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
130807da9f
TensorRT-LLMs/performance/useful-runtime-flags.html
Iman Tabrizian 130807da9f Update docs (#2732) 
* Update docs

* Update windows install version

Update gh pages (#2741)

update gh pages (#2743)

gh pages update (#2746)

Update gh-pages (#2764)

Update
2025-02-11 02:56:32 +00:00

417 lines
36 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" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>Useful Runtime Options &mdash; tensorrt_llm documentation</title>
<link rel="stylesheet" type="text/css" href="../_static/pygments.css?v=b86133f3" />
<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=9bcbadda"></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="../torch.html">PyTorch Backend</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>
<li class="toctree-l1"><a class="reference internal" href="../advanced/disaggregated-service.html">Disaggregated-Service (experimental)</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="perf-overview.html">Overview</a></li>
<li class="toctree-l1"><a class="reference internal" href="perf-benchmarking.html">Benchmarking</a></li>
<li class="toctree-l1"><a class="reference internal" href="introduction.html">Best Practices</a></li>
<li class="toctree-l1"><a class="reference internal" href="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 active">Useful Runtime Options</li>
<li class="wy-breadcrumbs-aside">
<a href="../_sources/performance/useful-runtime-flags.md.txt" rel="nofollow"> View page source</a>
</li>
</ul>
<hr/>
</div>
<div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
<div itemprop="articleBody">
<section id="useful-runtime-options">
<span id="useful-runtime-flags"></span><h1>Useful Runtime Options<a class="headerlink" href="#useful-runtime-options" title="Link to this heading"></a></h1>
<p>This part summarizes the runtime configuration knobs that can be tweaked to
enhance the performance of already built engines. As compared to previous examples where
the LLM-API was used to build and save an engine but not to process any requests,
runtime knobs would be specified when you are using the LLM-API to actually run inference
like in the <a class="reference internal" href="benchmarking-default-performance.html#before-you-begin-tensorrt-llm-llm-api"><span class="std std-ref">LLM-API end-to-end example</span></a></p>
<section id="capacity-scheduler-policy">
<h2>Capacity Scheduler Policy<a class="headerlink" href="#capacity-scheduler-policy" title="Link to this heading"></a></h2>
<p>TensorRT-LLM currently supports three batch scheduler policies: <code class="docutils literal notranslate"><span class="pre">GUARANTEED_NO_EVICT</span></code> (default),
<code class="docutils literal notranslate"><span class="pre">MAX_UTILIZATION</span></code> and <code class="docutils literal notranslate"><span class="pre">STATIC_BATCH</span></code>.</p>
<p>The scheduling policy can be set to <code class="docutils literal notranslate"><span class="pre">MAX_UTILIZATION</span></code> to pack as many
requests as possible at each iteration of the forward loop, when in-flight
sequence batching is enabled. It maximizes the utilization of the GPUs by
aggressively scheduling requests at the risk of having to pause requests if the
KV cache size limit is reached.</p>
<p>For a more conservative approach with respect to the KV cache limitations in
terms of memory allocation, <code class="docutils literal notranslate"><span class="pre">CapacitySchedulerPolicy</span></code> should be set to
<code class="docutils literal notranslate"><span class="pre">GUARANTEED_NO_EVICT</span></code> to guarantee that a started request is never paused.</p>
<p>If the goal is to maximizes the throughput, users should try <code class="docutils literal notranslate"><span class="pre">MAX_UTILIZATION</span></code>.
