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
305ff638ef
TensorRT-LLMs/_modules/tensorrt_llm/plugin/plugin.html
Kaiyu Xie 305ff638ef
Revert "doc:fix llm api reference blank issue. (#2669)" (#2700) 
This reverts commit 42d0585493.
2025-01-16 21:53:14 +08:00

921 lines
102 KiB
HTML
Raw Blame History

<!DOCTYPE html>
<html class="writer-html5" lang="en" data-content_root="../../../">
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>tensorrt_llm.plugin.plugin &mdash; tensorrt_llm documentation</title>
<link rel="stylesheet" type="text/css" href="../../../_static/pygments.css?v=80d5e7a1" />
<link rel="stylesheet" type="text/css" href="../../../_static/css/theme.css?v=e59714d7" />
<link rel="stylesheet" type="text/css" href="../../../_static/copybutton.css?v=76b2166b" />
<script src="../../../_static/jquery.js?v=5d32c60e"></script>
<script src="../../../_static/_sphinx_javascript_frameworks_compat.js?v=2cd50e6c"></script>
<script src="../../../_static/documentation_options.js?v=5929fcd5"></script>
<script src="../../../_static/doctools.js?v=9a2dae69"></script>
<script src="../../../_static/sphinx_highlight.js?v=dc90522c"></script>
<script src="../../../_static/clipboard.min.js?v=a7894cd8"></script>
<script src="../../../_static/copybutton.js?v=65e89d2a"></script>
<script src="../../../_static/js/theme.js"></script>
<link rel="index" title="Index" href="../../../genindex.html" />
<link rel="search" title="Search" href="../../../search.html" />
</head>
<body class="wy-body-for-nav">
<div class="wy-grid-for-nav">
<nav data-toggle="wy-nav-shift" class="wy-nav-side">
<div class="wy-side-scroll">
<div class="wy-side-nav-search" >
<a href="../../../index.html" class="icon icon-home">
tensorrt_llm
</a>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="../../../search.html" method="get">
<input type="text" name="q" placeholder="Search docs" aria-label="Search docs" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div><div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="Navigation menu">
<p class="caption" role="heading"><span class="caption-text">Getting Started</span></p>
<ul>
<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="../../../release-notes.html">Release Notes</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Installation</span></p>
<ul>
<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/windows.html">Installing on Windows</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../installation/build-from-source-windows.html">Building from Source Code on Windows</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../installation/grace-hopper.html">Installing on Grace Hopper</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">LLM API</span></p>
<ul>
<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 class="caption" role="heading"><span class="caption-text">LLM API Examples</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../../llm-api-examples/index.html">LLM Examples Introduction</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../llm-api-examples/customization.html">Common Customizations</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../llm-api-examples/llm_api_examples.html">Examples</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Model Definition API</span></p>
<ul>
<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 class="caption" role="heading"><span class="caption-text">C++ API</span></p>
<ul>
<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 class="caption" role="heading"><span class="caption-text">Command-Line Reference</span></p>
<ul>
<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 class="caption" role="heading"><span class="caption-text">Architecture</span></p>
<ul>
<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/core-concepts.html#compilation">Compilation</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../architecture/core-concepts.html#runtime">Runtime</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../architecture/core-concepts.html#multi-gpu-and-multi-node-support">Multi-GPU and Multi-Node Support</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 class="caption" role="heading"><span class="caption-text">Advanced</span></p>
<ul>
<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/inference-request.html#responses">Responses</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>
</ul>
<p class="caption" role="heading"><span class="caption-text">Performance</span></p>
<ul>
<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"><a class="reference internal" href="../../../performance/perf-best-practices.html">Best Practices</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../performance/perf-analysis.html">Performance Analysis</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Reference</span></p>
<ul>
<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 class="caption" role="heading"><span class="caption-text">Blogs</span></p>
<ul>
<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>
</div>
</nav>
<section data-toggle="wy-nav-shift" class="wy-nav-content-wrap"><nav class="wy-nav-top" aria-label="Mobile navigation menu" >
<i data-toggle="wy-nav-top" class="fa fa-bars"></i>
<a href="../../../index.html">tensorrt_llm</a>
</nav>
<div class="wy-nav-content">
<div class="rst-content">
<div role="navigation" aria-label="Page navigation">
<ul class="wy-breadcrumbs">
<li><a href="../../../index.html" class="icon icon-home" aria-label="Home"></a></li>
<li class="breadcrumb-item"><a href="../../index.html">Module code</a></li>
<li class="breadcrumb-item active">tensorrt_llm.plugin.plugin</li>
<li class="wy-breadcrumbs-aside">
</li>
</ul>
<hr/>
</div>
<div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
<div itemprop="articleBody">
<h1>Source code for tensorrt_llm.plugin.plugin</h1><div class="highlight"><pre>
<span></span><span class="c1"># SPDX-FileCopyrightText: Copyright (c) 2022-2024 NVIDIA CORPORATION &amp; AFFILIATES. All rights reserved.</span>
<span class="c1"># SPDX-License-Identifier: Apache-2.0</span>
<span class="c1">#</span>
<span class="c1"># Licensed under the Apache License, Version 2.0 (the &quot;License&quot;);</span>
<span class="c1"># you may not use this file except in compliance with the License.</span>
<span class="c1"># You may obtain a copy of the License at</span>
<span class="c1">#</span>
<span class="c1"># http://www.apache.org/licenses/LICENSE-2.0</span>
<span class="c1">#</span>
<span class="c1"># Unless required by applicable law or agreed to in writing, software</span>
<span class="c1"># distributed under the License is distributed on an &quot;AS IS&quot; BASIS,</span>
<span class="c1"># WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.</span>
<span class="c1"># See the License for the specific language governing permissions and</span>
<span class="c1"># limitations under the License.</span>
<span class="kn">import</span> <span class="nn">argparse</span>
<span class="kn">import</span> <span class="nn">ctypes</span>
<span class="kn">import</span> <span class="nn">platform</span>
<span class="kn">from</span> <span class="nn">collections</span> <span class="kn">import</span> <span class="n">OrderedDict</span>
<span class="kn">from</span> <span class="nn">dataclasses</span> <span class="kn">import</span> <span class="n">asdict</span><span class="p">,</span> <span class="n">dataclass</span><span class="p">,</span> <span class="n">field</span><span class="p">,</span> <span class="n">fields</span>
<span class="kn">from</span> <span class="nn">enum</span> <span class="kn">import</span> <span class="n">IntEnum</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">textwrap</span> <span class="kn">import</span> <span class="n">dedent</span>
<span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">List</span><span class="p">,</span> <span class="n">Optional</span><span class="p">,</span> <span class="n">Tuple</span>
<span class="kn">import</span> <span class="nn">tensorrt</span> <span class="k">as</span> <span class="nn">trt</span>
<span class="kn">from</span> <span class="nn">.._ipc_utils</span> <span class="kn">import</span> <span class="n">IpcMemory</span><span class="p">,</span> <span class="n">can_access_peer</span>
<span class="kn">from</span> <span class="nn">..bindings.internal.runtime</span> <span class="kn">import</span> <span class="n">lamport_initialize_all</span>
<span class="kn">from</span> <span class="nn">..logger</span> <span class="kn">import</span> <span class="n">logger</span>
<span class="kn">from</span> <span class="nn">..mapping</span> <span class="kn">import</span> <span class="n">Mapping</span>
<span class="n">TRT_LLM_PLUGIN_NAMESPACE</span> <span class="o">=</span> <span class="s1">&#39;tensorrt_llm&#39;</span>
