</span> <span class="kn">import</span> <span class="n">dataclass</span><span class="p">,</span> <span class="n">field</span>
<span class="kn">from</span> <span class="nn">multiprocessing.shared_memory</span> <span class="kn">import</span> <span class="n">SharedMemory</span>
<span class="kn">from</span> <span class="nn">pathlib</span> <span class="kn">import</span> <span class="n">Path</span>
<span class="kn">from</span> <span class="nn">queue</span> <span class="kn">import</span> <span class="n">Empty</span><span class="p">,</span> <span class="n">Queue</span>
<span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="p">(</span><span class="n">Any</span><span class="p">,</span> <span class="n">Callable</span><span class="p">,</span> <span class="n">Dict</span><span class="p">,</span> <span class="n">Generator</span><span class="p">,</span> <span class="n">List</span><span class="p">,</span> <span class="n">Literal</span><span class="p">,</span> <span class="n">NamedTuple</span><span class="p">,</span>
<span class="n">Optional</span><span class="p">,</span> <span class="n">Tuple</span><span class="p">,</span> <span class="n">Union</span><span class="p">)</span>
<span class="kn">from</span> <span class="nn">weakref</span> <span class="kn">import</span> <span class="n">WeakMethod</span>
<span class="kn">import</span> <span class="nn">numpy</span> <span class="k">as</span> <span class="nn">np</span>
<span class="kn">import</span> <span class="nn">torch</span>
<span class="kn">import</span> <span class="nn">zmq</span>
<span class="kn">from</span> <span class="nn">tensorrt_llm.logger</span> <span class="kn">import</span> <span class="n">logger</span><span class="p">,</span> <span class="n">set_level</span>
<span class="kn">from</span> <span class="nn">._utils</span> <span class="kn">import</span> <span class="n">mpi_rank</span><span class="p">,</span> <span class="n">mpi_world_size</span>
<span class="kn">from</span> <span class="nn">.bindings</span> <span class="kn">import</span> <span class="n">executor</span> <span class="k">as</span> <span class="n">tllm</span>
<span class="kn">from</span> <span class="nn">.builder</span> <span class="kn">import</span> <span class="n">ConfigEncoder</span><span class="p">,</span> <span class="n">Engine</span><span class="p">,</span> <span class="n">EngineConfig</span>
<span class="kn">from</span> <span class="nn">.llmapi.mpi_session</span> <span class="kn">import</span> <span class="p">(</span><span class="n">MpiPoolSession</span><span class="p">,</span> <span class="n">MpiSession</span><span class="p">,</span>
<span class="n">external_mpi_comm_available</span><span class="p">,</span>
<span class="n">need_spawn_mpi_workers</span><span class="p">)</span>
<span class="kn">from</span> <span class="nn">.llmapi.tracer</span> <span class="kn">import</span> <span class="p">(</span><span class="n">VizTracer</span><span class="p">,</span> <span class="n">enable_llm_tracer</span><span class="p">,</span> <span class="n">get_tracer</span><span class="p">,</span>
<span class="n">global_tracer</span><span class="p">,</span> <span class="n">set_global_tracer</span><span class="p">)</span>
<span class="kn">from</span> <span class="nn">.llmapi.utils</span> <span class="kn">import</span> <span class="p">(</span><span class="n">AsyncQueue</span><span class="p">,</span> <span class="n">ManagedThread</span><span class="p">,</span> <span class="n">_SyncQueue</span><span class="p">,</span>
<span class="n">enable_llm_debug</span><span class="p">,</span>
<span class="n">enable_worker_single_process_for_tp1</span><span class="p">,</span> <span class="n">print_colored</span><span class="p">)</span>
<span class="kn">from</span> <span class="nn">.lora_manager</span> <span class="kn">import</span> <span class="n">LoraManager</span>
<span class="kn">from</span> <span class="nn">.prompt_adapter_manager</span> <span class="kn">import</span> <span class="n">PromptAdapterManager</span>
<span class="kn">from</span> <span class="nn">.runtime</span> <span class="kn">import</span> <span class="n">ModelConfig</span>
<span class="kn">from</span> <span class="nn">.runtime.model_runner</span> <span class="kn">import</span> <span class="n">_engine_config_to_model_config</span>
<span class="kn">from</span> <span class="nn">.sampling_params</span> <span class="kn">import</span> <span class="n">SamplingParams</span>
<span class="n">unblock_corountine</span> <span class="o">=</span> <span class="kc">True</span>
<span class="n">BATCH_RESP_IN_AWAIT</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">getenv</span><span class="p">(</span><span class="s2">&quot;TLLM_EXECUTOR_BATCH_RESP_IN_AWAIT&quot;</span><span class="p">)</span> <span class="o">==</span> <span class="s2">&quot;1&quot;</span>
<span class="k">if</span> <span class="n">enable_llm_debug</span><span class="p">():</span>
<span class="c1"># Mainly enable more detailed logging from cpp runtime.</span>
<span class="n">set_level</span><span class="p">(</span><span class="s2">&quot;info&quot;</span><span class="p">)</span>
<span class="nd">@dataclass</span><span class="p">(</span><span class="n">slots</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="k">class</span> <span class="nc">LoRARequest</span><span class="p">:</span>
<span class="w"> </span><span class="sd">&quot;&quot;&quot; Request for a LoRA adapter. &quot;&quot;&quot;</span>
<span class="n">lora_name</span><span class="p">:</span> <span class="nb">str</span>
<span class="n">lora_int_id</span><span class="p">:</span> <span class="nb">int</span>
<span class="n">lora_path</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="s2">&quot;&quot;</span>
<span class="k">def</span> <span class="nf">__post_init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">if</span> <span class="ow">not</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">exists</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">lora_path</span><span class="p">):</span>
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;lora_path (</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">lora_path</span><span class="si">}</span><span class="s2">) does not exist.&quot;</span><span class="p">)</span>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">adapter_id</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_int_id</span>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">name</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_name</span>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">path</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_path</span>
<span class="nd">@dataclass</span><span class="p">(</span><span class="n">slots</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="k">class</span> <span class="nc">PromptAdapterRequest</span><span class="p">:</span>
<span class="w"> </span><span class="sd">&quot;&quot;&quot;</span>
<span class="sd"> Request for a Prompt adapter.</span>
<span class="sd"> &quot;&quot;&quot;</span>
<span class="n">prompt_adapter_name</span><span class="p">:</span> <span class="nb">str</span>
<span class="n">prompt_adapter_id</span><span class="p">:</span> <span class="nb">int</span>
<span class="n">prompt_adapter_local_path</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="s2">&quot;&quot;</span>
<span class="k">def</span> <span class="nf">__post_init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">if</span> <span class="ow">not</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">exists</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">prompt_adapter_local_path</span><span class="p">):</span>
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span>
<span class="sa">f</span><span class="s2">&quot;prompt_adapter_local_path (</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">prompt_adapter_local_path</span><span class="si">}</span><span class="s2">) does not exist.&quot;</span>
<span class="p">)</span>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">adapter_id</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">prompt_adapter_id</span>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">name</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">prompt_adapter_name</span>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">local_path</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">prompt_adapter_local_path</span>
<span class="k">class</span> <span class="nc">GenerationRequest</span><span class="p">:</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span>
<span class="bp">self</span><span class="p">,</span>
<span class="n">prompt_token_ids</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">np</span><span class="o">.</span><span class="n">ndarray</span><span class="p">,</span>
<span class="n">Union</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">],</span> <span class="n">List</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]]]],</span>
<span class="n">sampling_params</span><span class="p">:</span> <span class="n">SamplingParams</span><span class="p">,</span>
<span class="n">query_token_ids</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">np</span><span class="o">.</span><span class="n">ndarray</span><span class="p">,</span> <span class="nb">list</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">lora_request</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">LoRARequest</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">prompt_adapter_request</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">PromptAdapterRequest</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">streaming</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
<span class="n">prompt_tuning_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">list</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="p">):</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">prompt_token_ids</span><span class="p">,</span> <span class="nb">list</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">prompt_token_ids</span> <span class="o">=</span> <span class="n">prompt_token_ids</span>
<span class="bp">self</span><span class="o">.</span><span class="n">query_token_ids</span> <span class="o">=</span> <span class="n">query_token_ids</span>
<span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">prompt_token_ids</span><span class="p">,</span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">np</span><span class="o">.</span><span class="n">ndarray</span><span class="p">)):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">prompt_token_ids</span> <span class="o">=</span> <span class="n">prompt_token_ids</span><span class="o">.</span><span class="n">tolist</span><span class="p">()</span>
<span class="k">if</span> <span class="n">query_token_ids</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">query_token_ids</span> <span class="o">=</span> <span class="n">query_token_ids</span><span class="o">.</span><span class="n">tolist</span><span class="p">()</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">raise</span> <span class="ne">TypeError</span><span class="p">(</span>
<span class="sa">f</span><span class="s2">&quot;prompt_token_ids (</span><span class="si">{</span><span class="n">prompt_token_ids</span><span class="si">}</span><span class="s2">) should be an instance of torch.Tensor, np.ndarray or list&quot;</span>
<span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">sampling_params</span> <span class="o">=</span> <span class="n">sampling_params</span>
<span class="bp">self</span><span class="o">.</span><span class="n">lora_request</span> <span class="o">=</span> <span class="n">lora_request</span>
<span class="bp">self</span><span class="o">.</span><span class="n">prompt_adapter_request</span> <span class="o">=</span> <span class="n">prompt_adapter_request</span>
<span class="bp">self</span><span class="o">.</span><span class="n">streaming</span> <span class="o">=</span> <span class="n">streaming</span>
<span class="bp">self</span><span class="o">.</span><span class="n">prompt_tuning_config</span> <span class="o">=</span> <span class="n">prompt_tuning_config</span>
<span class="bp">self</span><span class="o">.</span><span class="n">id</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">def</span> <span class="nf">set_id</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="nb">id</span><span class="p">):</span>
<span class="k">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">id</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">,</span> <span class="sa">f</span><span class="s2">&quot;Request ID is already set: </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">id</span><span class="si">}</span><span class="s2">&quot;</span>
<span class="bp">self</span><span class="o">.</span><span class="n">id</span> <span class="o">=</span> <span class="nb">id</span>
<span class="k">return</span> <span class="bp">self</span>
<span class="nd">@dataclass</span><span class="p">(</span><span class="n">slots</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="k">class</span> <span class="nc">CompletionOutput</span><span class="p">:</span>
<span class="w"> </span><span class="sd">&quot;&quot;&quot;The output data of one completion output of a request.</span>
<span class="sd"> Args:</span>
<span class="sd"> index (int): The index of the output in the request.</span>
<span class="sd"> text (str): The generated output text. Defaults to &quot;&quot;.</span>
<span class="sd"> token_ids (List[int]): The token ids of the generated output text. Defaults to [].</span>
<span class="sd"> cumulative_logprob (float, optional): The cumulative log probability of the generated output text. Defaults to None.</span>
<span class="sd"> logprobs (List[float]): The log probabilities of the top probability words at each position if the logprobs are requested. Defaults to [].</span>
<span class="sd"> finish_reason (Literal[&#39;stop&#39;, &#39;length&#39;], optional): The reason why the sequence is finished. Defaults to None.</span>
<span class="sd"> stop_reason (int, str, optional): The stop string or token id that caused the completion to stop, None if the completion finished for some other reason. Defaults to None.</span>
<span class="sd"> generation_logits (torch.Tensor, optional): The logits on the generated output token ids. Defaults to None.</span>
<span class="sd"> Properties:</span>
<span class="sd"> length (int): The number of generated tokens.</span>
<span class="sd"> token_ids_diff (List[int]): Newly generated token ids.</span>
<span class="sd"> logprobs_diff (List[float]): Logprobs of newly generated tokens.</span>
<span class="sd"> text_diff (str): Newly generated tokens.</span>
<span class="sd"> &quot;&quot;&quot;</span>
<span class="n">index</span><span class="p">:</span> <span class="nb">int</span>
<span class="n">text</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="s2">&quot;&quot;</span>
<span class="n">token_ids</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span><span class="n">default_factory</span><span class="o">=</span><span class="nb">list</span><span class="p">)</span>
<span class="n">cumulative_logprob</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">float</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">logprobs</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="nb">float</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span><span class="n">default_factory</span><span class="o">=</span><span class="nb">list</span><span class="p">)</span>
<span class="n">finish_reason</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Literal</span><span class="p">[</span><span class="s1">&#39;stop&#39;</span><span class="p">,</span> <span class="s1">&#39;length&#39;</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">stop_reason</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span><span class="nb">int</span><span class="p">,</span> <span class="nb">str</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">generation_logits</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
<span class="c1"># hidden fields for tracking the diffs</span>
<span class="n">_last_text_len</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span> <span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="nb">repr</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
<span class="n">_last_token_ids_len</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span> <span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="nb">repr</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
<span class="n">_last_logprobs_len</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span> <span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="nb">repr</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
<span class="n">_incremental_states</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">dict</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="nb">repr</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">length</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="nb">len</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">token_ids</span><span class="p">)</span>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">text_diff</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">str</span><span class="p">:</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">text</span><span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">_last_text_len</span><span class="p">:]</span>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">token_ids_diff</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]:</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">token_ids</span><span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">_last_token_ids_len</span><span class="p">:]</span>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">logprobs_diff</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="n">List</span><span class="p">[</span><span class="nb">float</span><span class="p">]:</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">logprobs</span><span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">_last_logprobs_len</span><span class="p">:]</span>
<span class="k">class</span> <span class="nc">CppExecutorError</span><span class="p">(</span><span class="ne">RuntimeError</span><span class="p">):</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">message</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">message</span> <span class="o">=</span> <span class="n">message</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stack_trace</span> <span class="o">=</span> <span class="n">traceback</span><span class="o">.</span><span class="n">format_exc</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">message</span><span class="p">)</span>
<span class="k">def</span> <span class="fm">__str__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="sa">f</span><span class="s2">&quot;</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">message</span><span class="si">}</span><span class="se">\n</span><span class="s2">Stack trace:</span><span class="se">\n</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">stack_trace</span><span class="si">}</span><span class="s2">&quot;</span>
<div class="viewcode-block" id="RequestError">
<a class="viewcode-back" href="../../llm-api/reference.html#tensorrt_llm.llmapi.RequestError">[docs]</a>
<span class="k">class</span> <span class="nc">RequestError</span><span class="p">(</span><span class="ne">RuntimeError</span><span class="p">):</span>
<span class="w"> </span><span class="sd">&#39;&#39;&#39; The error raised when the request is failed. &#39;&#39;&#39;</span></div>
<span class="k">class</span> <span class="nc">GenerationResult</span><span class="p">:</span>
<span class="w"> </span><span class="sd">&#39;&#39;&#39;</span>
<span class="sd"> The result of a generation request. It can be used to wait for the completion of the request.</span>
<span class="sd"> Args:</span>
<span class="sd"> generation_request (GenerationRequest): The generation request object.</span>
<span class="sd"> background_error_handler (Callable, optional): The error handler to process the errors from the background threads/processes. Defaults to None.</span>
<span class="sd"> &#39;&#39;&#39;</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span>
<span class="n">generation_request</span><span class="p">:</span> <span class="n">GenerationRequest</span><span class="p">,</span>
<span class="n">background_error_handler</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Callable</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="kc">None</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_done</span> <span class="o">=</span> <span class="kc">False</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_cancelled</span> <span class="o">=</span> <span class="kc">False</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_generation_request</span> <span class="o">=</span> <span class="n">generation_request</span>
<span class="k">if</span> <span class="n">has_event_loop</span><span class="p">():</span>
<span class="bp">self</span><span class="o">.</span><span class="n">aqueue</span> <span class="o">=</span> <span class="n">AsyncQueue</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">queue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">aqueue</span><span class="o">.</span><span class="n">sync_q</span>
<span class="k">else</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">queue</span> <span class="o">=</span> <span class="n">Queue</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">aqueue</span> <span class="o">=</span> <span class="kc">None</span>
<span class="c1"># In Sampling mode, the Executor runtime will return best_of sequences</span>
<span class="c1"># in total, which the LLM API will select the n-best sequences among</span>
<span class="c1"># them based on their cumulative log probabilities.</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_outputs</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="n">CompletionOutput</span><span class="p">]</span> <span class="o">=</span> <span class="p">[</span>
<span class="n">CompletionOutput</span><span class="p">(</span><span class="n">i</span><span class="p">)</span>
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_generation_request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">best_of</span><span class="p">)</span>
<span class="p">]</span>
<span class="bp">self</span><span class="o">.</span><span class="n">context_logits</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_background_error_handler</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="n">background_error_handler</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_background_error_handler</span> <span class="o">=</span> <span class="n">WeakMethod</span><span class="p">(</span>
<span class="n">background_error_handler</span><span class="p">)</span>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">request_id</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">int</span><span class="p">:</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_generation_request</span><span class="o">.</span><span class="n">id</span>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">prompt_token_ids</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]:</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_generation_request</span><span class="o">.</span><span class="n">prompt_token_ids</span>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">finished</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">bool</span><span class="p">:</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_done</span>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">streaming</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_generation_request</span><span class="o">.</span><span class="n">streaming</span>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">outputs</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="n">List</span><span class="p">[</span><span class="n">CompletionOutput</span><span class="p">]:</span>