However, they need to keep in mind that it may have a negative impact on
latency if requests have to be paused.</p>
<p><code class="docutils literal notranslate"><span class="pre">STATIC_BATCH</span></code> is a legacy mode and is not recommended for production usage.</p>
<p>To switch the capacity scheduler policy from the default of <code class="docutils literal notranslate"><span class="pre">GUARANTEED_NO_EVICT</span></code> to <code class="docutils literal notranslate"><span class="pre">MAX_UTILIZATION</span></code>
you would modify the <a class="reference internal" href="benchmarking-default-performance.html#before-you-begin-tensorrt-llm-llm-api"><span class="std std-ref">LLM-API end-to-end example</span></a> to be:</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">from</span><span class="w"> </span><span class="nn">tensorrt_llm</span><span class="w"> </span><span class="kn">import</span> <span class="n">LLM</span><span class="p">,</span> <span class="n">SamplingParams</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">tensorrt_llm.bindings.executor</span><span class="w"> </span><span class="kn">import</span> <span class="n">SchedulerConfig</span><span class="p">,</span> <span class="n">CapacitySchedulerPolicy</span>
<span class="k">def</span><span class="w"> </span><span class="nf">main</span><span class="p">():</span>
<span class="n">prompts</span> <span class="o">=</span> <span class="p">[</span>
<span class="s2">&quot;Hello, I am&quot;</span><span class="p">,</span>
<span class="s2">&quot;The president of the United States is&quot;</span><span class="p">,</span>
<span class="s2">&quot;The capital of France is&quot;</span><span class="p">,</span>
<span class="s2">&quot;The future of AI is&quot;</span><span class="p">,</span>
<span class="p">]</span>
<span class="n">sampling_params</span> <span class="o">=</span> <span class="n">SamplingParams</span><span class="p">(</span><span class="n">temperature</span><span class="o">=</span><span class="mf">0.8</span><span class="p">,</span> <span class="n">top_p</span><span class="o">=</span><span class="mf">0.95</span><span class="p">)</span>
<span class="n">scheduler_config</span> <span class="o">=</span> <span class="n">SchedulerConfig</span><span class="p">(</span>
<span class="n">capacity_scheduler_policy</span><span class="o">=</span><span class="n">CapacitySchedulerPolicy</span><span class="o">.</span><span class="n">MAX_UTILIZATION</span>
<span class="p">)</span>
<span class="n">llm</span> <span class="o">=</span> <span class="n">LLM</span><span class="p">(</span>
<span class="n">model</span><span class="o">=</span><span class="s2">&quot;meta-llama/Llama-3.3-70B-Instruct&quot;</span><span class="p">,</span>
<span class="n">tensor_parallel_size</span><span class="o">=</span><span class="mi">4</span><span class="p">,</span>
<span class="n">scheduler_config</span><span class="o">=</span><span class="n">scheduler_config</span>
<span class="p">)</span>
<span class="n">outputs</span> <span class="o">=</span> <span class="n">llm</span><span class="o">.</span><span class="n">generate</span><span class="p">(</span><span class="n">prompts</span><span class="p">,</span> <span class="n">sampling_params</span><span class="p">)</span>
<span class="c1"># Print the outputs.</span>
<span class="k">for</span> <span class="n">output</span> <span class="ow">in</span> <span class="n">outputs</span><span class="p">:</span>
<span class="n">prompt</span> <span class="o">=</span> <span class="n">output</span><span class="o">.</span><span class="n">prompt</span>
<span class="n">generated_text</span> <span class="o">=</span> <span class="n">output</span><span class="o">.</span><span class="n">outputs</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">text</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;Prompt: </span><span class="si">{</span><span class="n">prompt</span><span class="si">!r}</span><span class="s2">, Generated text: </span><span class="si">{</span><span class="n">generated_text</span><span class="si">!r}</span><span class="s2">&quot;</span><span class="p">)</span>
<span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s1">&#39;__main__&#39;</span><span class="p">:</span>
<span class="n">main</span><span class="p">()</span>
</pre></div>
</div>
</section>
<section id="context-chunking-policy">
<h2>Context Chunking Policy<a class="headerlink" href="#context-chunking-policy" title="Link to this heading"></a></h2>
<p>As discussed <a class="reference internal" href="tuning-max-batch-size-and-max-num-tokens.html#revisiting-paged-context-attention-and-context-chunking"><span class="std std-ref">previously</span></a> context chunking will increase the chance of batch processing between
the context and the generation phase, thereby balancing the calculation amount
of each iteration and typically increasing throughput.</p>
<p>TensorRT-LLM currently supports two context chunking policies: <code class="docutils literal notranslate"><span class="pre">FIRST_COME_FIRST_SERVED</span></code> (default) which would prioritize scheduling all the context chunks of a request that comes in first,