<span class="k">def</span> <span class="nf">plugin_lib_path</span><span class="p">()</span> <span class="o">-&gt;</span> <span class="nb">str</span><span class="p">:</span>
<span class="n">project_dir</span> <span class="o">=</span> <span class="n">Path</span><span class="p">(</span><span class="vm">__file__</span><span class="p">)</span><span class="o">.</span><span class="n">parent</span><span class="o">.</span><span class="n">parent</span><span class="o">.</span><span class="n">absolute</span><span class="p">()</span>
<span class="n">dyn_lib</span> <span class="o">=</span> <span class="s2">&quot;libnvinfer_plugin_tensorrt_llm.so&quot;</span> <span class="k">if</span> <span class="n">platform</span><span class="o">.</span><span class="n">system</span><span class="p">(</span>
<span class="p">)</span> <span class="o">!=</span> <span class="s2">&quot;Windows&quot;</span> <span class="k">else</span> <span class="s2">&quot;nvinfer_plugin_tensorrt_llm.dll&quot;</span>
<span class="k">return</span> <span class="nb">str</span><span class="p">(</span><span class="n">project_dir</span><span class="o">.</span><span class="n">joinpath</span><span class="p">(</span><span class="s2">&quot;libs&quot;</span><span class="p">,</span> <span class="n">dyn_lib</span><span class="p">))</span>
<span class="k">def</span> <span class="nf">_load_plugin_lib</span><span class="p">():</span>
<span class="n">on_windows</span> <span class="o">=</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="s2">&quot;Windows&quot;</span>
<span class="n">winmode</span> <span class="o">=</span> <span class="mi">0</span> <span class="k">if</span> <span class="n">on_windows</span> <span class="k">else</span> <span class="kc">None</span>
<span class="n">handle</span> <span class="o">=</span> <span class="n">ctypes</span><span class="o">.</span><span class="n">CDLL</span><span class="p">(</span><span class="n">plugin_lib_path</span><span class="p">(),</span>
<span class="n">mode</span><span class="o">=</span><span class="n">ctypes</span><span class="o">.</span><span class="n">RTLD_GLOBAL</span><span class="p">,</span>
<span class="n">winmode</span><span class="o">=</span><span class="n">winmode</span><span class="p">)</span>
<span class="k">try</span><span class="p">:</span>
<span class="n">handle</span><span class="o">.</span><span class="n">initTrtLlmPlugins</span><span class="o">.</span><span class="n">argtypes</span> <span class="o">=</span> <span class="p">[</span><span class="n">ctypes</span><span class="o">.</span><span class="n">c_void_p</span><span class="p">,</span> <span class="n">ctypes</span><span class="o">.</span><span class="n">c_char_p</span><span class="p">]</span>
<span class="n">handle</span><span class="o">.</span><span class="n">initTrtLlmPlugins</span><span class="o">.</span><span class="n">restype</span> <span class="o">=</span> <span class="n">ctypes</span><span class="o">.</span><span class="n">c_bool</span>
<span class="k">except</span> <span class="ne">AttributeError</span> <span class="k">as</span> <span class="n">err</span><span class="p">:</span>
<span class="k">raise</span> <span class="ne">ImportError</span><span class="p">(</span><span class="s1">&#39;TensorRT-LLM Plugin is unavailable&#39;</span><span class="p">)</span> <span class="kn">from</span> <span class="nn">err</span>
<span class="k">try</span><span class="p">:</span>
<span class="k">assert</span> <span class="n">handle</span><span class="o">.</span><span class="n">initTrtLlmPlugins</span><span class="p">(</span>
<span class="kc">None</span><span class="p">,</span> <span class="n">TRT_LLM_PLUGIN_NAMESPACE</span><span class="o">.</span><span class="n">encode</span><span class="p">(</span><span class="s1">&#39;utf-8&#39;</span><span class="p">))</span>
<span class="k">except</span> <span class="ne">OSError</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
<span class="n">windows_err</span> <span class="o">=</span> <span class="s2">&quot;&quot;&quot;</span>
<span class="s2"> The error above may be caused by an outdated Microsoft Visual C++ Redistributable Version.</span>
<span class="s2"> Please install the latest MSVC from the link below and re-launch.</span>
<span class="s2"> https://learn.microsoft.com/en-us/cpp/windows/latest-supported-vc-redist?view=msvc-170#latest-microsoft-visual-c-redistributable-version</span>
<span class="s2"> &quot;&quot;&quot;</span>
<span class="n">err_msg</span> <span class="o">=</span> <span class="n">dedent</span><span class="p">(</span><span class="n">windows_err</span> <span class="k">if</span> <span class="n">on_windows</span> <span class="k">else</span> <span class="s2">&quot;Unknown error&quot;</span><span class="p">)</span>
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="n">err_msg</span><span class="p">)</span> <span class="kn">from</span> <span class="nn">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="k">raise</span> <span class="n">e</span>
<span class="k">class</span> <span class="nc">ContextFMHAType</span><span class="p">(</span><span class="n">IntEnum</span><span class="p">):</span>
<span class="n">disabled</span> <span class="o">=</span> <span class="mi">0</span>
<span class="c1"># FP16 I/O, FP16 Accumulation</span>
<span class="n">enabled</span> <span class="o">=</span> <span class="mi">1</span>
<span class="c1"># FP16 I/O, FP32 Accumulation</span>
<span class="n">enabled_with_fp32_acc</span> <span class="o">=</span> <span class="mi">2</span>
<span class="n">DEFAULT_PLUGIN_DTYPE_OPTIONS</span> <span class="o">=</span> <span class="p">[</span>
<span class="s2">&quot;auto&quot;</span><span class="p">,</span> <span class="s2">&quot;float16&quot;</span><span class="p">,</span> <span class="s2">&quot;float32&quot;</span><span class="p">,</span> <span class="s2">&quot;bfloat16&quot;</span><span class="p">,</span> <span class="s2">&quot;int32&quot;</span><span class="p">,</span> <span class="kc">None</span>
<span class="p">]</span>
<span class="n">PLUGIN_DTYPE_OPTIONS_MAP</span> <span class="o">=</span> <span class="p">{</span>
<span class="s2">&quot;gemm_swiglu_plugin&quot;</span><span class="p">:</span> <span class="p">[</span><span class="s2">&quot;fp8&quot;</span><span class="p">,</span> <span class="kc">None</span><span class="p">],</span>
<span class="s2">&quot;gemm_plugin&quot;</span><span class="p">:</span>
<span class="p">[</span><span class="s2">&quot;auto&quot;</span><span class="p">,</span> <span class="s2">&quot;float16&quot;</span><span class="p">,</span> <span class="s2">&quot;float32&quot;</span><span class="p">,</span> <span class="s2">&quot;bfloat16&quot;</span><span class="p">,</span> <span class="s2">&quot;int32&quot;</span><span class="p">,</span> <span class="s2">&quot;fp8&quot;</span><span class="p">,</span> <span class="kc">None</span><span class="p">],</span>
<span class="s2">&quot;low_latency_gemm_plugin&quot;</span><span class="p">:</span> <span class="p">[</span><span class="s2">&quot;fp8&quot;</span><span class="p">,</span> <span class="kc">None</span><span class="p">],</span>
<span class="s2">&quot;low_latency_gemm_swiglu_plugin&quot;</span><span class="p">:</span> <span class="p">[</span><span class="s2">&quot;fp8&quot;</span><span class="p">,</span> <span class="kc">None</span><span class="p">],</span>
<span class="p">}</span>
<span class="k">def</span> <span class="nf">_make_plugin_property</span><span class="p">(</span><span class="n">field_name</span><span class="p">:</span> <span class="nb">str</span><span class="p">,</span> <span class="n">field_type</span><span class="p">:</span> <span class="nb">type</span><span class="p">):</span>
<span class="k">def</span> <span class="nf">bind</span><span class="p">(</span><span class="n">field_name</span><span class="p">):</span>
<span class="n">storage_name</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">&#39;_</span><span class="si">{</span><span class="n">field_name</span><span class="si">}</span><span class="s1">&#39;</span>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">prop</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="n">field_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">storage_name</span><span class="p">)</span>
<span class="k">if</span> <span class="n">field_name</span> <span class="o">!=</span> <span class="s1">&#39;dtype&#39;</span> <span class="ow">and</span> <span class="n">field_value</span> <span class="o">==</span> <span class="s1">&#39;auto&#39;</span><span class="p">:</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">dtype</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">return</span> <span class="n">field_value</span>