<span class="n">sampling_param</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_generation_request</span><span class="o">.</span><span class="n">sampling_params</span>
<span class="k">if</span> <span class="p">(</span><span class="n">sampling_param</span><span class="o">.</span><span class="n">use_beam_search</span>
<span class="ow">or</span> <span class="n">sampling_param</span><span class="o">.</span><span class="n">n</span> <span class="o">==</span> <span class="n">sampling_param</span><span class="o">.</span><span class="n">best_of</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_outputs</span><span class="p">[:</span><span class="n">sampling_param</span><span class="o">.</span><span class="n">n</span><span class="p">]</span>
<span class="c1"># Pick the top-n outputs, sorted by cumulative log probs.</span>
<span class="n">sorted_outputs</span> <span class="o">=</span> <span class="nb">sorted</span><span class="p">(</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_outputs</span><span class="p">,</span>
<span class="n">key</span><span class="o">=</span><span class="k">lambda</span> <span class="n">x</span><span class="p">:</span>
<span class="p">(</span><span class="n">x</span><span class="o">.</span><span class="n">cumulative_logprob</span>
<span class="k">if</span> <span class="n">x</span><span class="o">.</span><span class="n">cumulative_logprob</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="k">else</span> <span class="nb">float</span><span class="p">(</span><span class="s1">&#39;-inf&#39;</span><span class="p">)),</span>
<span class="n">reverse</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="c1"># Reindex the sequence.</span>
<span class="k">for</span> <span class="n">i</span><span class="p">,</span> <span class="n">sorted_out</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">sorted_outputs</span><span class="p">):</span>
<span class="n">sorted_out</span><span class="o">.</span><span class="n">index</span> <span class="o">=</span> <span class="n">i</span>
<span class="k">return</span> <span class="n">sorted_outputs</span><span class="p">[:</span><span class="n">sampling_param</span><span class="o">.</span><span class="n">n</span><span class="p">]</span>
<span class="k">def</span> <span class="nf">handle_sequence</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">response</span><span class="p">:</span> <span class="s2">&quot;GenerationExecutor.Response&quot;</span><span class="p">,</span>
<span class="n">sequence_index</span><span class="p">:</span> <span class="nb">int</span><span class="p">):</span>
<span class="w"> </span><span class="sd">&quot;&quot;&quot; Handle a single sequence in the response. &quot;&quot;&quot;</span>
<span class="n">tensors</span> <span class="o">=</span> <span class="n">response</span><span class="o">.</span><span class="n">tensors</span>
<span class="k">assert</span> <span class="n">tensors</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
<span class="n">beam_search</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_generation_request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">use_beam_search</span>
<span class="n">seq_idx</span> <span class="o">=</span> <span class="n">sequence_index</span>
<span class="n">src_idx</span> <span class="o">=</span> <span class="n">sequence_index</span> <span class="k">if</span> <span class="n">beam_search</span> <span class="k">else</span> <span class="mi">0</span>
<span class="n">output</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_outputs</span><span class="p">[</span><span class="n">seq_idx</span><span class="p">]</span>
<span class="n">output</span><span class="o">.</span><span class="n">_last_token_ids_len</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="n">output</span><span class="o">.</span><span class="n">token_ids</span><span class="p">)</span>
<span class="n">output</span><span class="o">.</span><span class="n">token_ids</span><span class="o">.</span><span class="n">extend</span><span class="p">(</span><span class="n">tensors</span><span class="o">.</span><span class="n">output_token_ids</span><span class="p">[</span><span class="n">src_idx</span><span class="p">])</span>
<span class="k">if</span> <span class="n">tensors</span><span class="o">.</span><span class="n">cum_log_probs</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">output</span><span class="o">.</span><span class="n">cumulative_logprob</span> <span class="o">=</span> <span class="n">tensors</span><span class="o">.</span><span class="n">cum_log_probs</span><span class="p">[</span><span class="n">src_idx</span><span class="p">]</span>
<span class="k">if</span> <span class="n">tensors</span><span class="o">.</span><span class="n">log_probs</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">output</span><span class="o">.</span><span class="n">_last_logprobs_len</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="n">output</span><span class="o">.</span><span class="n">logprobs</span><span class="p">)</span>
<span class="n">output</span><span class="o">.</span><span class="n">logprobs</span> <span class="o">=</span> <span class="n">tensors</span><span class="o">.</span><span class="n">log_probs</span><span class="p">[</span><span class="n">src_idx</span><span class="p">]</span>
<span class="k">assert</span> <span class="nb">len</span><span class="p">(</span><span class="n">output</span><span class="o">.</span><span class="n">logprobs</span><span class="p">)</span> <span class="o">==</span> <span class="n">output</span><span class="o">.</span><span class="n">length</span>
<span class="k">if</span> <span class="n">tensors</span><span class="o">.</span><span class="n">generation_logits</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">output</span><span class="o">.</span><span class="n">generation_logits</span> <span class="o">=</span> <span class="n">tensors</span><span class="o">.</span><span class="n">generation_logits</span><span class="p">[</span>
<span class="n">src_idx</span><span class="p">,</span> <span class="p">:</span><span class="n">output</span><span class="o">.</span><span class="n">length</span><span class="p">]</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">finished</span><span class="p">:</span>
<span class="k">if</span> <span class="n">response</span><span class="o">.</span><span class="n">finish_reasons</span><span class="p">[</span><span class="n">src_idx</span><span class="p">]</span> <span class="o">==</span> <span class="n">tllm</span><span class="o">.</span><span class="n">FinishReason</span><span class="o">.</span><span class="n">END_ID</span><span class="p">:</span>
<span class="n">output</span><span class="o">.</span><span class="n">finish_reason</span> <span class="o">=</span> <span class="s1">&#39;stop&#39;</span>
<span class="k">elif</span> <span class="n">response</span><span class="o">.</span><span class="n">finish_reasons</span><span class="p">[</span>
<span class="n">src_idx</span><span class="p">]</span> <span class="o">==</span> <span class="n">tllm</span><span class="o">.</span><span class="n">FinishReason</span><span class="o">.</span><span class="n">STOP_WORDS</span><span class="p">:</span>
<span class="n">output</span><span class="o">.</span><span class="n">finish_reason</span> <span class="o">=</span> <span class="s1">&#39;stop&#39;</span>
<span class="n">sampling_params</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_generation_request</span><span class="o">.</span><span class="n">sampling_params</span>
<span class="k">for</span> <span class="n">stop_reason</span><span class="p">,</span> <span class="n">stop_ids</span> <span class="ow">in</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">_get_stop_reasons_and_words</span><span class="p">(</span>
<span class="p">):</span>
<span class="k">if</span> <span class="n">output</span><span class="o">.</span><span class="n">token_ids</span><span class="p">[</span><span class="o">-</span><span class="nb">len</span><span class="p">(</span><span class="n">stop_ids</span><span class="p">):]</span> <span class="o">==</span> <span class="n">stop_ids</span><span class="p">:</span>
<span class="n">output</span><span class="o">.</span><span class="n">stop_reason</span> <span class="o">=</span> <span class="n">stop_reason</span>
<span class="k">if</span> <span class="ow">not</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">include_stop_str_in_output</span><span class="p">:</span>
<span class="n">output</span><span class="o">.</span><span class="n">token_ids</span> <span class="o">=</span> <span class="n">output</span><span class="o">.</span><span class="n">token_ids</span><span class="p">[:</span><span class="o">-</span><span class="nb">len</span><span class="p">(</span><span class="n">stop_ids</span><span class="p">)]</span>
<span class="k">break</span>
<span class="k">elif</span> <span class="n">response</span><span class="o">.</span><span class="n">finish_reasons</span><span class="p">[</span><span class="n">src_idx</span><span class="p">]</span> <span class="o">==</span> <span class="n">tllm</span><span class="o">.</span><span class="n">FinishReason</span><span class="o">.</span><span class="n">LENGTH</span><span class="p">:</span>
<span class="n">output</span><span class="o">.</span><span class="n">finish_reason</span> <span class="o">=</span> <span class="s1">&#39;length&#39;</span>
<span class="k">def</span> <span class="nf">handle_response</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">response</span><span class="p">:</span> <span class="s2">&quot;GenerationExecutor.Response&quot;</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_done</span> <span class="o">=</span> <span class="n">response</span><span class="o">.</span><span class="n">is_final</span>
<span class="k">if</span> <span class="n">response</span><span class="o">.</span><span class="n">error</span><span class="p">:</span>
<span class="k">if</span> <span class="n">handler</span> <span class="o">:=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_background_error_handler</span><span class="p">():</span>
<span class="n">handler</span><span class="p">(</span><span class="n">response</span><span class="o">.</span><span class="n">error</span><span class="p">)</span>
<span class="n">tensors</span> <span class="o">=</span> <span class="n">response</span><span class="o">.</span><span class="n">tensors</span>
<span class="c1"># output_token_ids = (beams, tokens)</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_generation_request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">use_beam_search</span><span class="p">:</span>
<span class="k">for</span> <span class="n">beam_idx</span><span class="p">,</span> <span class="n">_</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">tensors</span><span class="o">.</span><span class="n">output_token_ids</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">handle_sequence</span><span class="p">(</span><span class="n">response</span><span class="p">,</span> <span class="n">beam_idx</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">handle_sequence</span><span class="p">(</span><span class="n">response</span><span class="p">,</span> <span class="n">response</span><span class="o">.</span><span class="n">sequence_index</span><span class="p">)</span>
<span class="k">if</span> <span class="n">tensors</span><span class="o">.</span><span class="n">context_logits</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">context_logits</span> <span class="o">=</span> <span class="n">tensors</span><span class="o">.</span><span class="n">context_logits</span>
<span class="c1"># Processing background errors here ASAF during generation.</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_background_error_handler</span> <span class="ow">and</span> <span class="p">(</span>
<span class="n">handler</span> <span class="o">:=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_background_error_handler</span><span class="p">()):</span>
<span class="n">handler</span><span class="p">()</span>
<span class="k">def</span> <span class="nf">result_step</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">timeout</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">float</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span>
<span class="n">response</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">queue</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="n">timeout</span><span class="o">=</span><span class="n">timeout</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">handle_response</span><span class="p">(</span><span class="n">response</span><span class="p">)</span>
<span class="k">async</span> <span class="k">def</span> <span class="nf">aresult_step</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">aqueue</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">,</span> <span class="s2">&quot;The asyncio event loop was not present during initialization, so async operations are not available.&quot;</span>
<span class="n">response</span> <span class="o">=</span> <span class="k">await</span> <span class="bp">self</span><span class="o">.</span><span class="n">aqueue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
<span class="n">global_tracer</span><span class="p">()</span><span class="o">.</span><span class="n">log_instant</span><span class="p">(</span><span class="s2">&quot;result_step.get&quot;</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">handle_response</span><span class="p">(</span><span class="n">response</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">result</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">timeout</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">float</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="s2">&quot;GenerationResult&quot;</span><span class="p">:</span>
<span class="k">while</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">_done</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">result_step</span><span class="p">(</span><span class="n">timeout</span><span class="p">)</span>
<span class="k">return</span> <span class="bp">self</span>
<span class="k">async</span> <span class="k">def</span> <span class="nf">aresult</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="s2">&quot;GenerationResult&quot;</span><span class="p">:</span>
<span class="k">while</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">_done</span><span class="p">:</span>
<span class="k">await</span> <span class="bp">self</span><span class="o">.</span><span class="n">aresult_step</span><span class="p">()</span>
<span class="k">return</span> <span class="bp">self</span>
<span class="k">def</span> <span class="fm">__await__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">aresult</span><span class="p">()</span><span class="o">.</span><span class="fm">__await__</span><span class="p">()</span>
<span class="k">def</span> <span class="fm">__iter__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span>
<span class="k">def</span> <span class="fm">__next__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_done</span><span class="p">:</span>
<span class="k">raise</span> <span class="ne">StopIteration</span>
<span class="bp">self</span><span class="o">.</span><span class="n">result_step</span><span class="p">()</span>
<span class="k">return</span> <span class="bp">self</span>
<span class="k">def</span> <span class="fm">__aiter__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span>
<span class="k">async</span> <span class="k">def</span> <span class="fm">__anext__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_done</span><span class="p">:</span>
<span class="k">raise</span> <span class="ne">StopAsyncIteration</span>
<span class="k">await</span> <span class="bp">self</span><span class="o">.</span><span class="n">aresult_step</span><span class="p">()</span>
<span class="k">return</span> <span class="bp">self</span>
<span class="k">def</span> <span class="nf">running</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">bool</span><span class="p">:</span>
<span class="k">return</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">_done</span>
<span class="k">def</span> <span class="nf">cancelled</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">bool</span><span class="p">:</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_cancelled</span>
<span class="k">def</span> <span class="nf">cancel</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">raise</span> <span class="ne">NotImplementedError</span>
<span class="k">def</span> <span class="nf">done</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">bool</span><span class="p">:</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_done</span>
<span class="k">def</span> <span class="nf">exception</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">timeout</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">float</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span>
<span class="k">try</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">result</span><span class="p">(</span><span class="n">timeout</span><span class="p">)</span>
<span class="k">except</span> <span class="ne">RuntimeError</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
<span class="k">return</span> <span class="n">e</span>
<span class="k">def</span> <span class="nf">_repr_fields</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="p">[</span>
<span class="s1">&#39;request_id&#39;</span><span class="p">,</span> <span class="s1">&#39;prompt_token_ids&#39;</span><span class="p">,</span> <span class="s1">&#39;outputs&#39;</span><span class="p">,</span> <span class="s1">&#39;finished&#39;</span><span class="p">,</span>
<span class="s2">&quot;context_logits&quot;</span>
<span class="p">]</span>
<span class="k">def</span> <span class="fm">__repr__</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">str</span><span class="p">:</span>
<span class="nb">repr</span> <span class="o">=</span> <span class="p">[]</span>
<span class="k">for</span> <span class="n">field</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">_repr_fields</span><span class="p">():</span>
<span class="n">value</span> <span class="o">=</span> <span class="nb">getattr</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">field</span><span class="p">)</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">value</span><span class="p">,</span> <span class="nb">str</span><span class="p">):</span>
<span class="nb">repr</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;</span><span class="si">{</span><span class="n">field</span><span class="si">}</span><span class="s2">=</span><span class="si">{</span><span class="n">value</span><span class="si">!r}</span><span class="s2">&quot;</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="nb">repr</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;</span><span class="si">{</span><span class="n">field</span><span class="si">}</span><span class="s2">=</span><span class="si">{</span><span class="n">value</span><span class="si">}</span><span class="s2">&quot;</span><span class="p">)</span>
<span class="nb">repr</span> <span class="o">=</span> <span class="s2">&quot;, &quot;</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="nb">repr</span><span class="p">)</span>
<span class="nb">repr</span> <span class="o">=</span> <span class="sa">f</span><span class="s2">&quot;</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="vm">__class__</span><span class="o">.</span><span class="vm">__name__</span><span class="si">}</span><span class="s2">(</span><span class="si">{</span><span class="nb">repr</span><span class="si">}</span><span class="s2">)&quot;</span>
<span class="k">return</span> <span class="nb">repr</span>
<span class="k">def</span> <span class="fm">__hash__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="nb">hash</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">request_id</span><span class="p">)</span>
<div class="viewcode-block" id="NoStatsAvailable">
<a class="viewcode-back" href="../../llm-api/reference.html#tensorrt_llm.llmapi.NoStatsAvailable">[docs]</a>
<span class="k">class</span> <span class="nc">NoStatsAvailable</span><span class="p">(</span><span class="ne">Exception</span><span class="p">):</span>
<span class="k">pass</span></div>
<span class="k">class</span> <span class="nc">GenerationExecutor</span><span class="p">(</span><span class="n">ABC</span><span class="p">):</span>
<span class="k">class</span> <span class="nc">ResponseTensors</span><span class="p">(</span><span class="n">NamedTuple</span><span class="p">):</span>
<span class="n">output_token_ids</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]]</span>
<span class="c1"># context_logits is a tensor or a string denoting the path to the shared memory.</span>
<span class="n">context_logits</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span> <span class="o">|</span> <span class="nb">str</span><span class="p">]</span>
<span class="c1"># generation_logits is a tensor or a string denoting the path to the shared memory.</span>
<span class="n">generation_logits</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span> <span class="o">|</span> <span class="nb">str</span><span class="p">]</span>
<span class="n">log_probs</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">list</span><span class="p">]</span>
<span class="n">cum_log_probs</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">list</span><span class="p">]</span>
<span class="k">class</span> <span class="nc">Response</span><span class="p">(</span><span class="n">NamedTuple</span><span class="p">):</span>
<span class="w"> </span><span class="sd">&quot;&quot;&quot; The response from the cpp-executor to the Python main thread. &quot;&quot;&quot;</span>
<span class="n">client_id</span><span class="p">:</span> <span class="nb">int</span>
<span class="n">tensors</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="s2">&quot;GenerationExecutor.ResponseTensors&quot;</span><span class="p">]</span>
<span class="n">finish_reasons</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="n">tllm</span><span class="o">.</span><span class="n">FinishReason</span><span class="p">]]</span>
<span class="n">is_final</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">bool</span><span class="p">]</span>
<span class="n">sequence_index</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span>
<span class="c1"># There are two types of errors:</span>
<span class="c1"># 1. str for the errors from the cpp-executor.await_responses, this will be dispatched to the user&#39;s</span>
<span class="c1"># generate_async as a per-request error, and won&#39;t stop the whole service.</span>
<span class="c1"># 2. Exception for the errors from the background threads/processes, this will be processed in the main thread,</span>
<span class="c1"># and stop the whole service.</span>
<span class="n">error</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span> <span class="o">|</span> <span class="ne">Exception</span><span class="p">]</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_stats</span> <span class="o">=</span> <span class="kc">None</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">atexit</span><span class="o">.</span><span class="n">register</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">shutdown</span><span class="p">)</span>