and <code class="docutils literal notranslate"><span class="pre">EQUAL_PROGRESS</span></code> which schedules context chunks from all requests before scheduling the next chunk of any request.</p>
<p><code class="docutils literal notranslate"><span class="pre">FIRST_COME_FIRST_SERVED</span></code> should achieve overall better performance, while
<code class="docutils literal notranslate"><span class="pre">EQUAL_PROGRESS</span></code> can be helpful in theory to make sure time to first token (TTFT)
for most requests are relatively similar.</p>
<p>To switch the context chunking policy from the default of <code class="docutils literal notranslate"><span class="pre">FIRST_COME_FIRST_SERVED</span></code> to <code class="docutils literal notranslate"><span class="pre">EQUAL_PROGRESS</span></code>
you would modify the <a class="reference internal" href="benchmarking-default-performance.html#before-you-begin-tensorrt-llm-llm-api"><span class="std std-ref">LLM-API end-to-end example</span></a> to be:</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">from</span><span class="w"> </span><span class="nn">tensorrt_llm</span><span class="w"> </span><span class="kn">import</span> <span class="n">LLM</span><span class="p">,</span> <span class="n">SamplingParams</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">tensorrt_llm.bindings.executor</span><span class="w"> </span><span class="kn">import</span> <span class="n">SchedulerConfig</span><span class="p">,</span> <span class="n">ContextChunkingPolicy</span>
<span class="k">def</span><span class="w"> </span><span class="nf">main</span><span class="p">():</span>
<span class="n">prompts</span> <span class="o">=</span> <span class="p">[</span>
<span class="s2">&quot;Hello, I am&quot;</span><span class="p">,</span>
<span class="s2">&quot;The president of the United States is&quot;</span><span class="p">,</span>
<span class="s2">&quot;The capital of France is&quot;</span><span class="p">,</span>
<span class="s2">&quot;The future of AI is&quot;</span><span class="p">,</span>
<span class="p">]</span>
<span class="n">sampling_params</span> <span class="o">=</span> <span class="n">SamplingParams</span><span class="p">(</span><span class="n">temperature</span><span class="o">=</span><span class="mf">0.8</span><span class="p">,</span> <span class="n">top_p</span><span class="o">=</span><span class="mf">0.95</span><span class="p">)</span>
<span class="n">scheduler_config</span> <span class="o">=</span> <span class="n">SchedulerConfig</span><span class="p">(</span>
<span class="n">context_chunking_policy</span><span class="o">=</span><span class="n">ContextChunkingPolicy</span><span class="o">.</span><span class="n">EQUAL_PROGRESS</span>
<span class="p">)</span>
<span class="n">llm</span> <span class="o">=</span> <span class="n">LLM</span><span class="p">(</span>
<span class="n">model</span><span class="o">=</span><span class="s2">&quot;meta-llama/Llama-3.3-70B-Instruct&quot;</span><span class="p">,</span>
<span class="n">tensor_parallel_size</span><span class="o">=</span><span class="mi">4</span><span class="p">,</span>
<span class="n">scheduler_config</span><span class="o">=</span><span class="n">scheduler_config</span>
<span class="p">)</span>
<span class="n">outputs</span> <span class="o">=</span> <span class="n">llm</span><span class="o">.</span><span class="n">generate</span><span class="p">(</span><span class="n">prompts</span><span class="p">,</span> <span class="n">sampling_params</span><span class="p">)</span>
<span class="c1"># Print the outputs.</span>
<span class="k">for</span> <span class="n">output</span> <span class="ow">in</span> <span class="n">outputs</span><span class="p">:</span>
<span class="n">prompt</span> <span class="o">=</span> <span class="n">output</span><span class="o">.</span><span class="n">prompt</span>
<span class="n">generated_text</span> <span class="o">=</span> <span class="n">output</span><span class="o">.</span><span class="n">outputs</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">text</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;Prompt: </span><span class="si">{</span><span class="n">prompt</span><span class="si">!r}</span><span class="s2">, Generated text: </span><span class="si">{</span><span class="n">generated_text</span><span class="si">!r}</span><span class="s2">&quot;</span><span class="p">)</span>
<span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s1">&#39;__main__&#39;</span><span class="p">:</span>
<span class="n">main</span><span class="p">()</span>
</pre></div>
</div>
</section>
<section id="max-tokens-in-paged-kv-cache-and-kv-cache-free-gpu-memory-fraction">
<h2>Max Tokens in Paged KV Cache and KV Cache Free GPU Memory Fraction<a class="headerlink" href="#max-tokens-in-paged-kv-cache-and-kv-cache-free-gpu-memory-fraction" title="Link to this heading"></a></h2>