<span class="nd">@prop</span><span class="o">.</span><span class="n">setter</span>
<span class="k">def</span> <span class="nf">prop</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">value</span><span class="p">):</span>
<span class="k">if</span> <span class="n">field_type</span> <span class="ow">is</span> <span class="nb">bool</span><span class="p">:</span>
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">value</span><span class="p">,</span> <span class="nb">bool</span><span class="p">),</span> \
<span class="sa">f</span><span class="s2">&quot;Plugin </span><span class="si">{</span><span class="n">field_name</span><span class="si">}</span><span class="s2"> expects </span><span class="si">{</span><span class="n">field_type</span><span class="si">}</span><span class="s2">, got </span><span class="si">{</span><span class="nb">type</span><span class="p">(</span><span class="n">value</span><span class="p">)</span><span class="si">}</span><span class="s2">&quot;</span>
<span class="k">elif</span> <span class="n">field_type</span> <span class="ow">in</span> <span class="p">(</span><span class="nb">str</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="n">plugin_dtype_options</span> <span class="o">=</span> <span class="n">DEFAULT_PLUGIN_DTYPE_OPTIONS</span>
<span class="k">if</span> <span class="n">field_name</span> <span class="ow">in</span> <span class="n">PLUGIN_DTYPE_OPTIONS_MAP</span><span class="p">:</span>
<span class="n">plugin_dtype_options</span> <span class="o">=</span> <span class="n">PLUGIN_DTYPE_OPTIONS_MAP</span><span class="p">[</span><span class="n">field_name</span><span class="p">]</span>
<span class="k">assert</span> <span class="n">value</span> <span class="ow">in</span> <span class="n">plugin_dtype_options</span><span class="p">,</span> \
<span class="sa">f</span><span class="s2">&quot;Plugin </span><span class="si">{</span><span class="n">field_name</span><span class="si">}</span><span class="s2"> expects values in </span><span class="si">{</span><span class="n">plugin_dtype_options</span><span class="si">}</span><span class="s2">, got </span><span class="si">{</span><span class="n">value</span><span class="si">}</span><span class="s2">&quot;</span>
<span class="k">if</span> <span class="n">field_name</span> <span class="o">==</span> <span class="s1">&#39;dtype&#39;</span><span class="p">:</span>
<span class="k">assert</span> <span class="n">value</span> <span class="ow">not</span> <span class="ow">in</span> <span class="p">[</span><span class="s1">&#39;auto&#39;</span><span class="p">,</span> <span class="kc">None</span><span class="p">],</span> \
<span class="s2">&quot;Plugin dtype cannot be auto or None&quot;</span>
<span class="nb">setattr</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">storage_name</span><span class="p">,</span> <span class="n">value</span><span class="p">)</span>
<span class="n">logger</span><span class="o">.</span><span class="n">info</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;Set </span><span class="si">{</span><span class="n">field_name</span><span class="si">}</span><span class="s2"> to </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="k">return</span> <span class="n">prop</span>
<span class="k">return</span> <span class="n">bind</span><span class="p">(</span><span class="n">field_name</span><span class="p">)</span>
<span class="k">class</span> <span class="nc">PluginConfigMeta</span><span class="p">(</span><span class="nb">type</span><span class="p">):</span>
<span class="k">def</span> <span class="fm">__new__</span><span class="p">(</span><span class="bp">cls</span><span class="p">,</span> <span class="n">name</span><span class="p">,</span> <span class="n">bases</span><span class="p">,</span> <span class="n">attrs</span><span class="p">):</span>
<span class="k">for</span> <span class="n">storage_name</span><span class="p">,</span> <span class="n">field_type</span> <span class="ow">in</span> <span class="n">attrs</span><span class="p">[</span><span class="s1">&#39;__annotations__&#39;</span><span class="p">]</span><span class="o">.</span><span class="n">items</span><span class="p">():</span>
<span class="k">assert</span> <span class="n">storage_name</span><span class="o">.</span><span class="n">startswith</span><span class="p">(</span><span class="s1">&#39;_&#39;</span><span class="p">)</span>
<span class="n">field_name</span> <span class="o">=</span> <span class="n">storage_name</span><span class="o">.</span><span class="n">lstrip</span><span class="p">(</span><span class="s1">&#39;_&#39;</span><span class="p">)</span>
<span class="n">attrs</span><span class="p">[</span><span class="n">field_name</span><span class="p">]</span> <span class="o">=</span> <span class="n">_make_plugin_property</span><span class="p">(</span><span class="n">field_name</span><span class="p">,</span> <span class="n">field_type</span><span class="p">)</span>
<span class="k">return</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__new__</span><span class="p">(</span><span class="bp">cls</span><span class="p">,</span> <span class="n">name</span><span class="p">,</span> <span class="n">bases</span><span class="p">,</span> <span class="n">attrs</span><span class="p">)</span>
<div class="viewcode-block" id="PluginConfig">
<a class="viewcode-back" href="../../../python-api/tensorrt_llm.plugin.html#tensorrt_llm.plugin.PluginConfig">[docs]</a>
<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">PluginConfig</span><span class="p">(</span><span class="n">metaclass</span><span class="o">=</span><span class="n">PluginConfigMeta</span><span class="p">):</span>
<span class="w"> </span><span class="sd">&quot;&quot;&quot;The config that manages plugin-related options.</span>
<span class="sd"> There are two option categories:</span>
<span class="sd"> * Plugin options (typically with xxx_plugin naming). These options can be assigned with:</span>
<span class="sd"> * &quot;float16&quot;/&quot;bfloat16&quot;/&quot;float32&quot;/&quot;int32&quot;, which means the plugin is enabled with the specified precision; (Some plugins only support limited dtype, i.e., gemm_swiglu_plugin and low_latency_gemm_swiglu_plugin only supports fp8 now)</span>
<span class="sd"> * &quot;auto&quot;, which means the plugin is enabled with the precision of `dtype` field (the `dtype` field must be same to model dtype, i.e., the one in PretrainedConfig);</span>
<span class="sd"> * None, which means the plugin is disabled.</span>
<span class="sd"> * Other features. These options can be assigned with boolean:</span>
<span class="sd"> * True, which means the plugin is enabled;</span>
<span class="sd"> * False, which means the plugin is disabled.</span>
<span class="sd"> Note: All the fields should use a prefix &quot;_&quot;; PluginConfigMeta will wrap each field as a property.</span>
<span class="sd"> This ensures the fields can only be assigned with allowed values.</span>
<span class="sd"> &quot;&quot;&quot;</span>
<span class="n">_dtype</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="s2">&quot;float16&quot;</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="c1"># Plugins</span>
<span class="n">_bert_attention_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="s2">&quot;auto&quot;</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;The plugin that uses efficient kernels and enables an in-place update of the KV cache for attention layer of BERT-like encoder models.&quot;</span>
<span class="p">})</span>
<span class="n">_gpt_attention_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="s2">&quot;auto&quot;</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;The plugin that uses efficient kernels and enables an in-place update of the KV cache for attention layer of GPT-like decoder models.&quot;</span>
<span class="p">})</span>
<span class="n">_gemm_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;The GEMM plugin that utilizes NVIDIA cuBLASLt to perform GEMM operations. &quot;</span>
<span class="s2">&quot;Note: it&#39;s only affective for non-quantized gemm operations (except FP8).&quot;</span>
<span class="s2">&quot;Note: For FP8, it also requires same calibration in checkpoint.&quot;</span>
<span class="p">})</span>
<span class="n">_gemm_swiglu_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;The GEMM + SwiGLU fusion in Gated-MLP combines two Matmul operations and &quot;</span>
<span class="s2">&quot;one SwiGLU operation into a single kernel. Currently this is only supported for FP8 precision on Hopper.&quot;</span>
<span class="p">})</span>
<span class="n">_fp8_rowwise_gemm_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;The quantized GEMM for fp8, which uses per token dynamic scales for &quot;</span>