<span class="c1"># This is used to capture the exceptions from the threads.</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span> <span class="o">=</span> <span class="n">Queue</span><span class="p">()</span>
<span class="c1"># A flag to avoid calling shutdown() recursively. This happens when the background threads raise errors.</span>
<span class="bp">self</span><span class="o">.</span><span class="n">doing_shutdown</span> <span class="o">=</span> <span class="kc">False</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_last_client_id</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">1</span>
<span class="nd">@abstractmethod</span>
<span class="k">def</span> <span class="nf">submit</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">request</span><span class="p">:</span> <span class="n">GenerationRequest</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="n">GenerationResult</span><span class="p">:</span>
<span class="k">pass</span>
<span class="k">def</span> <span class="nf">generate_async</span><span class="p">(</span>
<span class="bp">self</span><span class="p">,</span>
<span class="n">prompt_token_ids</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">],</span>
<span class="n">sampling_params</span><span class="p">:</span> <span class="n">SamplingParams</span><span class="p">,</span>
<span class="n">query_token_ids</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">np</span><span class="o">.</span><span class="n">ndarray</span><span class="p">,</span> <span class="nb">list</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">lora_request</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">LoRARequest</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">prompt_adapter_request</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">PromptAdapterRequest</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">streaming</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
<span class="n">prompt_tuning_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">list</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="p">)</span> <span class="o">-&gt;</span> <span class="n">GenerationResult</span><span class="p">:</span>
<span class="w"> </span><span class="sd">&quot;&quot;&quot;Generate output for the given prompt token ids in the asynchronous mode.</span>
<span class="sd"> Asynchronous generation accepts single prompt only.</span>
<span class="sd"> &quot;&quot;&quot;</span>
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">prompt_token_ids</span><span class="p">[</span><span class="mi">0</span><span class="p">],</span> <span class="nb">int</span><span class="p">)</span>
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">sampling_params</span><span class="p">,</span> <span class="n">SamplingParams</span><span class="p">)</span>
<span class="n">result</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">submit</span><span class="p">(</span>
<span class="n">GenerationRequest</span><span class="p">(</span><span class="n">prompt_token_ids</span><span class="p">,</span>
<span class="n">sampling_params</span><span class="o">=</span><span class="n">sampling_params</span><span class="p">,</span>
<span class="n">query_token_ids</span><span class="o">=</span><span class="n">query_token_ids</span><span class="p">,</span>
<span class="n">lora_request</span><span class="o">=</span><span class="n">lora_request</span><span class="p">,</span>
<span class="n">prompt_adapter_request</span><span class="o">=</span><span class="n">prompt_adapter_request</span><span class="p">,</span>
<span class="n">streaming</span><span class="o">=</span><span class="n">streaming</span><span class="p">,</span>
<span class="n">prompt_tuning_config</span><span class="o">=</span><span class="n">prompt_tuning_config</span><span class="p">))</span>
<span class="k">return</span> <span class="n">result</span>
<span class="k">def</span> <span class="nf">generate</span><span class="p">(</span>
<span class="bp">self</span><span class="p">,</span>
<span class="n">prompt_token_ids</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">],</span> <span class="n">List</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]]],</span>
<span class="n">sampling_params</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">SamplingParams</span><span class="p">,</span> <span class="n">List</span><span class="p">[</span><span class="n">SamplingParams</span><span class="p">]],</span>
<span class="n">query_token_ids</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">np</span><span class="o">.</span><span class="n">ndarray</span><span class="p">,</span> <span class="nb">list</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">lora_request</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span><span class="n">LoRARequest</span><span class="p">,</span> <span class="n">List</span><span class="p">[</span><span class="n">LoRARequest</span><span class="p">]]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">prompt_adapter_request</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span>
<span class="n">PromptAdapterRequest</span><span class="p">,</span> <span class="n">List</span><span class="p">[</span><span class="n">PromptAdapterRequest</span><span class="p">]]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="p">)</span> <span class="o">-&gt;</span> <span class="n">Union</span><span class="p">[</span><span class="n">GenerationResult</span><span class="p">,</span> <span class="n">List</span><span class="p">[</span><span class="n">GenerationResult</span><span class="p">]]:</span>
<span class="w"> </span><span class="sd">&quot;&quot;&quot;Generate output for the given prompt token ids in the synchronous mode.</span>
<span class="sd"> Synchronous generation accepts either single prompt or batched prompts.</span>
<span class="sd"> &quot;&quot;&quot;</span>
<span class="n">unbatched</span> <span class="o">=</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">prompt_token_ids</span><span class="p">[</span><span class="mi">0</span><span class="p">],</span> <span class="nb">int</span><span class="p">)</span>
<span class="k">if</span> <span class="n">unbatched</span><span class="p">:</span>
<span class="n">prompt_token_ids</span> <span class="o">=</span> <span class="p">[</span><span class="n">prompt_token_ids</span><span class="p">]</span>
<span class="k">if</span> <span class="n">query_token_ids</span><span class="p">:</span>
<span class="n">query_token_ids</span> <span class="o">=</span> <span class="p">[</span><span class="n">query_token_ids</span><span class="p">]</span>
<span class="n">futures</span> <span class="o">=</span> <span class="p">[]</span>
<span class="k">for</span> <span class="n">i</span><span class="p">,</span> <span class="n">p</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">prompt_token_ids</span><span class="p">):</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">sampling_params</span><span class="p">,</span> <span class="nb">list</span><span class="p">):</span>
<span class="n">sp</span> <span class="o">=</span> <span class="n">sampling_params</span><span class="p">[</span><span class="n">i</span><span class="p">]</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">sp</span> <span class="o">=</span> <span class="n">sampling_params</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">lora_request</span><span class="p">,</span> <span class="nb">list</span><span class="p">):</span>
<span class="n">lora_req</span> <span class="o">=</span> <span class="n">lora_request</span><span class="p">[</span><span class="n">i</span><span class="p">]</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">lora_req</span> <span class="o">=</span> <span class="n">lora_request</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">prompt_adapter_request</span><span class="p">,</span> <span class="nb">list</span><span class="p">):</span>
<span class="n">pa_req</span> <span class="o">=</span> <span class="n">prompt_adapter_request</span><span class="p">[</span><span class="n">i</span><span class="p">]</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">pa_req</span> <span class="o">=</span> <span class="n">prompt_adapter_request</span>
<span class="n">future</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">generate_async</span><span class="p">(</span><span class="n">p</span><span class="p">,</span>
<span class="n">sampling_params</span><span class="o">=</span><span class="n">sp</span><span class="p">,</span>
<span class="n">query_token_ids</span><span class="o">=</span><span class="n">query_token_ids</span><span class="p">,</span>
<span class="n">lora_request</span><span class="o">=</span><span class="n">lora_req</span><span class="p">,</span>
<span class="n">prompt_adapter_request</span><span class="o">=</span><span class="n">pa_req</span><span class="p">,</span>
<span class="n">streaming</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
<span class="n">futures</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">future</span><span class="p">)</span>
<span class="k">for</span> <span class="n">future</span> <span class="ow">in</span> <span class="n">futures</span><span class="p">:</span>
<span class="n">future</span><span class="o">.</span><span class="n">result</span><span class="p">()</span>
<span class="k">if</span> <span class="n">unbatched</span><span class="p">:</span>
<span class="n">futures</span> <span class="o">=</span> <span class="n">futures</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
<span class="k">return</span> <span class="n">futures</span>
<span class="k">def</span> <span class="nf">_get_next_client_id</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="c1"># (self._last_client_id + 1) % UINT64_MAX</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_last_client_id</span> <span class="o">=</span> <span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_last_client_id</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)</span> <span class="o">&amp;</span> <span class="p">((</span><span class="mi">1</span> <span class="o">&lt;&lt;</span> <span class="mi">64</span><span class="p">)</span> <span class="o">-</span> <span class="mi">1</span><span class="p">)</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_last_client_id</span>
<span class="k">def</span> <span class="nf">_handle_background_error</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">error</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="ne">Exception</span> <span class="o">|</span> <span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span>
<span class="w"> </span><span class="sd">&quot;&quot;&quot; Process the errors from the threads or processes.</span>
<span class="sd"> NOTE: This should be called in the main thread.</span>
<span class="sd"> &quot;&quot;&quot;</span>
<span class="k">if</span> <span class="n">error</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="c1"># For details please refer to the comment of `GenerationResult.error`</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">error</span><span class="p">,</span> <span class="ne">Exception</span><span class="p">):</span>
<span class="c1"># Serious error from background thread or process</span>
<span class="k">if</span> <span class="n">enable_llm_debug</span><span class="p">():</span>
<span class="n">print_colored</span><span class="p">(</span>
<span class="sa">f</span><span class="s2">&quot;Got background error: </span><span class="si">{</span><span class="nb">repr</span><span class="p">(</span><span class="n">error</span><span class="p">)</span><span class="si">}</span><span class="s2">, will shutdown the LLM instance</span><span class="se">\n</span><span class="s2">&quot;</span><span class="p">,</span>
<span class="s2">&quot;red&quot;</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
<span class="k">raise</span> <span class="n">error</span>
<span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">error</span><span class="p">,</span> <span class="nb">str</span><span class="p">):</span>
<span class="k">if</span> <span class="n">enable_llm_debug</span><span class="p">():</span>
<span class="n">print_colored</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;Got per-request error: </span><span class="si">{</span><span class="nb">repr</span><span class="p">(</span><span class="n">error</span><span class="p">)</span><span class="si">}</span><span class="se">\n</span><span class="s2">&quot;</span><span class="p">,</span>
<span class="s2">&quot;red&quot;</span><span class="p">)</span>
<span class="n">print_colored</span><span class="p">(</span><span class="nb">str</span><span class="p">(</span><span class="n">traceback</span><span class="o">.</span><span class="n">extract_stack</span><span class="p">())</span> <span class="o">+</span> <span class="s2">&quot;</span><span class="se">\n</span><span class="s2">&quot;</span><span class="p">,</span> <span class="s2">&quot;red&quot;</span><span class="p">)</span>
<span class="c1"># A per-request error, can be captured and ignored</span>
<span class="k">raise</span> <span class="n">RequestError</span><span class="p">(</span><span class="n">error</span><span class="p">)</span>
<span class="c1"># Here we raise the first error in the queue. This method will be called repeatedly and user can choose to catch</span>
<span class="c1"># more than one error.</span>
<span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="o">.</span><span class="n">empty</span><span class="p">():</span>
<span class="n">e</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="o">.</span><span class="n">task_done</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
<span class="c1"># We can catch some exceptions here.</span>
<span class="k">raise</span> <span class="n">e</span>
<span class="nd">@abstractmethod</span>
<span class="k">def</span> <span class="nf">shutdown</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">pass</span>
<span class="k">def</span> <span class="nf">create_stats_queue</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="c1"># Stats queue is created during first submission to ensure event loop exists if it is needed.</span>
<span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">_stats</span><span class="p">:</span>
<span class="k">if</span> <span class="n">has_event_loop</span><span class="p">():</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_stats</span> <span class="o">=</span> <span class="n">AsyncQueue</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_stats</span><span class="o">.</span><span class="n">sync_q</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stats_aqueue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_stats</span>
<span class="k">else</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_stats</span> <span class="o">=</span> <span class="n">Queue</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_stats</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stats_aqueue</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">def</span> <span class="nf">get_stats</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">timeout</span><span class="o">=</span><span class="kc">None</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">str</span><span class="p">:</span>
<span class="w"> </span><span class="sd">&#39;&#39;&#39; Get the stats from the runtime.</span>
<span class="sd"> Exceptions:</span>
<span class="sd"> NoStatsAvailable: If the stats are not available.</span>
<span class="sd"> Returns:</span>
<span class="sd"> str: The stats in JSON format.</span>
<span class="sd"> Known issue:</span>
<span class="sd"> The `get_stats` cannot mix with `aget_stats` in the same Executor instance.</span>
<span class="sd"> &#39;&#39;&#39;</span>
<span class="k">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span><span class="p">,</span> <span class="s2">&quot;The stats queue is not created. It is likely that `get_stats` and `aget_stats` methods&quot;</span> \
<span class="s2">&quot; are mixed.&quot;</span>
<span class="k">try</span><span class="p">:</span>
<span class="n">res</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="n">timeout</span><span class="o">=</span><span class="n">timeout</span><span class="p">)</span>
<span class="k">except</span> <span class="n">Empty</span><span class="p">:</span>
<span class="k">raise</span> <span class="n">NoStatsAvailable</span>
<span class="k">return</span> <span class="n">res</span>
<span class="k">async</span> <span class="k">def</span> <span class="nf">aget_stats</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">timeout</span><span class="o">=</span><span class="kc">None</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]:</span>
<span class="w"> </span><span class="sd">&#39;&#39;&#39; Get the stats from the runtime.</span>
<span class="sd"> Exceptions:</span>
<span class="sd"> NoStatsAvailable: If the stats are not available.</span>
<span class="sd"> Returns:</span>
<span class="sd"> str: The stats in JSON format.</span>
<span class="sd"> Known issue:</span>
<span class="sd"> The `aget_stats` cannot mix with `get_stats` in the same Executor instance.</span>
<span class="sd"> &#39;&#39;&#39;</span>
<span class="bp">self</span><span class="o">.</span><span class="n">create_stats_queue</span><span class="p">()</span>
<span class="k">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">stats_aqueue</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
<span class="k">if</span> <span class="ow">not</span> <span class="n">has_event_loop</span><span class="p">():</span>
<span class="k">raise</span> <span class="n">NoStatsAvailable</span>
<span class="k">try</span><span class="p">:</span>
<span class="n">res</span> <span class="o">=</span> <span class="k">await</span> <span class="bp">self</span><span class="o">.</span><span class="n">stats_aqueue</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="n">timeout</span><span class="o">=</span><span class="n">timeout</span><span class="p">)</span>
<span class="k">except</span> <span class="n">asyncio</span><span class="o">.</span><span class="n">TimeoutError</span><span class="p">:</span>
<span class="k">raise</span> <span class="n">NoStatsAvailable</span>
<span class="k">return</span> <span class="n">res</span>
<span class="nd">@staticmethod</span>
<span class="k">def</span> <span class="nf">create</span><span class="p">(</span>
<span class="n">engine</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">Path</span><span class="p">,</span> <span class="n">Engine</span><span class="p">],</span>
<span class="n">executor_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">tllm</span><span class="o">.</span><span class="n">ExecutorConfig</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">logits_post_processor_map</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Callable</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">model_world_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">1</span><span class="p">,</span>
<span class="n">world_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">0</span><span class="p">,</span>
<span class="n">mpi_session</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">MpiSession</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">reuse_mpi_comm</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
<span class="n">return_logits</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
<span class="p">)</span> <span class="o">-&gt;</span> <span class="n">Union</span><span class="p">[</span><span class="s2">&quot;ExecutorBindingsProxy&quot;</span><span class="p">,</span> <span class="s2">&quot;ExecutorBindingsWorker&quot;</span><span class="p">]:</span>
<span class="k">if</span> <span class="n">world_size</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
<span class="n">world_size</span> <span class="o">=</span> <span class="n">mpi_world_size</span><span class="p">()</span>
<span class="k">if</span> <span class="n">world_size</span> <span class="o">&gt;</span> <span class="mi">1</span> <span class="ow">and</span> <span class="n">world_size</span> <span class="o">&lt;</span> <span class="n">model_world_size</span><span class="p">:</span>
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span>
<span class="s2">&quot;Cannot instantiate Generator for engine built &quot;</span>
<span class="sa">f</span><span class="s2">&quot;for </span><span class="si">{</span><span class="n">model_world_size</span><span class="si">}</span><span class="s2"> ranks, while currently running &quot;</span>
<span class="sa">f</span><span class="s2">&quot;on </span><span class="si">{</span><span class="n">world_size</span><span class="si">}</span><span class="s2"> ranks.&quot;</span><span class="p">)</span>
<span class="n">worker_kwargs</span> <span class="o">=</span> <span class="p">{</span>
<span class="s2">&quot;engine&quot;</span><span class="p">:</span> <span class="n">engine</span><span class="p">,</span>
<span class="s2">&quot;executor_config&quot;</span><span class="p">:</span> <span class="n">executor_config</span><span class="p">,</span>
<span class="s2">&quot;logits_post_processor_map&quot;</span><span class="p">:</span> <span class="n">logits_post_processor_map</span><span class="p">,</span>
<span class="p">}</span>
<span class="c1"># The case where the Python main process is launched by mpirun</span>
<span class="n">mpirun_launch</span> <span class="o">=</span> <span class="n">external_mpi_comm_available</span><span class="p">(</span><span class="n">model_world_size</span><span class="p">)</span>
<span class="c1"># The case where the Python main process utilizes mpi4py to spawn MPI workers</span>
<span class="n">spawn_workers</span> <span class="o">=</span> <span class="n">need_spawn_mpi_workers</span><span class="p">(</span><span class="n">model_world_size</span><span class="p">)</span>
<span class="k">if</span> <span class="n">spawn_workers</span> <span class="ow">or</span> <span class="p">(</span><span class="n">mpirun_launch</span> <span class="ow">and</span> <span class="n">reuse_mpi_comm</span><span class="p">):</span>
<span class="k">if</span> <span class="n">reuse_mpi_comm</span><span class="p">:</span>
<span class="k">assert</span> <span class="n">mpi_session</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">,</span> <span class="s2">&quot;reuse_mpi_comm requires an external MPI session&quot;</span>
<span class="k">return</span> <span class="n">ExecutorBindingsProxy</span><span class="p">(</span><span class="n">worker_kwargs</span><span class="p">,</span>
<span class="n">model_world_size</span><span class="o">=</span><span class="n">model_world_size</span><span class="p">,</span>
<span class="n">mpi_session</span><span class="o">=</span><span class="n">mpi_session</span><span class="p">)</span>
<span class="c1"># WAR: For the performance of gathering logits, we use single process worker</span>
<span class="c1"># for TP1 to avoid the large overhead of IPC.</span>
<span class="c1"># WAR: Developers can enable this manually, this will be easier for TP1</span>
<span class="c1"># debugging. We will introduce a better solution in the future.</span>
<span class="k">if</span> <span class="n">return_logits</span> <span class="ow">or</span> <span class="n">enable_worker_single_process_for_tp1</span><span class="p">():</span>
<span class="n">logger</span><span class="o">.</span><span class="n">warning</span><span class="p">(</span>
<span class="s2">&quot;Using single process worker for TP1, this may hurt streaming generation performance.&quot;</span>
