mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-14 06:27:45 +08:00
1946 lines
255 KiB
HTML
1946 lines
255 KiB
HTML
|
|
|
|
<!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.executor — 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=888ff710"></script>
|
|
<script src="../../_static/sphinx_highlight.js?v=dc90522c"></script>
|
|
<script src="../../_static/clipboard.min.js?v=a7894cd8"></script>
|
|
<script src="../../_static/copybutton.js?v=65e89d2a"></script>
|
|
<script 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.executor</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.executor</h1><div class="highlight"><pre>
|
|
<span></span><span class="kn">import</span> <span class="nn">asyncio</span>
|
|
<span class="kn">import</span> <span class="nn">atexit</span>
|
|
<span class="kn">import</span> <span class="nn">concurrent.futures</span>
|
|
<span class="kn">import</span> <span class="nn">copy</span>
|
|
<span class="kn">import</span> <span class="nn">datetime</span>
|
|
<span class="kn">import</span> <span class="nn">faulthandler</span>
|
|
<span class="kn">import</span> <span class="nn">io</span>
|
|
<span class="kn">import</span> <span class="nn">json</span>
|
|
<span class="kn">import</span> <span class="nn">multiprocessing</span>
|
|
<span class="kn">import</span> <span class="nn">os</span>
|
|
<span class="kn">import</span> <span class="nn">pickle</span> <span class="c1"># nosec B403</span>
|
|
<span class="kn">import</span> <span class="nn">signal</span>
|
|
<span class="kn">import</span> <span class="nn">time</span>
|
|
<span class="kn">import</span> <span class="nn">traceback</span>
|
|
<span class="kn">from</span> <span class="nn">abc</span> <span class="kn">import</span> <span class="n">ABC</span><span class="p">,</span> <span class="n">abstractmethod</span>
|
|
<span class="kn">from</span> <span class="nn">concurrent.futures</span> <span class="kn">import</span> <span class="n">ProcessPoolExecutor</span>
|
|
<span class="kn">from</span> <span class="nn">dataclasses</span> <span class="kn">import</span> <span class="n">dataclass</span><span class="p">,</span> <span class="n">field</span>
|
|
<span class="kn">from</span> <span class="nn">multiprocessing.shared_memory</span> <span class="kn">import</span> <span class="n">SharedMemory</span>
|
|
<span class="kn">from</span> <span class="nn">pathlib</span> <span class="kn">import</span> <span class="n">Path</span>
|
|
<span class="kn">from</span> <span class="nn">queue</span> <span class="kn">import</span> <span class="n">Empty</span><span class="p">,</span> <span class="n">Queue</span>
|
|
<span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="p">(</span><span class="n">Any</span><span class="p">,</span> <span class="n">Callable</span><span class="p">,</span> <span class="n">Dict</span><span class="p">,</span> <span class="n">Generator</span><span class="p">,</span> <span class="n">List</span><span class="p">,</span> <span class="n">Literal</span><span class="p">,</span> <span class="n">NamedTuple</span><span class="p">,</span>
|
|
<span class="n">Optional</span><span class="p">,</span> <span class="n">Tuple</span><span class="p">,</span> <span class="n">Union</span><span class="p">)</span>
|
|
<span class="kn">from</span> <span class="nn">weakref</span> <span class="kn">import</span> <span class="n">WeakMethod</span>
|
|
|
|
<span class="kn">import</span> <span class="nn">numpy</span> <span class="k">as</span> <span class="nn">np</span>
|
|
<span class="kn">import</span> <span class="nn">torch</span>
|
|
<span class="kn">import</span> <span class="nn">zmq</span>
|
|
|
|
<span class="kn">from</span> <span class="nn">tensorrt_llm.logger</span> <span class="kn">import</span> <span class="n">set_level</span>
|
|
|
|
<span class="kn">from</span> <span class="nn">._utils</span> <span class="kn">import</span> <span class="n">mpi_rank</span><span class="p">,</span> <span class="n">mpi_world_size</span>
|
|
<span class="kn">from</span> <span class="nn">.bindings</span> <span class="kn">import</span> <span class="n">executor</span> <span class="k">as</span> <span class="n">tllm</span>
|
|
<span class="kn">from</span> <span class="nn">.builder</span> <span class="kn">import</span> <span class="n">ConfigEncoder</span><span class="p">,</span> <span class="n">Engine</span><span class="p">,</span> <span class="n">EngineConfig</span>
|
|
<span class="kn">from</span> <span class="nn">.llmapi.mpi_session</span> <span class="kn">import</span> <span class="p">(</span><span class="n">MpiPoolSession</span><span class="p">,</span> <span class="n">MpiSession</span><span class="p">,</span>
|
|
<span class="n">external_mpi_comm_available</span><span class="p">,</span>
|
|
<span class="n">need_spawn_mpi_workers</span><span class="p">)</span>
|
|
<span class="kn">from</span> <span class="nn">.llmapi.tracer</span> <span class="kn">import</span> <span class="p">(</span><span class="n">VizTracer</span><span class="p">,</span> <span class="n">enable_llm_tracer</span><span class="p">,</span> <span class="n">get_tracer</span><span class="p">,</span>
|
|
<span class="n">global_tracer</span><span class="p">,</span> <span class="n">set_global_tracer</span><span class="p">)</span>
|
|
<span class="kn">from</span> <span class="nn">.llmapi.utils</span> <span class="kn">import</span> <span class="p">(</span><span class="n">AsyncQueue</span><span class="p">,</span> <span class="n">ManagedThread</span><span class="p">,</span> <span class="n">_SyncQueue</span><span class="p">,</span>