<span class="s2">&quot;activation and per channel static scales for weights.&quot;</span>
<span class="s2">&quot;Note: It also requires same calibration in checkpoint.&quot;</span>
<span class="p">})</span>
<span class="n">_qserve_gemm_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;The quantized GEMM from [QServe](https://arxiv.org/abs/2405.04532), &quot;</span>
<span class="s2">&quot;which employs 4-bit quantization for weights and 8-bit quantization for activations.&quot;</span>
<span class="p">})</span>
<span class="n">_identity_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;The identity plugin simply copies inputs to outputs, it&#39;s used mostly for debugging purpose.&quot;</span>
<span class="p">})</span>
<span class="n">_nccl_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="s2">&quot;auto&quot;</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;The NCCL plugin wraps NCCL operators to support multi-GPU and even multi-nodes.&quot;</span>
<span class="p">})</span>
<span class="n">_lora_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span><span class="s2">&quot;help&quot;</span><span class="p">:</span> <span class="s2">&quot;Enable LoRA.&quot;</span><span class="p">})</span>
<span class="n">_weight_only_groupwise_quant_matmul_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Enable weight-only groupwise quantization matmul operators.&quot;</span>
<span class="p">})</span>
<span class="n">_weight_only_quant_matmul_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span><span class="s2">&quot;help&quot;</span><span class="p">:</span> <span class="s2">&quot;Enable weight-only quantization matmul operators.&quot;</span><span class="p">})</span>
<span class="n">_smooth_quant_plugins</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span> <span class="s2">&quot;Enable a group of plugins to support smooth quantization.&quot;</span>
<span class="p">})</span>
<span class="n">_smooth_quant_gemm_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Enable plugin that supports smooth quantization gemm kernels.&quot;</span>
<span class="p">})</span>
<span class="n">_layernorm_quantization_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Enable plugin that supports layernorm quantization kernels.&quot;</span>
<span class="p">})</span>
<span class="n">_rmsnorm_quantization_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span> <span class="s2">&quot;Enable plugin that supports rmsnorm quantization kernels.&quot;</span>
<span class="p">})</span>
<span class="n">_quantize_per_token_plugin</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span> <span class="s2">&quot;Enable plugin that supports per-token quantization.&quot;</span>
<span class="p">})</span>
<span class="n">_quantize_tensor_plugin</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span> <span class="s2">&quot;Enable plugin that supports per-tensor quantization.&quot;</span>
<span class="p">})</span>
<span class="n">_moe_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="s2">&quot;auto&quot;</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Enable some customized kernels to speed up the MoE layer of MoE models.&quot;</span>
<span class="p">})</span>
<span class="n">_mamba_conv1d_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="s2">&quot;auto&quot;</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Enable customized kernels to speed up conv1d operator for Mamba.&quot;</span>
<span class="p">})</span>
<span class="n">_fp4_gemm_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">metadata</span><span class="o">=</span><span class="p">{</span><span class="s2">&quot;help&quot;</span><span class="p">:</span> <span class="s2">&quot;Enable FP4 GEMM kernels.&quot;</span><span class="p">})</span>
<span class="n">_low_latency_gemm_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;The GEMM plugin that optimized specially for low latency scenarios.&quot;</span>
<span class="p">})</span>
<span class="n">_low_latency_gemm_swiglu_plugin</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;The GEMM + SwiGLU fusion plugin that optimized specially for low latency scenarios.&quot;</span>
<span class="p">})</span>
<span class="c1"># Features</span>
<span class="n">_context_fmha</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Enable the fused multi-head attention during the context phase, &quot;</span>
<span class="s2">&quot;will trigger a kernel that performs the MHA/MQA/GQA block using a single kernel.&quot;</span>
<span class="p">})</span>
<span class="n">_bert_context_fmha_fp32_acc</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Enable the FP32 accumulator for context FMHA in the bert_attention_plugin. &quot;</span>
<span class="s2">&quot;If disabled, FP16 is used, better performance but potentially worse accuracy is expected.&quot;</span>
<span class="p">})</span>
<span class="n">_paged_kv_cache</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">bool</span><span class="p">]</span> <span class="o">=</span> <span class="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="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Enable paged KV cache, which helps manage memory for the KV cache more efficiently, &quot;</span>
<span class="s2">&quot;and usually leads to an increase in the batch size and an improved efficiency.&quot;</span>
<span class="p">})</span>
<span class="n">_remove_input_padding</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Pack different tokens together, which reduces both the amount of computations and memory consumption.&quot;</span>
<span class="p">})</span>
<span class="n">_reduce_fusion</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Fuse the ResidualAdd and LayerNorm kernels after AllReduce into a single kernel, &quot;</span>
<span class="s2">&quot;resulting in improved end-to-end performance.&quot;</span>
<span class="p">})</span>
<span class="n">_user_buffer</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Eliminate extra copies from the local buffer to the shared buffer &quot;</span>
<span class="s2">&quot;in the communication kernel, leading to improved end-to-end performance. &quot;</span>
<span class="s2">&quot;This feature must be enabled with `--reduce_fusion enable` and &quot;</span>
<span class="s2">&quot;is currently only supported for the FP8 LLAMA model.&quot;</span>
<span class="p">})</span>
<span class="n">_tokens_per_block</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="mi">64</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Define how many tokens are contained in each paged kv cache block.&quot;</span>
<span class="p">})</span>
<span class="n">_use_paged_context_fmha</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Allow advanced features like KV cache reuse and chunked context.&quot;</span>
<span class="p">})</span>
<span class="n">_use_fp8_context_fmha</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;When FP8 quantization is activated, the attention can be further accelerated by enabling FP8 Context FMHA&quot;</span>
<span class="p">})</span>
<span class="n">_multiple_profiles</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Enables multiple TensorRT optimization profiles in the built engines, &quot;</span>
<span class="s2">&quot;will benefits the performance especially when GEMM plugin is disabled, &quot;</span>
<span class="s2">&quot;because more optimization profiles help TensorRT have more chances to select better kernels. &quot;</span>
<span class="s2">&quot;Note: This feature increases engine build time but no other adverse effects are expected.&quot;</span>
<span class="p">})</span>
<span class="n">_paged_state</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Enable paged state, which helps manage memory for the RNN state more efficiently.&quot;</span>
<span class="p">})</span>
<span class="n">_streamingllm</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Enable [StreamingLLM](https://arxiv.org/abs/2309.17453), which uses a window attention to perform efficient and stable LLM on long texts.&quot;</span>
<span class="p">})</span>
<span class="n">_manage_weights</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Enable TensorRT-LLM managed weights to speed up engine building process.&quot;</span>
<span class="p">})</span>
<span class="n">_use_fused_mlp</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Enable horizontal fusion in Gated-MLP that combines two Matmul &quot;</span>
<span class="s2">&quot;operations into a single one followed by a separate SwiGLU kernel.&quot;</span>
<span class="p">})</span>
<span class="n">_pp_reduce_scatter</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span>