<span class="p">)</span>
<span class="k">return</span> <span class="n">ExecutorBindingsWorker</span><span class="p">(</span><span class="o">**</span><span class="n">worker_kwargs</span><span class="p">)</span>
<span class="c1"># For single-gpu case:</span>
<span class="c1"># Partition the workload to multiple process for streaming performance.</span>
<span class="c1"># While this requires uses to protect their entrypoint to</span>
<span class="c1"># `if __name__ == &quot;__main__&quot;:`.</span>
<span class="k">if</span> <span class="ow">not</span> <span class="n">platform</span><span class="o">.</span><span class="n">system</span><span class="p">()</span> <span class="o">==</span> <span class="s1">&#39;Windows&#39;</span><span class="p">:</span>
<span class="k">return</span> <span class="n">ExecutorBindingsProxy</span><span class="p">(</span><span class="n">worker_kwargs</span><span class="p">,</span>
<span class="n">model_world_size</span><span class="o">=</span><span class="n">model_world_size</span><span class="p">,</span>
<span class="n">mpi_session</span><span class="o">=</span><span class="kc">None</span><span class="p">)</span> <span class="c1"># use mpi4py</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">ctx</span> <span class="o">=</span> <span class="n">multiprocessing</span><span class="o">.</span><span class="n">get_context</span><span class="p">(</span><span class="s2">&quot;spawn&quot;</span><span class="p">)</span>
<span class="c1"># The ProcessPoolExecutorSession is used to support Windows, as mpi4py cannot.</span>
<span class="n">mpi_session</span> <span class="o">=</span> <span class="n">ProcessPoolExecutorSession</span><span class="p">(</span><span class="n">n_workers</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span>
<span class="n">mp_context</span><span class="o">=</span><span class="n">ctx</span><span class="p">)</span>
<span class="k">return</span> <span class="n">ExecutorBindingsProxy</span><span class="p">(</span><span class="n">worker_kwargs</span><span class="p">,</span>
<span class="n">model_world_size</span><span class="o">=</span><span class="n">model_world_size</span><span class="p">,</span>
<span class="n">mpi_session</span><span class="o">=</span><span class="n">mpi_session</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">wait_first_completed</span><span class="p">(</span>
<span class="bp">self</span><span class="p">,</span> <span class="n">futures</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="n">GenerationResult</span><span class="p">]</span>
<span class="p">)</span> <span class="o">-&gt;</span> <span class="n">Generator</span><span class="p">[</span><span class="n">GenerationResult</span><span class="p">,</span> <span class="kc">None</span><span class="p">,</span> <span class="kc">None</span><span class="p">]:</span>
<span class="n">wait_set</span> <span class="o">=</span> <span class="nb">set</span><span class="p">(</span><span class="n">futures</span><span class="p">)</span>
<span class="c1"># clear already-finished requests</span>
<span class="k">for</span> <span class="n">f</span> <span class="ow">in</span> <span class="n">futures</span><span class="p">:</span>
<span class="k">if</span> <span class="n">f</span><span class="o">.</span><span class="n">_done</span><span class="p">:</span>
<span class="n">wait_set</span><span class="o">.</span><span class="n">pop</span><span class="p">(</span><span class="n">f</span><span class="p">)</span>
<span class="k">yield</span> <span class="n">f</span>
<span class="c1"># wait remaining active requests</span>
<span class="k">while</span> <span class="nb">len</span><span class="p">(</span><span class="n">wait_set</span><span class="p">)</span> <span class="o">&gt;</span> <span class="mi">0</span><span class="p">:</span>
<span class="n">fut</span> <span class="o">=</span> <span class="n">wait_set</span><span class="o">.</span><span class="n">pop</span><span class="p">()</span>
<span class="k">if</span> <span class="n">fut</span><span class="o">.</span><span class="n">request_id</span> <span class="ow">not</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="p">:</span>
<span class="k">yield</span> <span class="n">fut</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">wait_set</span><span class="o">.</span><span class="n">add</span><span class="p">(</span><span class="n">fut</span><span class="p">)</span>
<span class="k">class</span> <span class="nc">ProcessPoolExecutorSession</span><span class="p">(</span><span class="n">MpiSession</span><span class="p">):</span>
<span class="c1"># This process pool is introduced for better recoverable exceptions handling.</span>
<span class="c1"># It replaces MpiPoolExecutor for single-gpu case.</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">n_workers</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">n_workers</span> <span class="o">=</span> <span class="n">n_workers</span>
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_pool</span> <span class="o">=</span> <span class="n">ProcessPoolExecutor</span><span class="p">(</span><span class="n">max_workers</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">n_workers</span><span class="p">,</span>
<span class="o">**</span><span class="n">kwargs</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">submit</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">task</span><span class="p">:</span> <span class="n">Callable</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="o">-&gt;</span> <span class="n">List</span><span class="p">[</span><span class="n">concurrent</span><span class="o">.</span><span class="n">futures</span><span class="o">.</span><span class="n">Future</span><span class="p">]:</span>
<span class="k">return</span> <span class="p">[</span>
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_pool</span><span class="o">.</span><span class="n">submit</span><span class="p">(</span><span class="n">task</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">for</span> <span class="n">i</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">n_workers</span><span class="p">)</span>
<span class="p">]</span>
<span class="k">def</span> <span class="nf">submit_sync</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">task</span><span class="p">:</span> <span class="n">Callable</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="o">-&gt;</span> <span class="n">List</span><span class="p">[</span><span class="n">Any</span><span class="p">]:</span>
<span class="n">futures</span> <span class="o">=</span> <span class="p">[</span>
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_pool</span><span class="o">.</span><span class="n">submit</span><span class="p">(</span><span class="n">task</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">for</span> <span class="n">i</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">n_workers</span><span class="p">)</span>
<span class="p">]</span>
<span class="k">return</span> <span class="p">[</span><span class="n">future</span><span class="o">.</span><span class="n">result</span><span class="p">()</span> <span class="k">for</span> <span class="n">future</span> <span class="ow">in</span> <span class="n">futures</span><span class="p">]</span>
<span class="k">def</span> <span class="nf">shutdown</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_pool</span><span class="o">.</span><span class="n">shutdown</span><span class="p">(</span><span class="n">wait</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
<span class="k">class</span> <span class="nc">ExecutorBindingsWorker</span><span class="p">(</span><span class="n">GenerationExecutor</span><span class="p">):</span>
<span class="k">class</span> <span class="nc">WorkerExit</span><span class="p">(</span><span class="ne">GeneratorExit</span><span class="p">):</span>
<span class="k">pass</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span>
<span class="bp">self</span><span class="p">,</span>
<span class="n">engine</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">Path</span><span class="p">,</span> <span class="n">Engine</span><span class="p">],</span>
<span class="n">executor_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">tllm</span><span class="o">.</span><span class="n">ExecutorConfig</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">logits_post_processor_map</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Callable</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="p">)</span> <span class="o">-&gt;</span> <span class="kc">None</span><span class="p">:</span>
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">engine</span> <span class="o">=</span> <span class="kc">None</span>
<span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span> <span class="o">=</span> <span class="kc">None</span>
<span class="bp">self</span><span class="o">.</span><span class="n">rank</span> <span class="o">=</span> <span class="n">mpi_rank</span><span class="p">()</span>
<span class="c1"># mapping: client_id -&gt; GenerationResult</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">int</span><span class="p">,</span> <span class="n">GenerationResult</span><span class="p">]</span> <span class="o">=</span> <span class="p">{}</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">engine</span><span class="p">,</span> <span class="nb">list</span><span class="p">):</span>
<span class="n">engine</span> <span class="o">=</span> <span class="n">engine</span><span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">rank</span><span class="p">]</span>
<span class="k">if</span> <span class="n">executor_config</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">executor_config</span> <span class="o">=</span> <span class="n">tllm</span><span class="o">.</span><span class="n">ExecutorConfig</span><span class="p">(</span><span class="mi">1</span><span class="p">)</span>
<span class="k">if</span> <span class="n">logits_post_processor_map</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">executor_config</span><span class="o">.</span><span class="n">logits_post_processor_config</span> <span class="o">=</span> <span class="n">tllm</span><span class="o">.</span><span class="n">LogitsPostProcessorConfig</span><span class="p">(</span>
<span class="n">processor_map</span><span class="o">=</span><span class="n">logits_post_processor_map</span><span class="p">,</span> <span class="n">replicate</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">_create_engine</span><span class="p">():</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">engine</span><span class="p">,</span> <span class="n">Engine</span><span class="p">):</span>
<span class="k">return</span> <span class="n">tllm</span><span class="o">.</span><span class="n">Executor</span><span class="p">(</span><span class="n">engine</span><span class="o">.</span><span class="n">engine</span><span class="p">,</span>
<span class="n">json</span><span class="o">.</span><span class="n">dumps</span><span class="p">(</span><span class="n">engine</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">to_dict</span><span class="p">(),</span>
<span class="bp">cls</span><span class="o">=</span><span class="n">ConfigEncoder</span><span class="p">),</span>
<span class="n">tllm</span><span class="o">.</span><span class="n">ModelType</span><span class="o">.</span><span class="n">DECODER_ONLY</span><span class="p">,</span>
<span class="n">executor_config</span><span class="o">=</span><span class="n">executor_config</span><span class="p">,</span>
<span class="n">managed_weights</span><span class="o">=</span><span class="n">engine</span><span class="o">.</span><span class="n">managed_weights</span><span class="p">)</span>
<span class="n">use_default_executor</span> <span class="o">=</span> <span class="ow">not</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">executor_config</span><span class="p">,</span> <span class="s2">&quot;backend&quot;</span><span class="p">)</span>
<span class="k">if</span> <span class="n">use_default_executor</span><span class="p">:</span>
<span class="k">return</span> <span class="n">tllm</span><span class="o">.</span><span class="n">Executor</span><span class="p">(</span><span class="n">engine</span><span class="p">,</span> <span class="n">tllm</span><span class="o">.</span><span class="n">ModelType</span><span class="o">.</span><span class="n">DECODER_ONLY</span><span class="p">,</span>
<span class="n">executor_config</span><span class="p">)</span>
<span class="kn">from</span> <span class="nn">tensorrt_llm._torch.pyexecutor.backend_registries.backend_registry</span> <span class="kn">import</span> \
<span class="n">unique_create_executor</span>
<span class="k">return</span> <span class="n">unique_create_executor</span><span class="p">(</span><span class="n">engine</span><span class="p">,</span>
<span class="n">tllm</span><span class="o">.</span><span class="n">ModelType</span><span class="o">.</span><span class="n">DECODER_ONLY</span><span class="p">,</span>
<span class="n">executor_config</span><span class="o">=</span><span class="n">executor_config</span><span class="p">,</span>
<span class="n">device_id</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">rank</span> <span class="o">%</span>
<span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">device_count</span><span class="p">())</span>
<span class="bp">self</span><span class="o">.</span><span class="n">engine</span> <span class="o">=</span> <span class="n">_create_engine</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_lora_manager</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">LoraManager</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_prompt_adapter_manager</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">PromptAdapterManager</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_runtime_model_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">ModelConfig</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">rank</span> <span class="o">==</span> <span class="mi">0</span> <span class="ow">and</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="p">,</span> <span class="n">tllm</span><span class="o">.</span><span class="n">Executor</span><span class="p">):</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">engine</span><span class="p">,</span> <span class="n">Engine</span><span class="p">):</span>
<span class="n">engine_config</span> <span class="o">=</span> <span class="n">engine</span><span class="o">.</span><span class="n">config</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">engine_config</span> <span class="o">=</span> <span class="n">EngineConfig</span><span class="o">.</span><span class="n">from_json_file</span><span class="p">(</span>
<span class="sa">f</span><span class="s2">&quot;</span><span class="si">{</span><span class="n">engine</span><span class="si">}</span><span class="s2">/config.json&quot;</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_runtime_model_config</span> <span class="o">=</span> <span class="n">_engine_config_to_model_config</span><span class="p">(</span>
<span class="n">engine_config</span><span class="p">)</span>
<span class="k">if</span> <span class="n">engine_config</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">plugin_config</span><span class="o">.</span><span class="n">lora_plugin</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_lora_manager</span> <span class="o">=</span> <span class="n">LoraManager</span><span class="p">()</span>
<span class="k">if</span> <span class="n">engine_config</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">max_prompt_embedding_table_size</span> <span class="o">&gt;</span> <span class="mi">0</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_prompt_adapter_manager</span> <span class="o">=</span> <span class="n">PromptAdapterManager</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">await_response_thread</span> <span class="o">=</span> <span class="n">ManagedThread</span><span class="p">(</span>
<span class="bp">self</span><span class="o">.</span><span class="n">await_response_task</span><span class="p">,</span>
<span class="n">error_queue</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="p">,</span>
<span class="n">name</span><span class="o">=</span><span class="s2">&quot;await_response_thread&quot;</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span> <span class="o">=</span> <span class="n">ManagedThread</span><span class="p">(</span>
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_task</span><span class="p">,</span>
<span class="n">error_queue</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="p">,</span>
<span class="n">name</span><span class="o">=</span><span class="s2">&quot;dispatch_stats_thread&quot;</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">create_stats_queue</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="c1"># Stats queue is created during first submission to ensure event loop exists if it is needed.</span>
<span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">_stats</span><span class="p">:</span>
<span class="k">if</span> <span class="n">has_event_loop</span><span class="p">():</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_stats</span> <span class="o">=</span> <span class="n">AsyncQueue</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_stats</span><span class="o">.</span><span class="n">sync_q</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stats_aqueue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_stats</span>
<span class="k">else</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_stats</span> <span class="o">=</span> <span class="n">Queue</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_stats</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stats_aqueue</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">def</span> <span class="nf">set_result_queue</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">queue</span><span class="p">):</span>
<span class="w"> </span><span class="sd">&quot;&quot;&quot;In multi-gpu mode, result_queue will be set here to communicate between the proxy and the worker 0 process.&quot;&quot;&quot;</span>
<span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span> <span class="o">=</span> <span class="n">queue</span>
<span class="k">def</span> <span class="nf">set_stats_queue</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">queue</span><span class="p">):</span>
<span class="w"> </span><span class="sd">&quot;&quot;&quot;In multi-gpu mode, stats_queue will be set here to communicate between the proxy and the worker 0 process.&quot;&quot;&quot;</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_stats</span> <span class="o">=</span> <span class="n">queue</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_stats</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stats_aqueue</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">def</span> <span class="nf">return_queue</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">client_id</span><span class="p">:</span> <span class="nb">int</span><span class="p">):</span>
<span class="w"> </span><span class="sd">&quot;&quot;&quot; If a centralized result queue is registered (used for communication with the proxy)</span>
<span class="sd"> send the message there.</span>
<span class="sd"> Otherwise, push the result directly in the GenerationResult queue.</span>
<span class="sd"> &quot;&quot;&quot;</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="p">[</span><span class="n">client_id</span><span class="p">]</span><span class="o">.</span><span class="n">queue</span>
<span class="k">def</span> <span class="nf">start_awaiter_thread</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">can_enqueue_requests</span><span class="p">(</span>
<span class="p">)</span> <span class="ow">and</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">await_response_thread</span><span class="o">.</span><span class="n">is_alive</span><span class="p">():</span>
<span class="bp">self</span><span class="o">.</span><span class="n">await_response_thread</span><span class="o">.</span><span class="n">start</span><span class="p">()</span>
<span class="k">def</span> <span class="nf">start_stats_thread</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">can_enqueue_requests</span><span class="p">(</span>
<span class="p">)</span> <span class="ow">and</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span><span class="o">.</span><span class="n">is_alive</span><span class="p">():</span>
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span><span class="o">.</span><span class="n">start</span><span class="p">()</span>
<span class="k">def</span> <span class="nf">_engine_response_callback</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">response</span><span class="p">:</span> <span class="n">tllm</span><span class="o">.</span><span class="n">Response</span><span class="p">):</span>
<span class="k">return</span> <span class="n">response</span>
<span class="k">def</span> <span class="nf">await_response_task</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">bool</span><span class="p">:</span>
<span class="c1"># TODO(zhenhuan, chunwei): keep only batch branch after more tests</span>
<span class="k">if</span> <span class="n">BATCH_RESP_IN_AWAIT</span><span class="p">:</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_batch_await_response_task</span><span class="p">()</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_await_response_task</span><span class="p">()</span>
<span class="k">def</span> <span class="nf">_await_response_task</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">bool</span><span class="p">:</span>
<span class="c1"># Get responses and place in queue.</span>
<span class="n">async_queues</span> <span class="o">=</span> <span class="p">[]</span>
<span class="n">event_loop</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">for</span> <span class="n">response</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">await_responses</span><span class="p">(</span><span class="n">timeout</span><span class="o">=</span><span class="n">datetime</span><span class="o">.</span><span class="n">timedelta</span><span class="p">(</span>
<span class="n">milliseconds</span><span class="o">=</span><span class="mi">100</span><span class="p">)):</span>
<span class="n">response</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_engine_response_callback</span><span class="p">(</span><span class="n">response</span><span class="p">)</span>
<span class="k">if</span> <span class="n">response</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="k">continue</span>
<span class="n">client_id</span> <span class="o">=</span> <span class="n">response</span><span class="o">.</span><span class="n">client_id</span>
<span class="k">assert</span> <span class="n">client_id</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
<span class="k">if</span> <span class="n">response</span><span class="o">.</span><span class="n">has_error</span><span class="p">():</span>
<span class="c1"># This error will be dispatched to the user&#39;s generate_async for the corresponding request. It won&#39;t</span>
<span class="c1"># stop the whole service.</span>
<span class="n">rsp</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">Response</span><span class="p">(</span>
<span class="n">client_id</span><span class="p">,</span>
<span class="n">tensors</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="c1"># Note: error Response only has one finish reason.</span>
<span class="c1"># Since the error will be raised in the main thread, so the finish reason is not actually used.</span>