|
|
<span class="n">enable_llm_debug</span><span class="p">,</span> <span class="n">print_colored</span><span class="p">)</span>
|
|
<span class="kn">from</span> <span class="nn">.lora_manager</span> <span class="kn">import</span> <span class="n">LoraManager</span>
|
|
<span class="kn">from</span> <span class="nn">.prompt_adapter_manager</span> <span class="kn">import</span> <span class="n">PromptAdapterManager</span>
|
|
<span class="kn">from</span> <span class="nn">.runtime</span> <span class="kn">import</span> <span class="n">ModelConfig</span>
|
|
<span class="kn">from</span> <span class="nn">.runtime.model_runner</span> <span class="kn">import</span> <span class="n">_engine_config_to_model_config</span>
|
|
<span class="kn">from</span> <span class="nn">.sampling_params</span> <span class="kn">import</span> <span class="n">SamplingParams</span>
|
|
|
|
<span class="n">unblock_corountine</span> <span class="o">=</span> <span class="kc">True</span>
|
|
|
|
<span class="k">if</span> <span class="n">enable_llm_debug</span><span class="p">():</span>
|
|
<span class="c1"># Mainly enable more detailed logging from cpp runtime.</span>
|
|
<span class="n">set_level</span><span class="p">(</span><span class="s2">"info"</span><span class="p">)</span>
|
|
|
|
|
|
<span class="nd">@dataclass</span><span class="p">(</span><span class="n">slots</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
|
|
<span class="k">class</span> <span class="nc">LoRARequest</span><span class="p">:</span>
|
|
<span class="w"> </span><span class="sd">""" Request for a LoRA adapter. """</span>
|
|
<span class="n">lora_name</span><span class="p">:</span> <span class="nb">str</span>
|
|
<span class="n">lora_int_id</span><span class="p">:</span> <span class="nb">int</span>
|
|
<span class="n">lora_path</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="s2">""</span>
|
|
|
|
<span class="k">def</span> <span class="nf">__post_init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="ow">not</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">exists</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">lora_path</span><span class="p">):</span>
|
|
<span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="sa">f</span><span class="s2">"lora_path (</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">lora_path</span><span class="si">}</span><span class="s2">) does not exist."</span><span class="p">)</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span> <span class="nf">adapter_id</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_int_id</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span> <span class="nf">name</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_name</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span> <span class="nf">path</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_path</span>
|
|
|
|
|
|
<span class="nd">@dataclass</span><span class="p">(</span><span class="n">slots</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
|
|
<span class="k">class</span> <span class="nc">PromptAdapterRequest</span><span class="p">:</span>
|
|
<span class="w"> </span><span class="sd">"""</span>
|
|
<span class="sd"> Request for a Prompt adapter.</span>
|
|
<span class="sd"> """</span>
|
|
<span class="n">prompt_adapter_name</span><span class="p">:</span> <span class="nb">str</span>
|
|
<span class="n">prompt_adapter_id</span><span class="p">:</span> <span class="nb">int</span>
|
|
<span class="n">prompt_adapter_local_path</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="s2">""</span>
|
|
|
|
<span class="k">def</span> <span class="nf">__post_init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="ow">not</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">exists</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">prompt_adapter_local_path</span><span class="p">):</span>
|
|
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span>
|
|
<span class="sa">f</span><span class="s2">"prompt_adapter_local_path (</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">prompt_adapter_local_path</span><span class="si">}</span><span class="s2">) does not exist."</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span> <span class="nf">adapter_id</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">prompt_adapter_id</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span> <span class="nf">name</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">prompt_adapter_name</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span> <span class="nf">local_path</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">prompt_adapter_local_path</span>
|
|
|
|
|
|
<span class="k">class</span> <span class="nc">GenerationRequest</span><span class="p">:</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="p">,</span>
|
|
<span class="n">prompt_token_ids</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">np</span><span class="o">.</span><span class="n">ndarray</span><span class="p">,</span>
|
|
<span class="n">Union</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">],</span> <span class="n">List</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]]]],</span>
|
|
<span class="n">sampling_params</span><span class="p">:</span> <span class="n">SamplingParams</span><span class="p">,</span>
|
|