<span class="n">default</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">metadata</span><span class="o">=</span><span class="p">{</span>
<span class="s2">&quot;help&quot;</span><span class="p">:</span>
<span class="s2">&quot;Enable a pipeline parallelism optimization with &quot;</span>
<span class="s2">&quot;ReduceScatter + AllGather targeting large MoE models.&quot;</span>
<span class="p">})</span>
<span class="k">def</span> <span class="nf">update_from_dict</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">config</span><span class="p">:</span> <span class="nb">dict</span><span class="p">):</span>
<span class="k">for</span> <span class="n">name</span> <span class="ow">in</span> <span class="n">config</span><span class="o">.</span><span class="n">keys</span><span class="p">():</span>
<span class="k">if</span> <span class="nb">hasattr</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">name</span><span class="p">):</span>
<span class="n">value_to_be_update</span> <span class="o">=</span> <span class="n">config</span><span class="p">[</span><span class="n">name</span><span class="p">]</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="nb">getattr</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">name</span><span class="p">),</span>
<span class="nb">bool</span><span class="p">)</span> <span class="ow">or</span> <span class="n">name</span> <span class="o">==</span> <span class="s1">&#39;paged_kv_cache&#39;</span><span class="p">:</span>
<span class="k">if</span> <span class="n">value_to_be_update</span> <span class="o">==</span> <span class="s2">&quot;enable&quot;</span><span class="p">:</span>
<span class="n">value_to_be_update</span> <span class="o">=</span> <span class="kc">True</span>
<span class="k">elif</span> <span class="n">value_to_be_update</span> <span class="o">==</span> <span class="s2">&quot;disable&quot;</span><span class="p">:</span>
<span class="n">value_to_be_update</span> <span class="o">=</span> <span class="kc">False</span>
<span class="k">elif</span> <span class="n">value_to_be_update</span> <span class="o">==</span> <span class="s2">&quot;disable&quot;</span><span class="p">:</span>
<span class="n">value_to_be_update</span> <span class="o">=</span> <span class="kc">None</span>
<span class="nb">setattr</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">name</span><span class="p">,</span> <span class="n">value_to_be_update</span><span class="p">)</span>
<span class="nd">@classmethod</span>
<span class="k">def</span> <span class="nf">from_dict</span><span class="p">(</span><span class="bp">cls</span><span class="p">,</span> <span class="n">config</span><span class="p">:</span> <span class="nb">dict</span><span class="p">):</span>
<span class="n">plugin_config</span> <span class="o">=</span> <span class="bp">cls</span><span class="p">()</span>
<span class="n">plugin_config</span><span class="o">.</span><span class="n">update_from_dict</span><span class="p">(</span><span class="n">config</span><span class="p">)</span>
<span class="k">return</span> <span class="n">plugin_config</span>
<span class="nd">@classmethod</span>
<span class="k">def</span> <span class="nf">from_arguments</span><span class="p">(</span><span class="bp">cls</span><span class="p">,</span> <span class="n">args</span><span class="p">:</span> <span class="n">argparse</span><span class="o">.</span><span class="n">Namespace</span><span class="p">):</span>
<span class="k">return</span> <span class="bp">cls</span><span class="o">.</span><span class="n">from_dict</span><span class="p">(</span><span class="nb">vars</span><span class="p">(</span><span class="n">args</span><span class="p">))</span>
<span class="k">def</span> <span class="nf">to_dict</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="n">config</span> <span class="o">=</span> <span class="n">asdict</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span>
<span class="c1"># Remove prefix &quot;_&quot; of the storage name</span>
<span class="n">config</span> <span class="o">=</span> <span class="p">{</span><span class="n">key</span><span class="o">.</span><span class="n">lstrip</span><span class="p">(</span><span class="s1">&#39;_&#39;</span><span class="p">):</span> <span class="n">value</span> <span class="k">for</span> <span class="n">key</span><span class="p">,</span> <span class="n">value</span> <span class="ow">in</span> <span class="n">config</span><span class="o">.</span><span class="n">items</span><span class="p">()}</span>
<span class="k">return</span> <span class="n">config</span>
<div class="viewcode-block" id="PluginConfig.to_legacy_setting">
<a class="viewcode-back" href="../../../python-api/tensorrt_llm.plugin.html#tensorrt_llm.plugin.PluginConfig.to_legacy_setting">[docs]</a>
<span class="k">def</span> <span class="nf">to_legacy_setting</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="w"> </span><span class="sd">&#39;&#39;&#39;Legacy setting means that all of the plugins and features are</span>
<span class="sd"> disabled, this is needed for the legacy `build.py` script, which will be</span>
<span class="sd"> migrated to the centralized building script `tensorrt_llm/commands/build.py`.</span>
<span class="sd"> After the migration is done, this function may or may not be deleted.</span>
<span class="sd"> &#39;&#39;&#39;</span>
<span class="k">for</span> <span class="n">field</span> <span class="ow">in</span> <span class="n">fields</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="c1"># Remove prefix &quot;_&quot; of the storage name</span>
<span class="n">field_name</span> <span class="o">=</span> <span class="n">field</span><span class="o">.</span><span class="n">name</span><span class="o">.</span><span class="n">lstrip</span><span class="p">(</span><span class="s1">&#39;_&#39;</span><span class="p">)</span>
<span class="k">if</span> <span class="n">field_name</span> <span class="o">==</span> <span class="s1">&#39;dtype&#39;</span><span class="p">:</span>
<span class="k">continue</span>
<span class="k">if</span> <span class="n">field</span><span class="o">.</span><span class="n">type</span> <span class="ow">in</span> <span class="p">(</span><span class="nb">str</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="nb">setattr</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">field_name</span><span class="p">,</span> <span class="kc">None</span><span class="p">)</span>
<span class="k">elif</span> <span class="n">field</span><span class="o">.</span><span class="n">type</span> <span class="o">==</span> <span class="nb">bool</span> <span class="ow">or</span> <span class="n">field_name</span> <span class="o">==</span> <span class="s1">&#39;paged_kv_cache&#39;</span><span class="p">:</span>
<span class="nb">setattr</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">field_name</span><span class="p">,</span> <span class="kc">False</span><span class="p">)</span></div>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">context_fmha_type</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">bert_context_fmha_fp32_acc</span><span class="p">:</span>
<span class="k">return</span> <span class="n">ContextFMHAType</span><span class="o">.</span><span class="n">enabled_with_fp32_acc</span>
<span class="k">elif</span> <span class="bp">self</span><span class="o">.</span><span class="n">context_fmha</span><span class="p">:</span>
<span class="k">return</span> <span class="n">ContextFMHAType</span><span class="o">.</span><span class="n">enabled</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">return</span> <span class="n">ContextFMHAType</span><span class="o">.</span><span class="n">disabled</span>
<span class="k">def</span> <span class="nf">is_context_fmha_enabled</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">context_fmha_type</span> <span class="o">!=</span> <span class="n">ContextFMHAType</span><span class="o">.</span><span class="n">disabled</span>
<span class="nd">@context_fmha_type</span><span class="o">.</span><span class="n">setter</span>
<span class="k">def</span> <span class="nf">context_fmha_type</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">value</span><span class="p">):</span>
<span class="k">if</span> <span class="n">value</span> <span class="o">==</span> <span class="n">ContextFMHAType</span><span class="o">.</span><span class="n">disabled</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">context_fmha</span> <span class="o">=</span> <span class="kc">False</span>
<span class="bp">self</span><span class="o">.</span><span class="n">bert_context_fmha_fp32_acc</span> <span class="o">=</span> <span class="kc">False</span>
<span class="k">else</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">context_fmha</span> <span class="o">=</span> <span class="kc">True</span>