<span class="n">finish_reasons</span><span class="o">=</span><span class="p">[</span><span class="n">tllm</span><span class="o">.</span><span class="n">FinishReason</span><span class="o">.</span><span class="n">NOT_FINISHED</span><span class="p">],</span>
<span class="n">is_final</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
<span class="n">sequence_index</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">error</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">error_msg</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">tensors</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">ResponseTensors</span><span class="p">(</span>
<span class="n">output_token_ids</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">output_token_ids</span><span class="p">,</span>
<span class="n">context_logits</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">context_logits</span><span class="p">,</span>
<span class="n">generation_logits</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">generation_logits</span><span class="p">,</span>
<span class="n">log_probs</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">log_probs</span><span class="p">,</span>
<span class="n">cum_log_probs</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">cum_log_probs</span><span class="p">,</span>
<span class="p">)</span>
<span class="n">rsp</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">Response</span><span class="p">(</span>
<span class="n">client_id</span><span class="p">,</span>
<span class="n">tensors</span><span class="p">,</span>
<span class="n">finish_reasons</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">finish_reasons</span><span class="p">,</span>
<span class="n">is_final</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">is_final</span><span class="p">,</span>
<span class="n">sequence_index</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">sequence_index</span><span class="p">,</span>
<span class="n">error</span><span class="o">=</span><span class="kc">None</span><span class="p">)</span>
<span class="n">queue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">return_queue</span><span class="p">(</span><span class="n">client_id</span><span class="p">)</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_has_background_error</span><span class="p">():</span>
<span class="n">rsp</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_create_error_response</span><span class="p">(</span><span class="n">client_id</span><span class="p">)</span>
<span class="c1"># For AsyncQueue.sync_q, we will batch the events to avoid too many event notifications, thus put without</span>
<span class="c1"># wait here.</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">queue</span><span class="p">,</span> <span class="n">_SyncQueue</span><span class="p">):</span>
<span class="n">global_tracer</span><span class="p">()</span><span class="o">.</span><span class="n">log_instant</span><span class="p">(</span><span class="s2">&quot;worker-rsp.put&quot;</span><span class="p">)</span>
<span class="n">queue</span><span class="o">.</span><span class="n">put_nowait</span><span class="p">(</span><span class="n">rsp</span><span class="p">)</span>
<span class="n">async_queues</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">queue</span><span class="p">)</span>
<span class="c1"># all the loops are identical</span>
<span class="n">event_loop</span> <span class="o">=</span> <span class="n">event_loop</span> <span class="ow">or</span> <span class="n">queue</span><span class="o">.</span><span class="n">loop</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">global_tracer</span><span class="p">()</span><span class="o">.</span><span class="n">log_instant</span><span class="p">(</span><span class="s2">&quot;worker-rsp.put&quot;</span><span class="p">)</span>
<span class="n">queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">rsp</span><span class="p">)</span> <span class="c1"># This could be IPC</span>
<span class="c1"># Eliminate the finished GenerationRequest instances timely, which may take considerable memory.</span>
<span class="k">if</span> <span class="n">rsp</span><span class="o">.</span><span class="n">is_final</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="o">.</span><span class="n">pop</span><span class="p">(</span><span class="n">client_id</span><span class="p">,</span> <span class="kc">None</span><span class="p">)</span>
<span class="k">if</span> <span class="n">async_queues</span><span class="p">:</span>
<span class="n">_SyncQueue</span><span class="o">.</span><span class="n">notify_many</span><span class="p">(</span><span class="n">event_loop</span><span class="p">,</span> <span class="n">async_queues</span><span class="p">)</span>
<span class="k">return</span> <span class="kc">True</span> <span class="c1"># success</span>
<span class="k">def</span> <span class="nf">_batch_await_response_task</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">bool</span><span class="p">:</span>
<span class="c1"># Get responses and place in queue.</span>
<span class="n">result_queue_array</span> <span class="o">=</span> <span class="p">[]</span>
<span class="k">for</span> <span class="n">response</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">await_responses</span><span class="p">(</span><span class="n">timeout</span><span class="o">=</span><span class="n">datetime</span><span class="o">.</span><span class="n">timedelta</span><span class="p">(</span>
<span class="n">milliseconds</span><span class="o">=</span><span class="mi">100</span><span class="p">)):</span>
<span class="n">response</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_engine_response_callback</span><span class="p">(</span><span class="n">response</span><span class="p">)</span>
<span class="k">if</span> <span class="n">response</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="k">continue</span>
<span class="n">client_id</span> <span class="o">=</span> <span class="n">response</span><span class="o">.</span><span class="n">client_id</span>
<span class="k">assert</span> <span class="n">client_id</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
<span class="k">if</span> <span class="n">response</span><span class="o">.</span><span class="n">has_error</span><span class="p">():</span>
<span class="c1"># This error will be dispatched to the user&#39;s generate_async for the corresponding request. It won&#39;t</span>
<span class="c1"># stop the whole service.</span>
<span class="n">rsp</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">Response</span><span class="p">(</span>
<span class="n">client_id</span><span class="p">,</span>
<span class="n">tensors</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="c1"># Note: error Response only has one finish reason.</span>
<span class="c1"># Since the error will be raised in the main thread, so the finish reason is not actually used.</span>
<span class="n">finish_reasons</span><span class="o">=</span><span class="p">[</span><span class="n">tllm</span><span class="o">.</span><span class="n">FinishReason</span><span class="o">.</span><span class="n">NOT_FINISHED</span><span class="p">],</span>
<span class="n">is_final</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
<span class="n">sequence_index</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">error</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">error_msg</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">tensors</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">ResponseTensors</span><span class="p">(</span>
<span class="n">output_token_ids</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">output_token_ids</span><span class="p">,</span>
<span class="n">context_logits</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">context_logits</span><span class="p">,</span>
<span class="n">generation_logits</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">generation_logits</span><span class="p">,</span>
<span class="n">log_probs</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">log_probs</span><span class="p">,</span>
<span class="n">cum_log_probs</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">cum_log_probs</span><span class="p">,</span>
<span class="p">)</span>
<span class="n">rsp</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">Response</span><span class="p">(</span>
<span class="n">client_id</span><span class="p">,</span>
<span class="n">tensors</span><span class="p">,</span>
<span class="n">finish_reasons</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">finish_reasons</span><span class="p">,</span>
<span class="n">is_final</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">is_final</span><span class="p">,</span>
<span class="n">sequence_index</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">sequence_index</span><span class="p">,</span>
<span class="n">error</span><span class="o">=</span><span class="kc">None</span><span class="p">)</span>
<span class="n">queue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">return_queue</span><span class="p">(</span><span class="n">client_id</span><span class="p">)</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_has_background_error</span><span class="p">():</span>
<span class="n">rsp</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_create_error_response</span><span class="p">(</span><span class="n">client_id</span><span class="p">)</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">global_tracer</span><span class="p">()</span><span class="o">.</span><span class="n">log_instant</span><span class="p">(</span><span class="s2">&quot;worker-rsp.put&quot;</span><span class="p">)</span>
<span class="n">queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">rsp</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">result_queue_array</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">rsp</span><span class="p">)</span>
<span class="c1"># Eliminate the finished GenerationRequest instances timely, which may take considerable memory.</span>
<span class="k">if</span> <span class="n">rsp</span><span class="o">.</span><span class="n">is_final</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="o">.</span><span class="n">pop</span><span class="p">(</span><span class="n">client_id</span><span class="p">,</span> <span class="kc">None</span><span class="p">)</span>
<span class="k">if</span> <span class="n">result_queue_array</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">result_queue_array</span><span class="p">)</span>
<span class="k">return</span> <span class="kc">True</span> <span class="c1"># success</span>
<span class="k">def</span> <span class="nf">_has_background_error</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">bool</span><span class="p">:</span>
<span class="k">return</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="o">.</span><span class="n">empty</span><span class="p">()</span>
<span class="k">def</span> <span class="nf">_create_error_response</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">client_id</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">:</span>
<span class="n">bck_error</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="o">.</span><span class="n">get_nowait</span><span class="p">()</span>
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">bck_error</span><span class="p">,</span> <span class="ne">Exception</span><span class="p">)</span>
<span class="k">return</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">(</span><span class="n">client_id</span><span class="p">,</span>
<span class="n">tensors</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">finish_reasons</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">is_final</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">sequence_index</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">error</span><span class="o">=</span><span class="n">bck_error</span><span class="p">)</span>
<span class="n">stats_count</span> <span class="o">=</span> <span class="mi">0</span>
<span class="k">def</span> <span class="nf">dispatch_stats_task</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">bool</span><span class="p">:</span>
<span class="n">time</span><span class="o">.</span><span class="n">sleep</span><span class="p">(</span><span class="mf">0.1</span><span class="p">)</span>
<span class="c1"># Get stats and place in queue.</span>
<span class="k">for</span> <span class="n">stats</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">get_latest_iteration_stats</span><span class="p">():</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stats_count</span> <span class="o">+=</span> <span class="mi">1</span>
<span class="k">while</span> <span class="nb">hasattr</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span><span class="p">,</span> <span class="s2">&quot;full&quot;</span><span class="p">)</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span><span class="o">.</span><span class="n">full</span><span class="p">():</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
<span class="k">try</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">stats</span><span class="o">.</span><span class="n">to_json_str</span><span class="p">())</span>
<span class="k">except</span> <span class="n">AsyncQueue</span><span class="o">.</span><span class="n">EventLoopShutdownError</span><span class="p">:</span>
<span class="c1"># This happens in the last stats loop while the generate workflow is stopped.</span>
<span class="k">pass</span>
<span class="k">except</span> <span class="ne">Exception</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
<span class="k">raise</span> <span class="n">e</span>
<span class="k">return</span> <span class="kc">True</span> <span class="c1"># success</span>
<span class="k">def</span> <span class="nf">start</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">create_stats_queue</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">start_awaiter_thread</span><span class="p">()</span>
<span class="c1"># TODO: Replace this with a decent get_stats implementation</span>
<span class="c1">#self.start_stats_thread()</span>
<span class="k">def</span> <span class="nf">_load_lora_adapter</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">lora_request</span><span class="p">:</span> <span class="n">LoRARequest</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_lora_manager</span><span class="o">.</span><span class="n">load_from_ckpt</span><span class="p">(</span>
<span class="p">[</span><span class="n">lora_request</span><span class="o">.</span><span class="n">path</span><span class="p">],</span>
<span class="n">model_config</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_runtime_model_config</span><span class="p">,</span>
<span class="n">runtime_mapping</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">uids</span><span class="o">=</span><span class="p">[</span><span class="nb">str</span><span class="p">(</span><span class="n">lora_request</span><span class="o">.</span><span class="n">adapter_id</span><span class="p">)])</span>
<span class="k">def</span> <span class="nf">_load_prompt_adapter</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span>
<span class="n">prompt_adapter_request</span><span class="p">:</span> <span class="n">PromptAdapterRequest</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_prompt_adapter_manager</span><span class="o">.</span><span class="n">load_from_ckpt</span><span class="p">(</span>
<span class="p">[</span><span class="n">prompt_adapter_request</span><span class="o">.</span><span class="n">local_path</span><span class="p">],</span>
<span class="n">model_config</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_runtime_model_config</span><span class="p">,</span>
<span class="n">uids</span><span class="o">=</span><span class="p">[</span><span class="nb">str</span><span class="p">(</span><span class="n">prompt_adapter_request</span><span class="o">.</span><span class="n">adapter_id</span><span class="p">)])</span>
<span class="k">def</span> <span class="nf">_enqueue_request</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">request</span><span class="p">:</span> <span class="n">GenerationRequest</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">int</span><span class="p">:</span>
<span class="k">assert</span> <span class="n">request</span><span class="o">.</span><span class="n">id</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_lora_manager</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="n">request</span><span class="o">.</span><span class="n">lora_request</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_load_lora_adapter</span><span class="p">(</span><span class="n">request</span><span class="o">.</span><span class="n">lora_request</span><span class="p">)</span>
<span class="n">uid</span> <span class="o">=</span> <span class="nb">str</span><span class="p">(</span><span class="n">request</span><span class="o">.</span><span class="n">lora_request</span><span class="o">.</span><span class="n">adapter_id</span><span class="p">)</span>
<span class="n">lora_config</span> <span class="o">=</span> <span class="n">tllm</span><span class="o">.</span><span class="n">LoraConfig</span><span class="p">(</span>
<span class="n">task_id</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">lora_request</span><span class="o">.</span><span class="n">adapter_id</span><span class="p">,</span>
<span class="n">weights</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_lora_manager</span><span class="o">.</span><span class="n">cpp_lora_weights</span><span class="p">[</span><span class="n">uid</span><span class="p">],</span>
<span class="n">config</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_lora_manager</span><span class="o">.</span><span class="n">cpp_lora_config</span><span class="p">[</span><span class="n">uid</span><span class="p">])</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">lora_config</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">prompt_token_ids</span> <span class="o">=</span> <span class="n">copy</span><span class="o">.</span><span class="n">deepcopy</span><span class="p">(</span><span class="n">request</span><span class="o">.</span><span class="n">prompt_token_ids</span><span class="p">)</span>
<span class="n">prompt_tuning_config</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="n">request</span><span class="o">.</span><span class="n">prompt_adapter_request</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="k">assert</span> <span class="n">request</span><span class="o">.</span><span class="n">prompt_tuning_config</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">,</span> \
<span class="s2">&quot;cannot accept both prompt_adapter_request and prompt_tuning_config in one request&quot;</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_load_prompt_adapter</span><span class="p">(</span><span class="n">request</span><span class="o">.</span><span class="n">prompt_adapter_request</span><span class="p">)</span>
<span class="n">uid</span> <span class="o">=</span> <span class="nb">str</span><span class="p">(</span><span class="n">request</span><span class="o">.</span><span class="n">prompt_adapter_request</span><span class="o">.</span><span class="n">adapter_id</span><span class="p">)</span>
<span class="n">prompt_tuning_config</span> <span class="o">=</span> <span class="n">tllm</span><span class="o">.</span><span class="n">PromptTuningConfig</span><span class="p">(</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_prompt_adapter_manager</span><span class="o">.</span><span class="n">uid_to_weights</span><span class="p">[</span><span class="n">uid</span><span class="p">])</span>
<span class="n">vocab_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_runtime_model_config</span><span class="o">.</span><span class="n">vocab_size</span>
<span class="n">pa_length</span> <span class="o">=</span> <span class="n">prompt_tuning_config</span><span class="o">.</span><span class="n">embedding_table</span><span class="o">.</span><span class="n">size</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span>
<span class="n">prompt_token_ids</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="nb">range</span><span class="p">(</span>
<span class="n">vocab_size</span><span class="p">,</span> <span class="n">vocab_size</span> <span class="o">+</span> <span class="n">pa_length</span><span class="p">))</span> <span class="o">+</span> <span class="n">prompt_token_ids</span>
<span class="k">elif</span> <span class="n">request</span><span class="o">.</span><span class="n">prompt_tuning_config</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">prompt_tuning_config</span> <span class="o">=</span> <span class="n">tllm</span><span class="o">.</span><span class="n">PromptTuningConfig</span><span class="p">(</span>
<span class="n">request</span><span class="o">.</span><span class="n">prompt_tuning_config</span><span class="p">[</span><span class="mi">0</span><span class="p">])</span>
<span class="k">assert</span> <span class="n">request</span><span class="o">.</span><span class="n">id</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
<span class="k">try</span><span class="p">:</span>
<span class="n">executor_request</span> <span class="o">=</span> <span class="n">tllm</span><span class="o">.</span><span class="n">Request</span><span class="p">(</span>
<span class="n">client_id</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">id</span><span class="p">,</span>
<span class="n">input_token_ids</span><span class="o">=</span><span class="n">prompt_token_ids</span><span class="p">,</span>
<span class="n">max_tokens</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">max_tokens</span><span class="p">,</span>
<span class="n">max_new_tokens</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">max_new_tokens</span><span class="p">,</span>
<span class="n">streaming</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">streaming</span><span class="p">,</span>
<span class="n">sampling_config</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">_get_sampling_config</span><span class="p">(),</span>
<span class="n">end_id</span><span class="o">=-</span><span class="mi">1</span> <span class="k">if</span> <span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">ignore_eos</span> <span class="k">else</span>
<span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">end_id</span><span class="p">,</span>
<span class="n">pad_id</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">pad_id</span><span class="p">,</span>
<span class="n">output_config</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">_get_output_config</span><span class="p">(),</span>
<span class="n">lookahead_config</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">lookahead_config</span><span class="p">,</span>
<span class="n">guided_decoding_params</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span>
<span class="n">_get_guided_decoding_params</span><span class="p">(),</span>
<span class="n">bad_words</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">_get_bad_words</span><span class="p">(),</span>
<span class="n">stop_words</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">_get_stop_words</span><span class="p">(),</span>
<span class="n">embedding_bias</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">embedding_bias</span><span class="p">,</span>
<span class="n">external_draft_tokens_config</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span>
<span class="n">external_draft_tokens_config</span><span class="p">,</span>