<p>The <code class="docutils literal notranslate"><span class="pre">max_tokens_in_paged_kv_cache</span></code> and <code class="docutils literal notranslate"><span class="pre">kv_cache_free_gpu_mem_fraction</span></code>
parameters can be used to control the maximum number of tokens handled by the
KV cache manager. Setting them properly helps better control the amount of
available memory for the KV cache manager during inference. Keeping in mind
that increasing the amount of memory available to the KV cache manager tends to
translate to a higher achievable throughput.</p>
<p>The <code class="docutils literal notranslate"><span class="pre">max_tokens_in_paged_kv_cache</span></code> flag directly sets the maximum number of
tokens in the KV cache manager. When left unset, that value will be computed
based on the <code class="docutils literal notranslate"><span class="pre">kv_cache_free_gpu_mem_fraction</span></code> setting.</p>
<p>The <code class="docutils literal notranslate"><span class="pre">kv_cache_free_gpu_mem_fraction</span></code> is a floating-point number between <code class="docutils literal notranslate"><span class="pre">0.0</span></code>
and <code class="docutils literal notranslate"><span class="pre">1.0</span></code> that indicates the maximum fraction of GPU memory (after loading the
model) that will be used for the KV cache. The default value is <code class="docutils literal notranslate"><span class="pre">0.90</span></code> and
means that 90% of the free GPU memory will be used to save tokens in the KV
cache. Based on that value, TensorRT-LLM can determine the maximum number of
tokens in the KV cache manager.</p>
<p>When both parameters are set, the maximum number of tokens in the KV cache
manager will be set to the smaller value between <code class="docutils literal notranslate"><span class="pre">max_tokens_in_paged_kv_cache</span></code>
and the value computed from the amount of memory available for the KV cache.</p>
<p>Unless users clearly know the maximum number of tokens in the KV cache needed
by the model, it is recommended to leave <code class="docutils literal notranslate"><span class="pre">max_tokens_in_paged_kv_cache</span></code> unset.
For <code class="docutils literal notranslate"><span class="pre">kv_cache_free_gpu_mem_fraction</span></code>, if no other programs are executed on the
same GPU, it is recommended to test with a as high value as <code class="docutils literal notranslate"><span class="pre">0.95</span></code> to target a
high throughput. Note that the <code class="docutils literal notranslate"><span class="pre">kv_cache_free_gpu_mem_fraction</span></code> parameter
cannot be set to <code class="docutils literal notranslate"><span class="pre">1.0</span></code> because some amount of memory has to be reserved for
inputs and outputs.</p>
<p>To set <code class="docutils literal notranslate"><span class="pre">kv_cache_free_gpu_mem_fraction</span></code> you would modify the <a class="reference internal" href="benchmarking-default-performance.html#before-you-begin-tensorrt-llm-llm-api"><span class="std std-ref">LLM-API end-to-end example</span></a> to be:</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">from</span><span class="w"> </span><span class="nn">tensorrt_llm</span><span class="w"> </span><span class="kn">import</span> <span class="n">LLM</span><span class="p">,</span> <span class="n">SamplingParams</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">tensorrt_llm.bindings.executor</span><span class="w"> </span><span class="kn">import</span> <span class="n">KvCacheConfig</span>
<span class="k">def</span><span class="w"> </span><span class="nf">main</span><span class="p">():</span>
<span class="n">prompts</span> <span class="o">=</span> <span class="p">[</span>
<span class="s2">&quot;Hello, I am&quot;</span><span class="p">,</span>
<span class="s2">&quot;The president of the United States is&quot;</span><span class="p">,</span>
<span class="s2">&quot;The capital of France is&quot;</span><span class="p">,</span>
<span class="s2">&quot;The future of AI is&quot;</span><span class="p">,</span>
<span class="p">]</span>
<span class="n">sampling_params</span> <span class="o">=</span> <span class="n">SamplingParams</span><span class="p">(</span><span class="n">temperature</span><span class="o">=</span><span class="mf">0.8</span><span class="p">,</span> <span class="n">top_p</span><span class="o">=</span><span class="mf">0.95</span><span class="p">)</span>
<span class="n">kv_cache_config</span> <span class="o">=</span> <span class="n">KvCacheConfig</span><span class="p">(</span><span class="n">free_gpu_memory_fraction</span><span class="o">=</span><span class="mf">0.95</span><span class="p">)</span>
<span class="n">llm</span> <span class="o">=</span> <span class="n">LLM</span><span class="p">(</span>
<span class="n">model</span><span class="o">=</span><span class="s2">&quot;meta-llama/Llama-3.3-70B-Instruct&quot;</span><span class="p">,</span>
<span class="n">tensor_parallel_size</span><span class="o">=</span><span class="mi">8</span><span class="p">,</span>
<span class="n">kv_cache_config</span><span class="o">=</span><span class="n">kv_cache_config</span>