<span class="k">if</span> <span class="n">value</span> <span class="o">==</span> <span class="n">ContextFMHAType</span><span class="o">.</span><span class="n">enabled</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">bert_context_fmha_fp32_acc</span> <span class="o">=</span> <span class="kc">False</span>
<span class="k">elif</span> <span class="n">value</span> <span class="o">==</span> <span class="n">ContextFMHAType</span><span class="o">.</span><span class="n">enabled_with_fp32_acc</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">bert_context_fmha_fp32_acc</span> <span class="o">=</span> <span class="kc">True</span>
<span class="k">def</span> <span class="nf">set_smooth_quant_plugins</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">dtype</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="s2">&quot;auto&quot;</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">smooth_quant_gemm_plugin</span> <span class="o">=</span> <span class="n">dtype</span>
<span class="bp">self</span><span class="o">.</span><span class="n">rmsnorm_quantization_plugin</span> <span class="o">=</span> <span class="n">dtype</span>
<span class="bp">self</span><span class="o">.</span><span class="n">layernorm_quantization_plugin</span> <span class="o">=</span> <span class="n">dtype</span>
<span class="bp">self</span><span class="o">.</span><span class="n">quantize_per_token_plugin</span> <span class="o">=</span> <span class="kc">True</span>
<span class="bp">self</span><span class="o">.</span><span class="n">quantize_tensor_plugin</span> <span class="o">=</span> <span class="kc">True</span>
<span class="k">return</span> <span class="bp">self</span>
<span class="k">def</span> <span class="nf">set_qserve_plugins</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">dtype</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="s2">&quot;auto&quot;</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">qserve_gemm_plugin</span> <span class="o">=</span> <span class="n">dtype</span>
<span class="bp">self</span><span class="o">.</span><span class="n">rmsnorm_quantization_plugin</span> <span class="o">=</span> <span class="n">dtype</span>
<span class="bp">self</span><span class="o">.</span><span class="n">quantize_per_token_plugin</span> <span class="o">=</span> <span class="kc">True</span>
<span class="k">return</span> <span class="bp">self</span>
<span class="k">def</span> <span class="nf">set_fp8_rowwise_quant_plugins</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">dtype</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="s2">&quot;auto&quot;</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">fp8_rowwise_gemm_plugin</span> <span class="o">=</span> <span class="n">dtype</span>
<span class="bp">self</span><span class="o">.</span><span class="n">rmsnorm_quantization_plugin</span> <span class="o">=</span> <span class="n">dtype</span>
<span class="c1"># self.layernorm_quantization_plugin = dtype</span>
<span class="bp">self</span><span class="o">.</span><span class="n">quantize_per_token_plugin</span> <span class="o">=</span> <span class="kc">True</span>
<span class="bp">self</span><span class="o">.</span><span class="n">quantize_tensor_plugin</span> <span class="o">=</span> <span class="kc">True</span>
<span class="k">return</span> <span class="bp">self</span>
<span class="k">def</span> <span class="nf">set_context_fmha</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">context_fmha_type</span><span class="o">=</span><span class="n">ContextFMHAType</span><span class="o">.</span><span class="n">enabled</span><span class="p">):</span>
<span class="k">assert</span> <span class="nb">type</span><span class="p">(</span><span class="n">context_fmha_type</span><span class="p">)</span> <span class="o">==</span> <span class="n">ContextFMHAType</span>
<span class="bp">self</span><span class="o">.</span><span class="n">context_fmha_type</span> <span class="o">=</span> <span class="n">context_fmha_type</span>
<span class="k">return</span> <span class="bp">self</span>
<span class="k">def</span> <span class="nf">enable_paged_kv_cache</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">tokens_per_block</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">64</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">paged_kv_cache</span> <span class="o">=</span> <span class="kc">True</span>
<span class="bp">self</span><span class="o">.</span><span class="n">tokens_per_block</span> <span class="o">=</span> <span class="n">tokens_per_block</span>
<span class="k">return</span> <span class="bp">self</span>
<span class="k">def</span> <span class="nf">set_nccl_plugin</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">dtype</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="s2">&quot;auto&quot;</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">nccl_plugin</span> <span class="o">=</span> <span class="n">dtype</span>
<span class="n">init_all_reduce_helper</span><span class="p">()</span>
<span class="k">return</span> <span class="bp">self</span>
<span class="k">def</span> <span class="nf">set_lora_plugin</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">dtype</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">lora_plugin</span> <span class="o">=</span> <span class="n">dtype</span>
<span class="k">return</span> <span class="bp">self</span></div>
<span class="c1"># Only plugin configs in this list will be exposed as `trtllm-build` arguments,</span>
<span class="c1"># others are automatically enabled when needed, no need for users to control.</span>
<span class="n">cli_plugin_args</span> <span class="o">=</span> <span class="p">[</span>
<span class="c1"># Plugins</span>
<span class="s2">&quot;bert_attention_plugin&quot;</span><span class="p">,</span>
<span class="s2">&quot;gpt_attention_plugin&quot;</span><span class="p">,</span>
<span class="s2">&quot;gemm_plugin&quot;</span><span class="p">,</span>
<span class="s2">&quot;gemm_swiglu_plugin&quot;</span><span class="p">,</span>
<span class="s2">&quot;fp8_rowwise_gemm_plugin&quot;</span><span class="p">,</span>
<span class="s2">&quot;lora_plugin&quot;</span><span class="p">,</span>
<span class="s2">&quot;moe_plugin&quot;</span><span class="p">,</span>
<span class="s2">&quot;mamba_conv1d_plugin&quot;</span><span class="p">,</span>
<span class="s2">&quot;nccl_plugin&quot;</span><span class="p">,</span>
<span class="s2">&quot;low_latency_gemm_plugin&quot;</span><span class="p">,</span>
<span class="s2">&quot;low_latency_gemm_swiglu_plugin&quot;</span><span class="p">,</span>
<span class="c1"># Features</span>
<span class="s2">&quot;context_fmha&quot;</span><span class="p">,</span>
<span class="s2">&quot;bert_context_fmha_fp32_acc&quot;</span><span class="p">,</span>
<span class="s2">&quot;remove_input_padding&quot;</span><span class="p">,</span>
<span class="s2">&quot;tokens_per_block&quot;</span><span class="p">,</span>
<span class="s2">&quot;use_paged_context_fmha&quot;</span><span class="p">,</span>
<span class="s2">&quot;use_fp8_context_fmha&quot;</span><span class="p">,</span>
<span class="s2">&quot;multiple_profiles&quot;</span><span class="p">,</span>
<span class="s2">&quot;paged_state&quot;</span><span class="p">,</span>
<span class="s2">&quot;streamingllm&quot;</span><span class="p">,</span>
<span class="s2">&quot;reduce_fusion&quot;</span><span class="p">,</span>
<span class="s2">&quot;user_buffer&quot;</span><span class="p">,</span>
<span class="s2">&quot;use_fused_mlp&quot;</span><span class="p">,</span>
<span class="s2">&quot;pp_reduce_scatter&quot;</span><span class="p">,</span>
<span class="p">]</span>
<span class="k">def</span> <span class="nf">add_plugin_argument</span><span class="p">(</span><span class="n">parser</span><span class="p">:</span> <span class="n">argparse</span><span class="o">.</span><span class="n">ArgumentParser</span><span class="p">):</span>
<span class="n">plugin_config</span> <span class="o">=</span> <span class="n">PluginConfig</span><span class="p">()</span>
<span class="k">for</span> <span class="n">field</span> <span class="ow">in</span> <span class="n">fields</span><span class="p">(</span><span class="n">plugin_config</span><span class="p">):</span>
<span class="c1"># Remove prefix &quot;_&quot; of the storage name</span>
<span class="n">field_name</span> <span class="o">=</span> <span class="n">field</span><span class="o">.</span><span class="n">name</span><span class="o">.</span><span class="n">lstrip</span><span class="p">(</span><span class="s1">&#39;_&#39;</span><span class="p">)</span>
<span class="k">if</span> <span class="n">field_name</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">cli_plugin_args</span><span class="p">:</span>