<span class="n">lora_config</span><span class="o">=</span><span class="n">lora_config</span><span class="p">,</span>
<span class="n">prompt_tuning_config</span><span class="o">=</span><span class="n">prompt_tuning_config</span><span class="p">,</span>
<span class="n">logits_post_processor_name</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span>
<span class="n">logits_post_processor_name</span><span class="p">,</span>
<span class="p">)</span>
<span class="k">if</span> <span class="n">request</span><span class="o">.</span><span class="n">query_token_ids</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="c1"># pytorch star attention workflow</span>
<span class="c1"># a workaround to avoid public interface update</span>
<span class="n">req_id</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">enqueue_request</span><span class="p">(</span><span class="n">executor_request</span><span class="p">,</span>
<span class="n">request</span><span class="o">.</span><span class="n">query_token_ids</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">req_id</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">enqueue_request</span><span class="p">(</span><span class="n">executor_request</span><span class="p">)</span>
<span class="k">return</span> <span class="n">req_id</span>
<span class="k">except</span> <span class="ne">Exception</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
<span class="k">raise</span> <span class="n">RequestError</span><span class="p">(</span><span class="nb">str</span><span class="p">(</span><span class="n">e</span><span class="p">))</span>
<span class="k">def</span> <span class="nf">submit</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">request</span><span class="p">:</span> <span class="n">GenerationRequest</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="n">GenerationResult</span><span class="p">:</span>
<span class="w"> </span><span class="sd">&quot;&quot;&quot; Low-level API to the executor. Return a &quot;future&quot; GenerationResult which can be waited. &quot;&quot;&quot;</span>
<span class="bp">self</span><span class="o">.</span><span class="n">start</span><span class="p">()</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">rank</span> <span class="o">!=</span> <span class="mi">0</span><span class="p">:</span>
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span>
<span class="s2">&quot;Only rank 0 can submit requests.</span><span class="se">\n</span><span class="s2">&quot;</span>
<span class="s2">&quot;To fix this, ensure that the llm.generate(...) method is &quot;</span>
<span class="s2">&quot;guarded with the `if __name__ == &#39;__main__&#39;:` block.&quot;</span><span class="p">)</span>
<span class="n">client_id</span> <span class="o">=</span> <span class="n">request</span><span class="o">.</span><span class="n">id</span> <span class="k">if</span> <span class="n">request</span><span class="o">.</span><span class="n">id</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="k">else</span> <span class="bp">self</span><span class="o">.</span><span class="n">_get_next_client_id</span><span class="p">(</span>
<span class="p">)</span>
<span class="k">if</span> <span class="n">request</span><span class="o">.</span><span class="n">id</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">request</span><span class="o">.</span><span class="n">set_id</span><span class="p">(</span><span class="n">client_id</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_enqueue_request</span><span class="p">(</span><span class="n">request</span><span class="p">)</span>
<span class="n">result</span> <span class="o">=</span> <span class="n">GenerationResult</span><span class="p">(</span>
<span class="n">request</span><span class="p">,</span> <span class="n">background_error_handler</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_handle_background_error</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="p">[</span><span class="n">client_id</span><span class="p">]</span> <span class="o">=</span> <span class="n">result</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_handle_background_error</span><span class="p">()</span>
<span class="k">return</span> <span class="n">result</span>
<span class="k">def</span> <span class="nf">shutdown</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">if</span> <span class="n">enable_llm_debug</span><span class="p">():</span>
<span class="k">try</span><span class="p">:</span>
<span class="n">print_colored</span><span class="p">(</span><span class="s1">&#39;Proxy.shutdown...</span><span class="se">\n</span><span class="s1">&#39;</span><span class="p">,</span> <span class="s2">&quot;yellow&quot;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="n">traceback</span><span class="o">.</span><span class="n">extract_stack</span><span class="p">())</span>
<span class="k">except</span> <span class="ne">ValueError</span><span class="p">:</span>
<span class="k">pass</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">doing_shutdown</span><span class="p">:</span>
<span class="k">return</span>
<span class="k">else</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">doing_shutdown</span> <span class="o">=</span> <span class="kc">True</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">engine</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">can_enqueue_requests</span><span class="p">():</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">await_response_thread</span><span class="o">.</span><span class="n">is_alive</span><span class="p">():</span>
<span class="bp">self</span><span class="o">.</span><span class="n">await_response_thread</span><span class="o">.</span><span class="n">stop</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">await_response_thread</span><span class="o">.</span><span class="n">join</span><span class="p">()</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span><span class="o">.</span><span class="n">is_alive</span><span class="p">():</span>
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span><span class="o">.</span><span class="n">stop</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span><span class="o">.</span><span class="n">join</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">engine</span> <span class="o">=</span> <span class="kc">None</span>
<span class="c1"># Check if there are any errors from the threads before shutdown.</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_handle_background_error</span><span class="p">()</span>
<span class="k">def</span> <span class="nf">block_subordinates</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">rank</span> <span class="o">!=</span> <span class="mi">0</span><span class="p">:</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="p">,</span> <span class="n">tllm</span><span class="o">.</span><span class="n">Executor</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
<span class="k">raise</span> <span class="bp">self</span><span class="o">.</span><span class="n">WorkerExit</span><span class="p">(</span>
<span class="s2">&quot;block_subordinates() should be used in a `with ExecutorBindingsWorker() as ...:` block&quot;</span>
<span class="p">)</span>
<span class="kn">from</span> <span class="nn">tensorrt_llm._torch.pyexecutor.py_executor</span> <span class="kn">import</span> <span class="n">PyExecutor</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="p">,</span> <span class="n">PyExecutor</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">wait_shutdown</span><span class="p">()</span>
<span class="k">def</span> <span class="fm">__enter__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span>
<span class="k">def</span> <span class="fm">__exit__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">exc_type</span><span class="p">,</span> <span class="n">exc_value</span><span class="p">,</span> <span class="n">traceback</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">bool</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
<span class="k">return</span> <span class="n">exc_type</span> <span class="ow">is</span> <span class="kc">None</span> <span class="ow">or</span> <span class="n">exc_type</span> <span class="o">==</span> <span class="n">ExecutorBindingsWorker</span><span class="o">.</span><span class="n">WorkerExit</span>
<span class="k">def</span> <span class="fm">__del__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
<span class="k">class</span> <span class="nc">ZeroMqQueue</span><span class="p">:</span>
<span class="w"> </span><span class="sd">&#39;&#39;&#39; A Queue-like container for IPC using ZeroMQ. &#39;&#39;&#39;</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">address</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span> <span class="o">*</span><span class="p">,</span> <span class="n">is_server</span><span class="p">:</span> <span class="nb">bool</span><span class="p">):</span>
<span class="w"> </span><span class="sd">&#39;&#39;&#39;</span>
<span class="sd"> Parameters:</span>
<span class="sd"> address (Tuple[str, str], optional): The address (tcp-ip_port, authkey) for the IPC. Defaults to None.</span>
<span class="sd"> is_server (bool): Whether the current process is the server or the client.</span>
<span class="sd"> &#39;&#39;&#39;</span>
<span class="bp">self</span><span class="o">.</span><span class="n">address</span> <span class="o">=</span> <span class="n">address</span> <span class="ow">or</span> <span class="s2">&quot;tcp://127.0.0.1:*&quot;</span>
<span class="bp">self</span><span class="o">.</span><span class="n">is_server</span> <span class="o">=</span> <span class="n">is_server</span>
<span class="bp">self</span><span class="o">.</span><span class="n">context</span> <span class="o">=</span> <span class="n">zmq</span><span class="o">.</span><span class="n">Context</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">poller</span> <span class="o">=</span> <span class="kc">None</span>
<span class="bp">self</span><span class="o">.</span><span class="n">socket</span> <span class="o">=</span> <span class="kc">None</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_setup_done</span> <span class="o">=</span> <span class="kc">False</span>
<span class="bp">self</span><span class="o">.</span><span class="n">socket</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">context</span><span class="o">.</span><span class="n">socket</span><span class="p">(</span><span class="n">zmq</span><span class="o">.</span><span class="n">PAIR</span><span class="p">)</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">is_server</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">socket</span><span class="o">.</span><span class="n">bind</span><span class="p">(</span>
<span class="bp">self</span><span class="o">.</span><span class="n">address</span>
<span class="p">)</span> <span class="c1"># Binds to the address and occupy a port immediately</span>
<span class="bp">self</span><span class="o">.</span><span class="n">address</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">socket</span><span class="o">.</span><span class="n">getsockopt</span><span class="p">(</span><span class="n">zmq</span><span class="o">.</span><span class="n">LAST_ENDPOINT</span><span class="p">)</span><span class="o">.</span><span class="n">decode</span><span class="p">()</span>
<span class="k">def</span> <span class="nf">setup_lazily</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_setup_done</span><span class="p">:</span>
<span class="k">return</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_setup_done</span> <span class="o">=</span> <span class="kc">True</span>
<span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">is_server</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">socket</span><span class="o">.</span><span class="n">connect</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">address</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">poller</span> <span class="o">=</span> <span class="n">zmq</span><span class="o">.</span><span class="n">Poller</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">poller</span><span class="o">.</span><span class="n">register</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">socket</span><span class="p">,</span> <span class="n">zmq</span><span class="o">.</span><span class="n">POLLIN</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">poll</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">timeout</span><span class="p">:</span> <span class="nb">int</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">bool</span><span class="p">:</span>
<span class="w"> </span><span class="sd">&quot;&quot;&quot;</span>
<span class="sd"> Parameters:</span>
<span class="sd"> timeout (int): Timeout in seconds</span>
<span class="sd"> &quot;&quot;&quot;</span>
<span class="bp">self</span><span class="o">.</span><span class="n">setup_lazily</span><span class="p">()</span>
<span class="n">events</span> <span class="o">=</span> <span class="nb">dict</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">poller</span><span class="o">.</span><span class="n">poll</span><span class="p">(</span><span class="n">timeout</span><span class="o">=</span><span class="n">timeout</span> <span class="o">*</span> <span class="mi">1000</span><span class="p">))</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">socket</span> <span class="ow">in</span> <span class="n">events</span> <span class="ow">and</span> <span class="n">events</span><span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">socket</span><span class="p">]</span> <span class="o">==</span> <span class="n">zmq</span><span class="o">.</span><span class="n">POLLIN</span><span class="p">:</span>
<span class="k">return</span> <span class="kc">True</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">return</span> <span class="kc">False</span>
<span class="k">def</span> <span class="nf">put</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">obj</span><span class="p">:</span> <span class="n">Any</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">setup_lazily</span><span class="p">()</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">obj</span><span class="p">,</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">):</span>
<span class="n">tensors</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_store_tensors_in_shmm</span><span class="p">(</span><span class="n">obj</span><span class="o">.</span><span class="n">tensors</span><span class="p">)</span>
<span class="n">obj</span> <span class="o">=</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">(</span><span class="n">client_id</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">client_id</span><span class="p">,</span>
<span class="n">tensors</span><span class="o">=</span><span class="n">tensors</span><span class="p">,</span>
<span class="n">finish_reasons</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">finish_reasons</span><span class="p">,</span>
<span class="n">is_final</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">is_final</span><span class="p">,</span>
<span class="n">error</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">error</span><span class="p">)</span>
<span class="n">message</span> <span class="o">=</span> <span class="n">pickle</span><span class="o">.</span><span class="n">dumps</span><span class="p">(</span><span class="n">obj</span><span class="p">)</span> <span class="c1"># nosec B301</span>
<span class="bp">self</span><span class="o">.</span><span class="n">socket</span><span class="o">.</span><span class="n">send</span><span class="p">(</span><span class="n">message</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">get</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="n">Any</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">setup_lazily</span><span class="p">()</span>
<span class="n">message</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">socket</span><span class="o">.</span><span class="n">recv</span><span class="p">()</span>
<span class="n">obj</span> <span class="o">=</span> <span class="n">pickle</span><span class="o">.</span><span class="n">loads</span><span class="p">(</span><span class="n">message</span><span class="p">)</span> <span class="c1"># nosec B301</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">obj</span><span class="p">,</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">):</span>
<span class="n">tensors</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_load_tensors_from_shmm</span><span class="p">(</span><span class="n">obj</span><span class="o">.</span><span class="n">tensors</span><span class="p">)</span>
<span class="n">obj</span> <span class="o">=</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">(</span><span class="n">client_id</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">client_id</span><span class="p">,</span>
<span class="n">tensors</span><span class="o">=</span><span class="n">tensors</span><span class="p">,</span>
<span class="n">finish_reasons</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">finish_reasons</span><span class="p">,</span>
<span class="n">is_final</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">is_final</span><span class="p">,</span>
<span class="n">error</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">error</span><span class="p">)</span>
<span class="k">return</span> <span class="n">obj</span>
<span class="k">def</span> <span class="nf">close</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">socket</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">socket</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">socket</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">context</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">context</span><span class="o">.</span><span class="n">term</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">context</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">def</span> <span class="nf">_store_tensors_in_shmm</span><span class="p">(</span>
<span class="bp">self</span><span class="p">,</span> <span class="n">tensors</span><span class="p">:</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">ResponseTensors</span>
<span class="p">)</span> <span class="o">-&gt;</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">ResponseTensors</span><span class="p">:</span>
<span class="k">if</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="c1"># The tensors are huge and cannot be transferred through socket directly. We need to store them in shared memory,</span>
<span class="c1"># and replace the tensors with the shared memory path.</span>
<span class="k">def</span> <span class="nf">store_tensor</span><span class="p">(</span><span class="n">tensor</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">])</span> <span class="o">-&gt;</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]:</span>
<span class="k">if</span> <span class="n">tensor</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="k">return</span> <span class="kc">None</span>
<span class="c1"># NOTE: We create random shmm here rather than two specific shmm for context and generation logit, since the</span>
<span class="c1"># shmm may not be read timely by the IpcQueue.get() in the other side, so there might be multiple alive shmm</span>
<span class="c1"># for logits.</span>
<span class="c1"># A known issue: the shmm instance may leak if the IpcQueue.get() thread is stopped before the IpcQueue.put()</span>
<span class="c1"># thread. This is not a big issue since the shmm will be automatically cleaned up when the process exits.</span>
<span class="n">shm</span> <span class="o">=</span> <span class="n">SharedMemory</span><span class="p">(</span><span class="n">create</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">size</span><span class="o">=</span><span class="n">tensor</span><span class="o">.</span><span class="n">nbytes</span> <span class="o">+</span> <span class="mi">2048</span><span class="p">)</span>
<span class="n">torch</span><span class="o">.</span><span class="n">save</span><span class="p">(</span><span class="n">tensor</span><span class="p">,</span> <span class="n">shm</span><span class="o">.</span><span class="n">_mmap</span><span class="p">)</span>
<span class="n">shm</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
<span class="k">return</span> <span class="n">shm</span><span class="o">.</span><span class="n">name</span>
<span class="k">return</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">ResponseTensors</span><span class="p">(</span>
<span class="n">output_token_ids</span><span class="o">=</span><span class="n">tensors</span><span class="o">.</span><span class="n">output_token_ids</span><span class="p">,</span>
<span class="n">context_logits</span><span class="o">=</span><span class="n">store_tensor</span><span class="p">(</span><span class="n">tensors</span><span class="o">.</span><span class="n">context_logits</span><span class="p">),</span>
<span class="n">generation_logits</span><span class="o">=</span><span class="n">store_tensor</span><span class="p">(</span><span class="n">tensors</span><span class="o">.</span><span class="n">generation_logits</span><span class="p">),</span>
<span class="n">log_probs</span><span class="o">=</span><span class="n">tensors</span><span class="o">.</span><span class="n">log_probs</span><span class="p">,</span>
<span class="n">cum_log_probs</span><span class="o">=</span><span class="n">tensors</span><span class="o">.</span><span class="n">cum_log_probs</span><span class="p">,</span>
<span class="p">)</span>
<span class="k">def</span> <span class="nf">_load_tensors_from_shmm</span><span class="p">(</span>
<span class="bp">self</span><span class="p">,</span> <span class="n">tensors</span><span class="p">:</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">ResponseTensors</span>
<span class="p">)</span> <span class="o">-&gt;</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">ResponseTensors</span><span class="p">:</span>
<span class="k">if</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">def</span> <span class="nf">load_tensor</span><span class="p">(</span><span class="n">tensor</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">])</span> <span class="o">-&gt;</span> <span class="n">Optional</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">]:</span>
<span class="k">if</span> <span class="n">tensor</span> <span class="ow">is</span> <span class="kc">None</span> <span class="ow">or</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">tensor</span><span class="p">,</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">):</span>
<span class="k">return</span> <span class="n">tensor</span>
<span class="n">shm</span> <span class="o">=</span> <span class="n">SharedMemory</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="n">tensor</span><span class="p">,</span> <span class="n">create</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
<span class="n">tensor</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="n">io</span><span class="o">.</span><span class="n">BytesIO</span><span class="p">(</span><span class="n">shm</span><span class="o">.</span><span class="n">buf</span><span class="p">))</span>
<span class="n">shm</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
<span class="n">shm</span><span class="o">.</span><span class="n">unlink</span><span class="p">()</span>
<span class="k">return</span> <span class="n">tensor</span>
<span class="k">return</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">ResponseTensors</span><span class="p">(</span>