<span class="p">)</span>
<span class="n">outputs</span> <span class="o">=</span> <span class="n">llm</span><span class="o">.</span><span class="n">generate</span><span class="p">(</span><span class="n">prompts</span><span class="p">,</span> <span class="n">sampling_params</span><span class="p">)</span>
<span class="c1"># Print the outputs.</span>
<span class="k">for</span> <span class="n">output</span> <span class="ow">in</span> <span class="n">outputs</span><span class="p">:</span>
<span class="n">prompt</span> <span class="o">=</span> <span class="n">output</span><span class="o">.</span><span class="n">prompt</span>
<span class="n">generated_text</span> <span class="o">=</span> <span class="n">output</span><span class="o">.</span><span class="n">outputs</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">text</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;Prompt: </span><span class="si">{</span><span class="n">prompt</span><span class="si">!r}</span><span class="s2">, Generated text: </span><span class="si">{</span><span class="n">generated_text</span><span class="si">!r}</span><span class="s2">&quot;</span><span class="p">)</span>
<span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s1">&#39;__main__&#39;</span><span class="p">:</span>
<span class="n">main</span><span class="p">()</span>
</pre></div>
</div>
<p>If you wanted to set <code class="docutils literal notranslate"><span class="pre">max_tokens_in_paged_kv_cache</span></code> instead, you would replace <code class="docutils literal notranslate"><span class="pre">free_gpu_memory_fraction</span></code> with <code class="docutils literal notranslate"><span class="pre">max_tokens</span></code> and specify the number.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span> <span class="n">kv_cache_config</span> <span class="o">=</span> <span class="n">KvCacheConfig</span><span class="p">(</span><span class="n">max_tokens</span><span class="o">=&lt;</span><span class="n">number</span> <span class="n">of</span> <span class="n">tokens</span><span class="o">&gt;</span><span class="p">)</span>
</pre></div>
</div>
</section>
<section id="maximum-attention-window-size">
<h2>Maximum Attention Window Size<a class="headerlink" href="#maximum-attention-window-size" title="Link to this heading"></a></h2>
<p>The <code class="docutils literal notranslate"><span class="pre">max_attention_window_size</span></code> flag sets the maximum number of tokens that are
attended to in order to generate one token when using techniques like sliding window
attention. See this
<a class="reference internal" href="../advanced/gpt-attention.html#sliding-window-attention-cyclic-rolling-buffer-kv-cache"><span class="std std-ref">Document</span></a>
for more details. It defaults to the maximum sequence length
(<code class="docutils literal notranslate"><span class="pre">max_seq_len</span></code> when building the engine), which means
that the feature is disabled by default.</p>
<p>When set to a smaller value than <code class="docutils literal notranslate"><span class="pre">max_seq_len</span></code> (during
engine build), only the KV cache of the last <code class="docutils literal notranslate"><span class="pre">max_attention_window_size</span></code> tokens
will be stored. If the input sequence length at runtime exceeds the
<code class="docutils literal notranslate"><span class="pre">max_attention_window_size</span></code> value, the accuracy may start dropping, but the
runtime performance will be better (due to the reduction in terms of
computations and GPU memory allocation). Users can modify that value to
increase runtime performance at the expense of reduced accuracy.</p>
<p>Just like <a class="reference internal" href="#max-tokens-in-paged-kv-cache-and-kv-cache-free-gpu-memory-fraction"><span class="std std-ref"><code class="docutils literal notranslate"><span class="pre">kv_cache_free_gpu_mem_fraction</span></code></span></a>, <code class="docutils literal notranslate"><span class="pre">max_attention_window_size</span></code> can be specified in the LLM-API
via <code class="docutils literal notranslate"><span class="pre">KVCacheConfig</span></code>. To specify <code class="docutils literal notranslate"><span class="pre">max_attention_window_size</span></code> you would instantiate <code class="docutils literal notranslate"><span class="pre">KVCacheConfig</span></code> like so</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span> <span class="n">kv_cache_config</span> <span class="o">=</span> <span class="n">KvCacheConfig</span><span class="p">(</span><span class="n">max_attention_window</span><span class="o">=&lt;</span><span class="n">number</span> <span class="n">of</span> <span class="n">tokens</span><span class="o">&gt;</span><span class="p">)</span>
</pre></div>
</div>
</section>
</section>
</div>
</div>
<footer>
<hr/>
<div role="contentinfo">
<jinja2.runtime.BlockReference object at 0x7f6400533620>
<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