<span class="k">continue</span>
<span class="k">if</span> <span class="n">field</span><span class="o">.</span><span class="n">metadata</span> <span class="ow">and</span> <span class="s2">&quot;help&quot;</span> <span class="ow">in</span> <span class="n">field</span><span class="o">.</span><span class="n">metadata</span><span class="p">:</span>
<span class="n">help_message</span> <span class="o">=</span> <span class="n">field</span><span class="o">.</span><span class="n">metadata</span><span class="p">[</span><span class="s2">&quot;help&quot;</span><span class="p">]</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">raise</span> <span class="ne">AttributeError</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;Please add help message for </span><span class="si">{</span><span class="n">field_name</span><span class="si">}</span><span class="s2">.&quot;</span><span class="p">)</span>
<span class="k">if</span> <span class="n">field</span><span class="o">.</span><span class="n">type</span> <span class="ow">in</span> <span class="p">(</span><span class="nb">str</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="n">plugin_dtype_options</span> <span class="o">=</span> <span class="n">DEFAULT_PLUGIN_DTYPE_OPTIONS</span>
<span class="k">if</span> <span class="n">field_name</span> <span class="ow">in</span> <span class="n">PLUGIN_DTYPE_OPTIONS_MAP</span><span class="p">:</span>
<span class="n">plugin_dtype_options</span> <span class="o">=</span> <span class="n">PLUGIN_DTYPE_OPTIONS_MAP</span><span class="p">[</span><span class="n">field_name</span><span class="p">]</span>
<span class="n">parser</span><span class="o">.</span><span class="n">add_argument</span><span class="p">(</span>
<span class="s2">&quot;--&quot;</span> <span class="o">+</span> <span class="n">field_name</span><span class="p">,</span>
<span class="nb">type</span><span class="o">=</span><span class="nb">str</span><span class="p">,</span>
<span class="n">default</span><span class="o">=</span><span class="n">field</span><span class="o">.</span><span class="n">default</span> <span class="k">if</span> <span class="n">field</span><span class="o">.</span><span class="n">default</span> <span class="k">else</span> <span class="s2">&quot;disable&quot;</span><span class="p">,</span>
<span class="n">choices</span><span class="o">=</span><span class="p">[</span><span class="n">x</span> <span class="k">if</span> <span class="n">x</span> <span class="k">else</span> <span class="s2">&quot;disable&quot;</span> <span class="k">for</span> <span class="n">x</span> <span class="ow">in</span> <span class="n">plugin_dtype_options</span><span class="p">],</span>
<span class="n">help</span><span class="o">=</span><span class="n">help_message</span><span class="p">)</span>
<span class="k">elif</span> <span class="n">field</span><span class="o">.</span><span class="n">type</span> <span class="o">==</span> <span class="nb">bool</span><span class="p">:</span>
<span class="n">parser</span><span class="o">.</span><span class="n">add_argument</span><span class="p">(</span>
<span class="s2">&quot;--&quot;</span> <span class="o">+</span> <span class="n">field_name</span><span class="p">,</span>
<span class="nb">type</span><span class="o">=</span><span class="nb">str</span><span class="p">,</span>
<span class="n">default</span><span class="o">=</span><span class="s2">&quot;enable&quot;</span> <span class="k">if</span> <span class="n">field</span><span class="o">.</span><span class="n">default</span> <span class="k">else</span> <span class="s2">&quot;disable&quot;</span><span class="p">,</span>
<span class="n">choices</span><span class="o">=</span><span class="p">[</span><span class="s2">&quot;enable&quot;</span><span class="p">,</span> <span class="s2">&quot;disable&quot;</span><span class="p">],</span>
<span class="n">help</span><span class="o">=</span><span class="n">help_message</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">parser</span><span class="o">.</span><span class="n">add_argument</span><span class="p">(</span><span class="s2">&quot;--&quot;</span> <span class="o">+</span> <span class="n">field_name</span><span class="p">,</span>
<span class="nb">type</span><span class="o">=</span><span class="n">field</span><span class="o">.</span><span class="n">type</span><span class="p">,</span>
<span class="n">default</span><span class="o">=</span><span class="n">field</span><span class="o">.</span><span class="n">default</span><span class="p">,</span>
<span class="n">help</span><span class="o">=</span><span class="n">help_message</span><span class="p">)</span>
<span class="k">return</span> <span class="n">parser</span>
<span class="k">class</span> <span class="nc">CustomAllReduceHelper</span><span class="p">:</span>
<span class="w"> </span><span class="sd">&quot;&quot;&quot;</span>
<span class="sd"> Globally visible class to help usage of custom_all_reduce plugin.</span>
<span class="sd"> Provides the following utilities:</span>
<span class="sd"> workspace: Tensor</span>
<span class="sd"> When using CUSTOM or AUTO mode, a tensor containing pointers to memory</span>
<span class="sd"> visible to all GPUs. It should be 3 pointers per TP rank -</span>
<span class="sd"> ptr to data buffer, ptr to barriers in, ptr to barriers out.</span>
<span class="sd"> It must be initialized using IpcMemory class.</span>
<span class="sd"> Usage:</span>
<span class="sd"> - Set custom_all_reduce_helper.workspace with the required tensor.</span>
<span class="sd"> Then, each instance of allreduce will reference that tensor automatically.</span>
<span class="sd"> &quot;&quot;&quot;</span>
<span class="n">POINTERS_PER_RANK</span> <span class="o">=</span> <span class="mi">7</span>
<span class="n">POINTERS_OF_COUNTER</span> <span class="o">=</span> <span class="mi">2</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="o">-&gt;</span> <span class="kc">None</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">workspace</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">def</span> <span class="nf">set_workspace_tensor</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span>
<span class="n">mapping</span><span class="p">:</span> <span class="n">Mapping</span><span class="p">,</span>
<span class="n">num_profiles</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="p">):</span>
<span class="kn">from</span> <span class="nn">..functional</span> <span class="kn">import</span> <span class="n">Tensor</span>
<span class="n">workspace_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">POINTERS_PER_RANK</span> <span class="o">*</span> <span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span> <span class="o">+</span> <span class="bp">self</span><span class="o">.</span><span class="n">POINTERS_OF_COUNTER</span>
<span class="n">dim_range</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="n">num_profiles</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">dim_range</span> <span class="o">=</span> <span class="n">OrderedDict</span><span class="p">([(</span><span class="s1">&#39;all_reduce_size&#39;</span><span class="p">,</span>
<span class="p">[</span><span class="n">workspace_size</span><span class="p">]</span> <span class="o">*</span> <span class="n">num_profiles</span><span class="p">)])</span>
<span class="bp">self</span><span class="o">.</span><span class="n">workspace</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="s1">&#39;all_reduce_workspace&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int64</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="n">workspace_size</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">dim_range</span><span class="p">,</span>
<span class="p">)</span>
<span class="nd">@staticmethod</span>
<span class="k">def</span> <span class="nf">max_workspace_size_auto</span><span class="p">(</span><span class="n">tp_size</span><span class="p">:</span> <span class="nb">int</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">int</span><span class="p">:</span>
<span class="k">if</span> <span class="n">tp_size</span> <span class="o">&lt;=</span> <span class="mi">2</span><span class="p">:</span>
<span class="k">return</span> <span class="mi">16_000_000</span>
<span class="k">return</span> <span class="mi">8_000_000</span>
<span class="nd">@staticmethod</span>
<span class="k">def</span> <span class="nf">allocate_workspace</span><span class="p">(</span><span class="n">mapping</span><span class="p">:</span> <span class="n">Mapping</span><span class="p">,</span>