<span class="n">query_token_ids</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">np</span><span class="o">.</span><span class="n">ndarray</span><span class="p">,</span> <span class="nb">list</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">lora_request</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">LoRARequest</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">prompt_adapter_request</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">PromptAdapterRequest</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">streaming</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">prompt_tuning_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">list</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="p">):</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">prompt_token_ids</span><span class="p">,</span> <span class="nb">list</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">prompt_token_ids</span> <span class="o">=</span> <span class="n">prompt_token_ids</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">query_token_ids</span> <span class="o">=</span> <span class="n">query_token_ids</span>
|
|
<span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">prompt_token_ids</span><span class="p">,</span> <span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">np</span><span class="o">.</span><span class="n">ndarray</span><span class="p">)):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">prompt_token_ids</span> <span class="o">=</span> <span class="n">prompt_token_ids</span><span class="o">.</span><span class="n">tolist</span><span class="p">()</span>
|
|
<span class="k">if</span> <span class="n">query_token_ids</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">query_token_ids</span> <span class="o">=</span> <span class="n">query_token_ids</span><span class="o">.</span><span class="n">tolist</span><span class="p">()</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="k">raise</span> <span class="ne">TypeError</span><span class="p">(</span>
|
|
<span class="sa">f</span><span class="s2">"prompt_token_ids (</span><span class="si">{</span><span class="n">prompt_token_ids</span><span class="si">}</span><span class="s2">) should be an instance of torch.Tensor, np.ndarray or list"</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">sampling_params</span> <span class="o">=</span> <span class="n">sampling_params</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">lora_request</span> <span class="o">=</span> <span class="n">lora_request</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">prompt_adapter_request</span> <span class="o">=</span> <span class="n">prompt_adapter_request</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">streaming</span> <span class="o">=</span> <span class="n">streaming</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">prompt_tuning_config</span> <span class="o">=</span> <span class="n">prompt_tuning_config</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">id</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="k">def</span> <span class="nf">set_id</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="nb">id</span><span class="p">):</span>
|
|
<span class="k">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">id</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">,</span> <span class="sa">f</span><span class="s2">"Request ID is already set: </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">id</span><span class="si">}</span><span class="s2">"</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">id</span> <span class="o">=</span> <span class="nb">id</span>
|
|
<span class="k">return</span> <span class="bp">self</span>
|
|
|
|
|
|
<span class="nd">@dataclass</span><span class="p">(</span><span class="n">slots</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
|
|
<span class="k">class</span> <span class="nc">CompletionOutput</span><span class="p">:</span>
|
|
<span class="w"> </span><span class="sd">"""The output data of one completion output of a request.</span>
|
|
|
|
<span class="sd"> Args:</span>
|
|
<span class="sd"> index (int): The index of the output in the request.</span>
|
|
<span class="sd"> text (str): The generated output text. Defaults to "".</span>
|
|
<span class="sd"> token_ids (List[int]): The token ids of the generated output text. Defaults to [].</span>
|
|
<span class="sd"> cumulative_logprob (float, optional): The cumulative log probability of the generated output text. Defaults to None.</span>
|
|
<span class="sd"> logprobs (List[float]): The log probabilities of the top probability words at each position if the logprobs are requested. Defaults to [].</span>
|
|
<span class="sd"> finish_reason (Literal['stop', 'length'], optional): The reason why the sequence is finished. Defaults to None.</span>
|
|
<span class="sd"> stop_reason (int, str, optional): The stop string or token id that caused the completion to stop, None if the completion finished for some other reason. Defaults to None.</span>
|
|
<span class="sd"> generation_logits (torch.Tensor, optional): The logits on the generated output token ids. Defaults to None.</span>
|
|
|
|
<span class="sd"> Properties:</span>
|
|
<span class="sd"> length (int): The number of generated tokens.</span>
|
|
<span class="sd"> token_ids_diff (List[int]): Newly generated token ids.</span>
|
|
<span class="sd"> logprobs_diff (List[float]): Logprobs of newly generated tokens.</span>
|
|
<span class="sd"> text_diff (str): Newly generated tokens.</span>
|
|
<span class="sd"> """</span>
|
|
<span class="n">index</span><span class="p">:</span> <span class="nb">int</span>