<span class="n">output_token_ids</span><span class="o">=</span><span class="n">tensors</span><span class="o">.</span><span class="n">output_token_ids</span><span class="p">,</span>
<span class="n">context_logits</span><span class="o">=</span><span class="n">load_tensor</span><span class="p">(</span><span class="n">tensors</span><span class="o">.</span><span class="n">context_logits</span><span class="p">),</span>
<span class="n">generation_logits</span><span class="o">=</span><span class="n">load_tensor</span><span class="p">(</span><span class="n">tensors</span><span class="o">.</span><span class="n">generation_logits</span><span class="p">),</span>
<span class="n">log_probs</span><span class="o">=</span><span class="n">tensors</span><span class="o">.</span><span class="n">log_probs</span><span class="p">,</span>
<span class="n">cum_log_probs</span><span class="o">=</span><span class="n">tensors</span><span class="o">.</span><span class="n">cum_log_probs</span><span class="p">,</span>
<span class="p">)</span>
<span class="k">def</span> <span class="fm">__del__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
<span class="n">IpcQueue</span> <span class="o">=</span> <span class="n">ZeroMqQueue</span>
<span class="k">class</span> <span class="nc">FusedIpcQueue</span><span class="p">:</span>
<span class="w"> </span><span class="sd">&#39;&#39;&#39; A Queue-like container for IPC with optional message batched. &#39;&#39;&#39;</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span>
<span class="n">address</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="o">*</span><span class="p">,</span>
<span class="n">is_server</span><span class="p">:</span> <span class="nb">bool</span><span class="p">,</span>
<span class="n">fuse_message</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">fuse_size</span><span class="o">=</span><span class="mi">100000</span><span class="p">,</span>
<span class="n">error_queue</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">queue_cls</span><span class="o">=</span><span class="n">ZeroMqQueue</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">queue</span> <span class="o">=</span> <span class="n">queue_cls</span><span class="p">(</span><span class="n">address</span><span class="o">=</span><span class="n">address</span><span class="p">,</span> <span class="n">is_server</span><span class="o">=</span><span class="n">is_server</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">fuse_message</span> <span class="o">=</span> <span class="n">fuse_message</span>
<span class="bp">self</span><span class="o">.</span><span class="n">error_queue</span> <span class="o">=</span> <span class="n">error_queue</span>
<span class="bp">self</span><span class="o">.</span><span class="n">fuse_size</span> <span class="o">=</span> <span class="n">fuse_size</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_message_counter</span> <span class="o">=</span> <span class="mi">0</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_obj_counter</span> <span class="o">=</span> <span class="mi">0</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_send_thread</span> <span class="o">=</span> <span class="kc">None</span>
<span class="bp">self</span><span class="o">.</span><span class="n">sending_queue</span> <span class="o">=</span> <span class="n">Queue</span><span class="p">()</span> <span class="k">if</span> <span class="n">fuse_message</span> <span class="k">else</span> <span class="kc">None</span>
<span class="k">def</span> <span class="nf">setup_sender</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">fuse_message</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">_send_thread</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="k">return</span>
<span class="k">def</span> <span class="nf">send_task</span><span class="p">():</span>
<span class="k">while</span> <span class="kc">True</span><span class="p">:</span>
<span class="n">qsize</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">sending_queue</span><span class="o">.</span><span class="n">qsize</span><span class="p">()</span>
<span class="k">if</span> <span class="n">qsize</span> <span class="o">&gt;</span> <span class="mi">0</span><span class="p">:</span>
<span class="n">qsize</span> <span class="o">=</span> <span class="nb">min</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">fuse_size</span><span class="p">,</span> <span class="n">qsize</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_obj_counter</span> <span class="o">+=</span> <span class="n">qsize</span>
<span class="n">message</span> <span class="o">=</span> <span class="p">[</span>
<span class="bp">self</span><span class="o">.</span><span class="n">sending_queue</span><span class="o">.</span><span class="n">get_nowait</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="n">qsize</span><span class="p">)</span>
<span class="p">]</span>
<span class="bp">self</span><span class="o">.</span><span class="n">queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">message</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_message_counter</span> <span class="o">+=</span> <span class="mi">1</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">time</span><span class="o">.</span><span class="n">sleep</span><span class="p">(</span><span class="mf">0.001</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_send_thread</span> <span class="o">=</span> <span class="n">ManagedThread</span><span class="p">(</span><span class="n">send_task</span><span class="p">,</span>
<span class="n">name</span><span class="o">=</span><span class="s2">&quot;fused_send_thread&quot;</span><span class="p">,</span>
<span class="n">error_queue</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">error_queue</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_send_thread</span><span class="o">.</span><span class="n">start</span><span class="p">()</span>
<span class="k">def</span> <span class="nf">put</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">obj</span><span class="p">:</span> <span class="n">Any</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">setup_sender</span><span class="p">()</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">fuse_message</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">sending_queue</span><span class="o">.</span><span class="n">put_nowait</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_prepare_message</span><span class="p">(</span><span class="n">obj</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">queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_prepare_message</span><span class="p">(</span><span class="n">obj</span><span class="p">))</span>
<span class="k">def</span> <span class="nf">get</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="n">Any</span><span class="p">:</span>
<span class="n">obj</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">obj</span><span class="p">,</span> <span class="nb">list</span><span class="p">):</span>
<span class="k">return</span> <span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">_process_message</span><span class="p">(</span><span class="n">o</span><span class="p">)</span> <span class="k">for</span> <span class="n">o</span> <span class="ow">in</span> <span class="n">obj</span><span class="p">]</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_process_message</span><span class="p">(</span><span class="n">obj</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">_prepare_message</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">obj</span><span class="p">:</span> <span class="n">Any</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="n">Any</span><span class="p">:</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">obj</span><span class="p">,</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">):</span>
<span class="n">tensors</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">queue</span><span class="o">.</span><span class="n">_store_tensors_in_shmm</span><span class="p">(</span><span class="n">obj</span><span class="o">.</span><span class="n">tensors</span><span class="p">)</span>
<span class="k">return</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">(</span>
<span class="n">client_id</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">client_id</span><span class="p">,</span>
<span class="n">tensors</span><span class="o">=</span><span class="n">tensors</span><span class="p">,</span>
<span class="n">finish_reasons</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">finish_reasons</span><span class="p">,</span>
<span class="n">is_final</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">is_final</span><span class="p">,</span>
<span class="n">sequence_index</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">sequence_index</span><span class="p">,</span>
<span class="n">error</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">error</span><span class="p">)</span>
<span class="k">return</span> <span class="n">obj</span>
<span class="k">def</span> <span class="nf">_process_message</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">obj</span><span class="p">:</span> <span class="n">Any</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="n">Any</span><span class="p">:</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">obj</span><span class="p">,</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">):</span>
<span class="n">tensors</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">queue</span><span class="o">.</span><span class="n">_load_tensors_from_shmm</span><span class="p">(</span><span class="n">obj</span><span class="o">.</span><span class="n">tensors</span><span class="p">)</span>
<span class="k">return</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">(</span>
<span class="n">client_id</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">client_id</span><span class="p">,</span>
<span class="n">tensors</span><span class="o">=</span><span class="n">tensors</span><span class="p">,</span>
<span class="n">finish_reasons</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">finish_reasons</span><span class="p">,</span>
<span class="n">is_final</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">is_final</span><span class="p">,</span>
<span class="n">sequence_index</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">sequence_index</span><span class="p">,</span>
<span class="n">error</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">error</span><span class="p">)</span>
<span class="k">return</span> <span class="n">obj</span>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">address</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="n">Tuple</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">int</span><span class="p">,</span> <span class="nb">bytes</span><span class="p">]:</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">queue</span><span class="o">.</span><span class="n">address</span>
<span class="k">def</span> <span class="fm">__del__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
<span class="k">def</span> <span class="nf">print_fuse_stats</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_message_counter</span> <span class="o">&gt;</span> <span class="mi">0</span><span class="p">:</span>
<span class="n">print_colored</span><span class="p">(</span>
<span class="sa">f</span><span class="s2">&quot;IPCQueue: </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">_message_counter</span><span class="si">}</span><span class="s2"> messages, </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">_obj_counter</span><span class="si">}</span><span class="s2"> objects sent, average: </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">_obj_counter</span><span class="o">/</span><span class="bp">self</span><span class="o">.</span><span class="n">_message_counter</span><span class="si">}</span><span class="s2">.</span><span class="se">\n</span><span class="s2">&quot;</span><span class="p">,</span>
<span class="s2">&quot;green&quot;</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">close</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">queue</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_send_thread</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_send_thread</span><span class="o">.</span><span class="n">stop</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_send_thread</span><span class="o">.</span><span class="n">join</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_send_thread</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="n">enable_llm_debug</span><span class="p">():</span>
<span class="bp">self</span><span class="o">.</span><span class="n">print_fuse_stats</span><span class="p">()</span>
<span class="k">class</span> <span class="nc">ExecutorBindingsProxy</span><span class="p">(</span><span class="n">GenerationExecutor</span><span class="p">):</span>
<span class="n">READY_SIGNAL</span> <span class="o">=</span> <span class="sa">b</span><span class="s2">&quot;READY&quot;</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span>
<span class="n">workers_kwargs</span><span class="p">:</span> <span class="nb">dict</span><span class="p">,</span>
<span class="n">model_world_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">1</span><span class="p">,</span>
<span class="n">mpi_session</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">MpiSession</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="o">*</span><span class="p">,</span>
<span class="n">worker_cls</span><span class="p">:</span> <span class="nb">type</span> <span class="o">=</span> <span class="n">ExecutorBindingsWorker</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="kc">None</span><span class="p">:</span>
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">workers_started</span> <span class="o">=</span> <span class="kc">False</span>
<span class="bp">self</span><span class="o">.</span><span class="n">worker_cls</span> <span class="o">=</span> <span class="n">worker_cls</span>
<span class="bp">self</span><span class="o">.</span><span class="n">request_queue</span> <span class="o">=</span> <span class="n">IpcQueue</span><span class="p">(</span><span class="n">is_server</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">request_error_queue</span> <span class="o">=</span> <span class="n">IpcQueue</span><span class="p">(</span><span class="n">is_server</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="c1"># TODO(zhenhuan, chunwei): keep only batch branch after more tests</span>
<span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span> <span class="o">=</span> <span class="n">FusedIpcQueue</span><span class="p">(</span><span class="n">is_server</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
<span class="n">fuse_message</span><span class="o">=</span><span class="ow">not</span> <span class="n">BATCH_RESP_IN_AWAIT</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">mp_stats_queue</span> <span class="o">=</span> <span class="n">FusedIpcQueue</span><span class="p">(</span>
<span class="n">is_server</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">fuse_message</span><span class="o">=</span><span class="ow">not</span> <span class="n">BATCH_RESP_IN_AWAIT</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">int</span><span class="p">,</span> <span class="n">GenerationResult</span><span class="p">]</span> <span class="o">=</span> <span class="p">{}</span>
<span class="k">if</span> <span class="n">mpi_session</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span> <span class="o">=</span> <span class="n">MpiPoolSession</span><span class="p">(</span><span class="n">n_workers</span><span class="o">=</span><span class="n">model_world_size</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">mpi_session</span> <span class="o">=</span> <span class="n">mpi_session</span>
<span class="bp">self</span><span class="o">.</span><span class="n">model_world_size</span> <span class="o">=</span> <span class="n">model_world_size</span>
<span class="bp">self</span><span class="o">.</span><span class="n">workers_kwargs</span> <span class="o">=</span> <span class="nb">dict</span><span class="p">(</span>
<span class="o">**</span><span class="n">workers_kwargs</span><span class="p">,</span>
<span class="n">request_queue_addr</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">request_queue</span><span class="o">.</span><span class="n">address</span><span class="p">,</span>
<span class="n">request_error_queue_addr</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">request_error_queue</span><span class="o">.</span><span class="n">address</span><span class="p">,</span>
<span class="n">result_queue_addr</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span><span class="o">.</span><span class="n">address</span><span class="p">,</span>
<span class="n">stats_queue_addr</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">mp_stats_queue</span><span class="o">.</span><span class="n">address</span><span class="p">,</span>
<span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_result_thread</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">ManagedThread</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">ManagedThread</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_start_executor_workers</span><span class="p">()</span>
<span class="nd">@staticmethod</span>
<span class="k">def</span> <span class="nf">workers_main</span><span class="p">(</span>
<span class="n">engine</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">Path</span><span class="p">,</span> <span class="n">Engine</span><span class="p">],</span>
<span class="n">request_queue_addr</span><span class="p">:</span> <span class="n">Tuple</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">int</span><span class="p">,</span> <span class="nb">bytes</span><span class="p">],</span>
<span class="n">request_error_queue_addr</span><span class="p">:</span> <span class="n">Tuple</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">int</span><span class="p">,</span> <span class="nb">bytes</span><span class="p">],</span>
<span class="n">result_queue_addr</span><span class="p">:</span> <span class="n">Tuple</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">int</span><span class="p">,</span> <span class="nb">bytes</span><span class="p">],</span>
<span class="n">stats_queue_addr</span><span class="p">:</span> <span class="n">Tuple</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">int</span><span class="p">,</span> <span class="nb">bytes</span><span class="p">],</span>
<span class="n">executor_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">tllm</span><span class="o">.</span><span class="n">ExecutorConfig</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">logits_post_processor_map</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Callable</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">worker_cls</span><span class="p">:</span> <span class="nb">type</span> <span class="o">=</span> <span class="n">ExecutorBindingsWorker</span><span class="p">,</span>
<span class="n">tracer_init_kwargs</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">dict</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">_torch_model_class_mapping</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">dict</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="p">)</span> <span class="o">-&gt;</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">result_queue</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="n">tracer_init_kwargs</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="n">mpi_rank</span><span class="p">()</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
<span class="n">tracer</span> <span class="o">=</span> <span class="n">VizTracer</span><span class="p">(</span><span class="o">**</span><span class="n">tracer_init_kwargs</span><span class="p">)</span>
<span class="n">tracer</span><span class="o">.</span><span class="n">register_exit</span><span class="p">()</span>
<span class="n">tracer</span><span class="o">.</span><span class="n">start</span><span class="p">()</span>
<span class="n">set_global_tracer</span><span class="p">(</span><span class="n">tracer</span><span class="p">)</span>
<span class="k">if</span> <span class="n">_torch_model_class_mapping</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="kn">from</span> <span class="nn">tensorrt_llm._torch.models.modeling_auto</span> <span class="kn">import</span> \
<span class="n">MODEL_CLASS_MAPPING</span>
<span class="n">MODEL_CLASS_MAPPING</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="o">**</span><span class="n">_torch_model_class_mapping</span><span class="p">)</span>
<span class="k">if</span> <span class="n">mpi_rank</span><span class="p">()</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
<span class="n">request_queue</span> <span class="o">=</span> <span class="n">IpcQueue</span><span class="p">(</span><span class="n">request_queue_addr</span><span class="p">,</span> <span class="n">is_server</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
<span class="n">request_error_queue</span> <span class="o">=</span> <span class="n">IpcQueue</span><span class="p">(</span><span class="n">request_error_queue_addr</span><span class="p">,</span>
<span class="n">is_server</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
<span class="c1"># TODO(zhenhuan, chunwei): keep only batch branch after more tests</span>
<span class="n">result_queue</span> <span class="o">=</span> <span class="n">FusedIpcQueue</span><span class="p">(</span><span class="n">result_queue_addr</span><span class="p">,</span>
<span class="n">is_server</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">fuse_message</span><span class="o">=</span><span class="ow">not</span> <span class="n">BATCH_RESP_IN_AWAIT</span><span class="p">)</span>
<span class="n">mp_stats_queue</span> <span class="o">=</span> <span class="n">FusedIpcQueue</span><span class="p">(</span><span class="n">stats_queue_addr</span><span class="p">,</span>
<span class="n">is_server</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">fuse_message</span><span class="o">=</span><span class="ow">not</span> <span class="n">BATCH_RESP_IN_AWAIT</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">notify_proxy_threads_to_quit</span><span class="p">():</span>
<span class="c1"># Signal the dispatcher thread in the proxy to quit</span>
<span class="n">result_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="kc">None</span><span class="p">)</span>
<span class="c1"># Signal the stats thread in the proxy to quit</span>
<span class="n">mp_stats_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="kc">None</span><span class="p">)</span>
<span class="c1"># Error handling in the Worker/MPI process</span>
<span class="c1"># 1. During Executor initialization, the errors will be captured and send back via request_error_queue.</span>
<span class="c1"># 2. During execution, the errors will be captured by ManagedThreads</span>
<span class="c1"># a) For per-request error, the error will be send back via result_queue, and eventually raised in</span>
<span class="c1"># handle_response() in the main thread.</span>
<span class="c1"># b) For system error, the error will be raised in the MPI process and handled by future.done_callback,</span>
<span class="c1"># that will propagate the error to the error_queue in the main thread.</span>
<span class="k">try</span><span class="p">:</span>
<span class="n">executor</span> <span class="o">=</span> <span class="n">worker_cls</span><span class="p">(</span><span class="n">engine</span><span class="p">,</span> <span class="n">executor_config</span><span class="p">,</span>