<span class="n">size</span><span class="p">:</span> <span class="nb">int</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="n">Tuple</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="n">IpcMemory</span><span class="p">],</span> <span class="s2">&quot;torch.tensor&quot;</span><span class="p">]:</span>
<span class="kn">import</span> <span class="nn">torch</span>
<span class="n">is_p2p_supported</span> <span class="o">=</span> <span class="n">can_access_peer</span><span class="p">(</span><span class="n">mapping</span><span class="p">)</span>
<span class="n">ipc_buffers_ping</span> <span class="o">=</span> <span class="n">IpcMemory</span><span class="p">(</span><span class="n">mapping</span><span class="p">,</span> <span class="n">size</span> <span class="o">*</span> <span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
<span class="n">is_p2p_supported</span><span class="p">)</span>
<span class="n">ipc_buffers_pong</span> <span class="o">=</span> <span class="n">IpcMemory</span><span class="p">(</span><span class="n">mapping</span><span class="p">,</span> <span class="n">size</span> <span class="o">*</span> <span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
<span class="n">is_p2p_supported</span><span class="p">)</span>
<span class="n">ipc_barriers_in</span> <span class="o">=</span> <span class="n">IpcMemory</span><span class="p">(</span>
<span class="n">mapping</span><span class="p">,</span> <span class="n">IpcMemory</span><span class="o">.</span><span class="n">IPC_BARRIERS_SIZE_PER_GPU</span> <span class="o">*</span> <span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span> <span class="o">*</span> <span class="mi">2</span><span class="p">,</span>
<span class="n">is_p2p_supported</span><span class="p">)</span>
<span class="n">ipc_barriers_out</span> <span class="o">=</span> <span class="n">IpcMemory</span><span class="p">(</span>
<span class="n">mapping</span><span class="p">,</span> <span class="n">IpcMemory</span><span class="o">.</span><span class="n">IPC_BARRIERS_SIZE_PER_GPU</span> <span class="o">*</span> <span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span> <span class="o">*</span> <span class="mi">2</span><span class="p">,</span>
<span class="n">is_p2p_supported</span><span class="p">)</span>
<span class="n">lamport_buffers_0</span> <span class="o">=</span> <span class="n">IpcMemory</span><span class="p">(</span><span class="n">mapping</span><span class="p">,</span> <span class="n">size</span> <span class="o">*</span> <span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
<span class="n">is_p2p_supported</span><span class="p">)</span>
<span class="n">lamport_buffers_1</span> <span class="o">=</span> <span class="n">IpcMemory</span><span class="p">(</span><span class="n">mapping</span><span class="p">,</span> <span class="n">size</span> <span class="o">*</span> <span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
<span class="n">is_p2p_supported</span><span class="p">)</span>
<span class="n">lamport_buffers_2</span> <span class="o">=</span> <span class="n">IpcMemory</span><span class="p">(</span><span class="n">mapping</span><span class="p">,</span> <span class="n">size</span> <span class="o">*</span> <span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
<span class="n">is_p2p_supported</span><span class="p">)</span>
<span class="n">rank</span> <span class="o">=</span> <span class="n">mapping</span><span class="o">.</span><span class="n">rank</span>
<span class="n">tp_rank</span> <span class="o">=</span> <span class="n">mapping</span><span class="o">.</span><span class="n">tp_rank</span>
<span class="k">if</span> <span class="n">rank</span> <span class="o">==</span> <span class="n">tp_rank</span> <span class="ow">and</span> <span class="n">is_p2p_supported</span><span class="p">:</span>
<span class="n">lamport_initialize_all</span><span class="p">(</span>
<span class="n">lamport_buffers_0</span><span class="o">.</span><span class="n">local_ptr</span><span class="p">,</span>
<span class="n">lamport_buffers_1</span><span class="o">.</span><span class="n">local_ptr</span><span class="p">,</span>
<span class="n">lamport_buffers_2</span><span class="o">.</span><span class="n">local_ptr</span><span class="p">,</span>
<span class="n">size</span> <span class="o">*</span> <span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
<span class="p">)</span>
<span class="n">buffers</span> <span class="o">=</span> <span class="p">[</span>
<span class="n">ipc_buffers_ping</span><span class="p">,</span> <span class="n">ipc_buffers_pong</span><span class="p">,</span> <span class="n">ipc_barriers_in</span><span class="p">,</span>
<span class="n">ipc_barriers_out</span><span class="p">,</span> <span class="n">lamport_buffers_0</span><span class="p">,</span> <span class="n">lamport_buffers_1</span><span class="p">,</span>
<span class="n">lamport_buffers_2</span>
<span class="p">]</span>
<span class="k">return</span> <span class="n">buffers</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">ipc_buffers_ping</span><span class="o">.</span><span class="n">serialize</span><span class="p">()</span> <span class="o">+</span> <span class="n">ipc_buffers_pong</span><span class="o">.</span><span class="n">serialize</span><span class="p">()</span> <span class="o">+</span>
<span class="n">ipc_barriers_in</span><span class="o">.</span><span class="n">serialize</span><span class="p">()</span> <span class="o">+</span> <span class="n">ipc_barriers_out</span><span class="o">.</span><span class="n">serialize</span><span class="p">()</span> <span class="o">+</span>
<span class="n">lamport_buffers_0</span><span class="o">.</span><span class="n">serialize</span><span class="p">()</span> <span class="o">+</span> <span class="n">lamport_buffers_1</span><span class="o">.</span><span class="n">serialize</span><span class="p">()</span> <span class="o">+</span>
<span class="n">lamport_buffers_2</span><span class="o">.</span><span class="n">serialize</span><span class="p">()</span> <span class="o">+</span> <span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">+</span> <span class="p">[</span><span class="mi">0</span><span class="p">],</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">int64</span><span class="p">,</span>
<span class="n">device</span><span class="o">=</span><span class="s2">&quot;cpu&quot;</span><span class="p">)</span>
<span class="n">custom_all_reduce_helper</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">def</span> <span class="nf">init_all_reduce_helper</span><span class="p">():</span>
<span class="k">global</span> <span class="n">custom_all_reduce_helper</span>
<span class="n">custom_all_reduce_helper</span> <span class="o">=</span> <span class="n">CustomAllReduceHelper</span><span class="p">()</span>
<span class="k">def</span> <span class="nf">current_all_reduce_helper</span><span class="p">():</span>
<span class="k">global</span> <span class="n">custom_all_reduce_helper</span>
<span class="k">assert</span> <span class="n">custom_all_reduce_helper</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">,</span> <span class="s2">&quot;You must call `init_all_reduce_helper` first&quot;</span>
<span class="k">return</span> <span class="n">custom_all_reduce_helper</span>
</pre></div>
</div>
</div>
<footer>
<hr/>
<div role="contentinfo">
<jinja2.runtime.BlockReference object at 0x7f5e6e49fef0>
<div class="footer">
<p>
Copyright © 2024 NVIDIA Corporation
</p>
<p>
<a class="Link" href="https://www.nvidia.com/en-us/about-nvidia/privacy-policy/" target="_blank" rel="noopener"
data-cms-ai="0">Privacy Policy</a> |
<a class="Link" href="https://www.nvidia.com/en-us/about-nvidia/privacy-center/" target="_blank" rel="noopener"
data-cms-ai="0">Manage My Privacy</a> |
<a class="Link" href="https://www.nvidia.com/en-us/preferences/start/" target="_blank" rel="noopener"
data-cms-ai="0">Do Not Sell or Share My Data</a> |
<a class="Link" href="https://www.nvidia.com/en-us/about-nvidia/terms-of-service/" target="_blank"
rel="noopener" data-cms-ai="0">Terms of Service</a> |
<a class="Link" href="https://www.nvidia.com/en-us/about-nvidia/accessibility/" target="_blank" rel="noopener"
data-cms-ai="0">Accessibility</a> |
<a class="Link" href="https://www.nvidia.com/en-us/about-nvidia/company-policies/" target="_blank"
rel="noopener" data-cms-ai="0">Corporate Policies</a> |
<a class="Link" href="https://www.nvidia.com/en-us/product-security/" target="_blank" rel="noopener"
data-cms-ai="0">Product Security</a> |
<a class="Link" href="https://www.nvidia.com/en-us/contact/" target="_blank" rel="noopener"
data-cms-ai="0">Contact</a>
</p>
</div>
</div>
</footer>
</div>
</div>
</section>
</div>
<script>
jQuery(function () {
SphinxRtdTheme.Navigation.enable(true);
});
</script>
</body>
</html>
Reference in New Issue View Git Blame Copy Permalink