<span class="n">logits_post_processor_map</span><span class="p">)</span>
<span class="k">except</span> <span class="ne">Exception</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
<span class="nb">print</span><span class="p">(</span><span class="n">traceback</span><span class="o">.</span><span class="n">format_exc</span><span class="p">())</span>
<span class="k">if</span> <span class="n">mpi_rank</span><span class="p">()</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
<span class="n">request_error_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">e</span><span class="p">)</span>
<span class="k">return</span>
<span class="k">with</span> <span class="n">executor</span><span class="p">:</span>
<span class="k">try</span><span class="p">:</span>
<span class="n">executor</span><span class="o">.</span><span class="n">block_subordinates</span><span class="p">()</span>
<span class="k">if</span> <span class="n">mpi_rank</span><span class="p">()</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
<span class="n">executor</span><span class="o">.</span><span class="n">set_result_queue</span><span class="p">(</span><span class="n">result_queue</span><span class="p">)</span>
<span class="n">executor</span><span class="o">.</span><span class="n">set_stats_queue</span><span class="p">(</span><span class="n">mp_stats_queue</span><span class="p">)</span>
<span class="n">request_error_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">ExecutorBindingsProxy</span><span class="o">.</span><span class="n">READY_SIGNAL</span><span class="p">)</span>
<span class="k">while</span> <span class="p">(</span><span class="n">req</span> <span class="o">:=</span> <span class="n">request_queue</span><span class="o">.</span><span class="n">get</span><span class="p">())</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="k">try</span><span class="p">:</span>
<span class="n">result</span> <span class="o">=</span> <span class="n">executor</span><span class="o">.</span><span class="n">submit</span><span class="p">(</span><span class="n">req</span><span class="p">)</span>
<span class="n">request_error_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="kc">None</span><span class="p">)</span> <span class="c1"># None means success</span>
<span class="k">except</span> <span class="n">RequestError</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
<span class="n">request_error_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">e</span><span class="p">)</span>
<span class="n">notify_proxy_threads_to_quit</span><span class="p">()</span>
<span class="k">except</span> <span class="n">ExecutorBindingsWorker</span><span class="o">.</span><span class="n">WorkerExit</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
<span class="c1"># This will capture by the with-statement and exit normally.</span>
<span class="k">raise</span> <span class="n">e</span>
<span class="k">except</span> <span class="ne">Exception</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span> <span class="c1"># other critical errors</span>
<span class="k">if</span> <span class="n">mpi_rank</span><span class="p">()</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
<span class="n">notify_proxy_threads_to_quit</span><span class="p">()</span>
<span class="n">err</span> <span class="o">=</span> <span class="n">CppExecutorError</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;Failed during generation: </span><span class="si">{</span><span class="n">e</span><span class="si">}</span><span class="s2">&quot;</span><span class="p">)</span>
<span class="k">if</span> <span class="n">mpi_rank</span><span class="p">()</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
<span class="n">request_error_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">err</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">dispatch_result_task</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">bool</span><span class="p">:</span>
<span class="k">if</span> <span class="p">(</span><span class="n">res</span> <span class="o">:=</span> <span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span><span class="o">.</span><span class="n">get</span><span class="p">())</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="k">return</span> <span class="kc">False</span> <span class="c1"># shutdown the thread</span>
<span class="n">async_queues</span> <span class="o">=</span> <span class="p">[]</span>
<span class="n">event_loop</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">def</span> <span class="nf">process_res</span><span class="p">(</span><span class="n">res</span><span class="p">):</span>
<span class="n">client_id</span> <span class="o">=</span> <span class="n">res</span><span class="o">.</span><span class="n">client_id</span>
<span class="k">nonlocal</span> <span class="n">event_loop</span>
<span class="k">nonlocal</span> <span class="n">async_queues</span>
<span class="n">queue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="p">[</span><span class="n">client_id</span><span class="p">]</span><span class="o">.</span><span class="n">queue</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">queue</span><span class="p">,</span> <span class="n">_SyncQueue</span><span class="p">):</span>
<span class="n">queue</span><span class="o">.</span><span class="n">put_nowait</span><span class="p">(</span><span class="n">res</span><span class="p">)</span>
<span class="n">async_queues</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">queue</span><span class="p">)</span>
<span class="c1"># all the loops are identical</span>
<span class="n">event_loop</span> <span class="o">=</span> <span class="n">event_loop</span> <span class="ow">or</span> <span class="n">queue</span><span class="o">.</span><span class="n">loop</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">res</span><span class="p">)</span>
<span class="k">if</span> <span class="n">res</span><span class="o">.</span><span class="n">is_final</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="o">.</span><span class="n">pop</span><span class="p">(</span><span class="n">client_id</span><span class="p">,</span> <span class="kc">None</span><span class="p">)</span>
<span class="n">res</span> <span class="o">=</span> <span class="n">res</span> <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">res</span><span class="p">,</span> <span class="nb">list</span><span class="p">)</span> <span class="k">else</span> <span class="p">[</span><span class="n">res</span><span class="p">]</span>
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="n">res</span><span class="p">:</span>
<span class="n">global_tracer</span><span class="p">()</span><span class="o">.</span><span class="n">log_instant</span><span class="p">(</span><span class="s2">&quot;IPC.get&quot;</span><span class="p">)</span>
<span class="k">if</span> <span class="n">i</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="k">return</span> <span class="kc">False</span>
<span class="n">process_res</span><span class="p">(</span><span class="n">i</span><span class="p">)</span>
<span class="k">if</span> <span class="n">async_queues</span><span class="p">:</span>
<span class="n">_SyncQueue</span><span class="o">.</span><span class="n">notify_many</span><span class="p">(</span><span class="n">event_loop</span><span class="p">,</span> <span class="n">async_queues</span><span class="p">)</span>
<span class="k">return</span> <span class="kc">True</span> <span class="c1"># success</span>
<span class="k">def</span> <span class="nf">dispatch_stats_task</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">bool</span><span class="p">:</span>
<span class="c1"># get-stats is not urgent, so we can sleep a bit</span>
<span class="n">time</span><span class="o">.</span><span class="n">sleep</span><span class="p">(</span><span class="mf">0.1</span><span class="p">)</span>
<span class="k">try</span><span class="p">:</span>
<span class="n">stats</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">mp_stats_queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
<span class="k">except</span><span class="p">:</span>
<span class="k">return</span> <span class="kc">False</span>
<span class="k">if</span> <span class="n">stats</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="k">return</span> <span class="kc">False</span>
<span class="n">stats</span> <span class="o">=</span> <span class="n">stats</span> <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">stats</span><span class="p">,</span> <span class="nb">list</span><span class="p">)</span> <span class="k">else</span> <span class="p">[</span><span class="n">stats</span><span class="p">]</span>
<span class="k">while</span> <span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span><span class="o">.</span><span class="n">full</span><span class="p">():</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
<span class="k">try</span><span class="p">:</span>
<span class="k">for</span> <span class="n">s</span> <span class="ow">in</span> <span class="n">stats</span><span class="p">:</span>
<span class="k">if</span> <span class="n">s</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="k">return</span> <span class="kc">False</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">s</span><span class="p">)</span>
<span class="k">except</span> <span class="n">AsyncQueue</span><span class="o">.</span><span class="n">EventLoopShutdownError</span><span class="p">:</span>
<span class="c1"># This happens in the last stats loop while the generate workflow is stopped.</span>
<span class="k">pass</span>
<span class="k">except</span> <span class="ne">Exception</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
<span class="k">raise</span> <span class="n">e</span>
<span class="k">return</span> <span class="kc">True</span> <span class="c1"># success</span>
<span class="k">def</span> <span class="nf">_start_dispatch_threads</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">dispatch_result_thread</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_result_thread</span> <span class="o">=</span> <span class="n">ManagedThread</span><span class="p">(</span>
<span class="n">weakref</span><span class="o">.</span><span class="n">WeakMethod</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">dispatch_result_task</span><span class="p">),</span>
<span class="n">error_queue</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="p">,</span>
<span class="n">name</span><span class="o">=</span><span class="s2">&quot;proxy_dispatch_result_thread&quot;</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span> <span class="o">=</span> <span class="n">ManagedThread</span><span class="p">(</span>
<span class="n">weakref</span><span class="o">.</span><span class="n">WeakMethod</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_task</span><span class="p">),</span>
<span class="n">error_queue</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="p">,</span>
<span class="n">name</span><span class="o">=</span><span class="s2">&quot;proxy_dispatch_stats_thread&quot;</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_result_thread</span><span class="o">.</span><span class="n">start</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">create_stats_queue</span><span class="p">()</span>
<span class="c1"># TODO: clean up the stats thread, and replace with a decent get_stats API</span>
<span class="c1">#self.dispatch_stats_thread.start()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_handle_background_error</span><span class="p">()</span>
<span class="k">def</span> <span class="nf">_start_executor_workers</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="n">self_ref</span> <span class="o">=</span> <span class="n">weakref</span><span class="o">.</span><span class="n">ref</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">mpi_done_callback</span><span class="p">(</span><span class="n">future</span><span class="p">:</span> <span class="n">concurrent</span><span class="o">.</span><span class="n">futures</span><span class="o">.</span><span class="n">Future</span><span class="p">):</span>
<span class="c1"># This is called when the MPI worker is done, so future.exception() will not block.</span>
<span class="k">if</span> <span class="n">future</span><span class="o">.</span><span class="n">exception</span><span class="p">()</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="k">if</span> <span class="n">self_</span> <span class="o">:=</span> <span class="n">self_ref</span><span class="p">():</span>
<span class="n">self_</span><span class="o">.</span><span class="n">_error_queue</span><span class="o">.</span><span class="n">put_nowait</span><span class="p">(</span><span class="n">future</span><span class="o">.</span><span class="n">exception</span><span class="p">())</span>
<span class="n">tracer_init_kwargs</span> <span class="o">=</span> <span class="n">get_tracer</span><span class="p">()</span><span class="o">.</span><span class="n">init_kwargs</span> <span class="k">if</span> <span class="n">enable_llm_tracer</span><span class="p">(</span>
<span class="p">)</span> <span class="k">else</span> <span class="kc">None</span>
<span class="kn">from</span> <span class="nn">tensorrt_llm._torch.models.modeling_auto</span> <span class="kn">import</span> <span class="n">MODEL_CLASS_MAPPING</span>
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_futures</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span><span class="o">.</span><span class="n">submit</span><span class="p">(</span>
<span class="n">ExecutorBindingsProxy</span><span class="o">.</span><span class="n">workers_main</span><span class="p">,</span>
<span class="o">**</span><span class="bp">self</span><span class="o">.</span><span class="n">workers_kwargs</span><span class="p">,</span>
<span class="n">worker_cls</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">worker_cls</span><span class="p">,</span>
<span class="n">tracer_init_kwargs</span><span class="o">=</span><span class="n">tracer_init_kwargs</span><span class="p">,</span>
<span class="n">_torch_model_class_mapping</span><span class="o">=</span><span class="n">MODEL_CLASS_MAPPING</span><span class="p">,</span>
<span class="p">)</span>
<span class="k">for</span> <span class="n">fut</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">mpi_futures</span><span class="p">:</span>
<span class="n">fut</span><span class="o">.</span><span class="n">add_done_callback</span><span class="p">(</span><span class="n">mpi_done_callback</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">workers_started</span> <span class="o">=</span> <span class="kc">True</span>
<span class="k">while</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">request_error_queue</span><span class="o">.</span><span class="n">poll</span><span class="p">(</span><span class="mi">1</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_handle_background_error</span><span class="p">()</span>
<span class="n">ready_signal</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">request_error_queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
<span class="k">if</span> <span class="n">ready_signal</span> <span class="o">!=</span> <span class="n">ExecutorBindingsProxy</span><span class="o">.</span><span class="n">READY_SIGNAL</span><span class="p">:</span>
<span class="k">raise</span> <span class="n">ready_signal</span>
<span class="k">def</span> <span class="nf">shutdown</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">if</span> <span class="n">enable_llm_debug</span><span class="p">():</span>
<span class="k">try</span><span class="p">:</span>
<span class="n">print_colored</span><span class="p">(</span><span class="s1">&#39;Proxy.shutdown...</span><span class="se">\n</span><span class="s1">&#39;</span><span class="p">,</span> <span class="s2">&quot;yellow&quot;</span><span class="p">)</span>
<span class="n">print_colored</span><span class="p">(</span><span class="nb">str</span><span class="p">(</span><span class="n">traceback</span><span class="o">.</span><span class="n">format_exc</span><span class="p">())</span> <span class="o">+</span> <span class="s2">&quot;</span><span class="se">\n</span><span class="s2">&quot;</span><span class="p">,</span> <span class="s2">&quot;yellow&quot;</span><span class="p">)</span>
<span class="k">except</span> <span class="ne">ValueError</span><span class="p">:</span>
<span class="k">pass</span>
<span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">workers_started</span><span class="p">:</span>
<span class="k">return</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">doing_shutdown</span><span class="p">:</span>
<span class="k">return</span>
<span class="k">else</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">doing_shutdown</span> <span class="o">=</span> <span class="kc">True</span>
<span class="c1"># step1: notify the workers to quit</span>
<span class="k">if</span> <span class="nb">all</span><span class="p">(</span><span class="ow">not</span> <span class="n">f</span><span class="o">.</span><span class="n">done</span><span class="p">()</span> <span class="k">for</span> <span class="n">f</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">mpi_futures</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">request_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="kc">None</span><span class="p">)</span>
<span class="k">for</span> <span class="n">f</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">mpi_futures</span><span class="p">:</span>
<span class="k">try</span><span class="p">:</span>
<span class="n">f</span><span class="o">.</span><span class="n">result</span><span class="p">()</span>
<span class="k">except</span><span class="p">:</span>
<span class="c1"># The errors are already captured in mpi_done_callback, ignored here</span>
<span class="k">pass</span>
<span class="c1"># step2: notify the background threads to quit</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">dispatch_result_thread</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">dispatch_result_thread</span><span class="o">.</span><span class="n">is_alive</span><span class="p">(</span>
<span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_result_thread</span><span class="o">.</span><span class="n">stop</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_result_thread</span><span class="o">.</span><span class="n">join</span><span class="p">()</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span><span class="o">.</span><span class="n">is_alive</span><span class="p">(</span>
<span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span><span class="o">.</span><span class="n">stop</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span><span class="o">.</span><span class="n">join</span><span class="p">()</span>
<span class="c1"># step3: finish all remaining work</span>
<span class="c1"># close all the sockets</span>
<span class="bp">self</span><span class="o">.</span><span class="n">request_queue</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">request_error_queue</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">mp_stats_queue</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">workers_started</span> <span class="o">=</span> <span class="kc">False</span>
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
<span class="c1"># Process the errors in-case error during shutting down the threads</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_handle_background_error</span><span class="p">()</span>
<span class="k">def</span> <span class="nf">submit</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">request</span><span class="p">:</span> <span class="n">GenerationRequest</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="n">GenerationResult</span><span class="p">:</span>
<span class="w"> </span><span class="sd">&quot;&quot;&quot;</span>
<span class="sd"> Low-level API to the executor. Return a &quot;future&quot; GenerationResult which can be waited.</span>
<span class="sd"> Forwards the request to the workers through the request queue.</span>
<span class="sd"> &quot;&quot;&quot;</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_start_dispatch_threads</span><span class="p">()</span>
<span class="n">request</span><span class="o">.</span><span class="n">set_id</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_get_next_client_id</span><span class="p">())</span>
<span class="n">result</span> <span class="o">=</span> <span class="n">GenerationResult</span><span class="p">(</span>
<span class="n">request</span><span class="p">,</span> <span class="n">background_error_handler</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_handle_background_error</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="p">[</span><span class="n">request</span><span class="o">.</span><span class="n">id</span><span class="p">]</span> <span class="o">=</span> <span class="n">result</span>
<span class="bp">self</span><span class="o">.</span><span class="n">request_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">request</span><span class="p">)</span>
<span class="n">error</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">request_error_queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">error</span><span class="p">,</span> <span class="ne">Exception</span><span class="p">):</span>
<span class="k">raise</span> <span class="n">error</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_handle_background_error</span><span class="p">()</span>
<span class="k">return</span> <span class="n">result</span>
<span class="k">def</span> <span class="fm">__del__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
<span class="k">def</span> <span class="fm">__enter__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">self</span>
<span class="k">def</span> <span class="fm">__exit__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">exc_type</span><span class="p">,</span> <span class="n">exc_value</span><span class="p">,</span> <span class="n">traceback</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
<span class="k">return</span> <span class="kc">False</span> <span class="c1"># propagate the exception</span>
<span class="k">if</span> <span class="n">enable_llm_debug</span><span class="p">():</span>
<span class="n">print_colored</span><span class="p">(</span><span class="s2">&quot;LLM debug mode enabled.</span><span class="se">\n</span><span class="s2">&quot;</span><span class="p">,</span> <span class="s2">&quot;yellow&quot;</span><span class="p">)</span>
<span class="c1"># This will dump all the alive threads when the process is interrupted by SIGINT.</span>
<span class="n">faulthandler</span><span class="o">.</span><span class="n">register</span><span class="p">(</span><span class="n">signal</span><span class="o">.</span><span class="n">SIGINT</span><span class="p">,</span> <span class="n">all_threads</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">has_event_loop</span><span class="p">()</span> <span class="o">-&gt;</span> <span class="nb">bool</span><span class="p">:</span>
<span class="k">try</span><span class="p">:</span>
<span class="n">asyncio</span><span class="o">.</span><span class="n">get_running_loop</span><span class="p">()</span>
<span class="k">except</span> <span class="ne">RuntimeError</span><span class="p">:</span>
<span class="k">return</span> <span class="kc">False</span>
<span class="k">return</span> <span class="kc">True</span>
</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 © 2024, NVidia.
<br/>
</p>
</div>
</div>
</div>
</footer>
</body>
</html>
Reference in New Issue View Git Blame Copy Permalink