|
|
<span class="n">text</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="s2">""</span>
|
|
<span class="n">token_ids</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span><span class="n">default_factory</span><span class="o">=</span><span class="nb">list</span><span class="p">)</span>
|
|
<span class="n">cumulative_logprob</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">float</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="n">logprobs</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="nb">float</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span><span class="n">default_factory</span><span class="o">=</span><span class="nb">list</span><span class="p">)</span>
|
|
<span class="n">finish_reason</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Literal</span><span class="p">[</span><span class="s1">'stop'</span><span class="p">,</span> <span class="s1">'length'</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="n">stop_reason</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span><span class="nb">int</span><span class="p">,</span> <span class="nb">str</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="n">generation_logits</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="c1"># hidden fields for tracking the diffs</span>
|
|
<span class="n">_last_text_len</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span> <span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="nb">repr</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
|
|
<span class="n">_last_token_ids_len</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span> <span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="nb">repr</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
|
|
<span class="n">_last_logprobs_len</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span> <span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="nb">repr</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
|
|
<span class="n">_incremental_states</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">dict</span><span class="p">]</span> <span class="o">=</span> <span class="n">field</span><span class="p">(</span><span class="n">default</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">init</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="nb">repr</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span> <span class="nf">length</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="nb">len</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">token_ids</span><span class="p">)</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span> <span class="nf">text_diff</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="nb">str</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">text</span><span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">_last_text_len</span><span class="p">:]</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span> <span class="nf">token_ids_diff</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]:</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">token_ids</span><span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">_last_token_ids_len</span><span class="p">:]</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span> <span class="nf">logprobs_diff</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="n">List</span><span class="p">[</span><span class="nb">float</span><span class="p">]:</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">logprobs</span><span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">_last_logprobs_len</span><span class="p">:]</span>
|
|
|
|
|
|
<span class="k">class</span> <span class="nc">CppExecutorError</span><span class="p">(</span><span class="ne">RuntimeError</span><span class="p">):</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">message</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">message</span> <span class="o">=</span> <span class="n">message</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">stack_trace</span> <span class="o">=</span> <span class="n">traceback</span><span class="o">.</span><span class="n">format_exc</span><span class="p">()</span>
|
|
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">(</span><span class="n">message</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__str__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">message</span><span class="si">}</span><span class="se">\n</span><span class="s2">Stack trace:</span><span class="se">\n</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">stack_trace</span><span class="si">}</span><span class="s2">"</span>
|
|
|
|
|
|
<div class="viewcode-block" id="RequestError">
|
|
<a class="viewcode-back" href="../../llm-api/reference.html#tensorrt_llm.llmapi.RequestError">[docs]</a>
|
|
<span class="k">class</span> <span class="nc">RequestError</span><span class="p">(</span><span class="ne">RuntimeError</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">''' The error raised when the request is failed. '''</span></div>
|
|
|
|
|
|
|
|
<span class="k">class</span> <span class="nc">GenerationResult</span><span class="p">:</span>
|
|
<span class="w"> </span><span class="sd">'''</span>
|
|
<span class="sd"> The result of a generation request. It can be used to wait for the completion of the request.</span>
|
|
|
|
<span class="sd"> Args:</span>
|
|
<span class="sd"> generation_request (GenerationRequest): The generation request object.</span>
|
|
<span class="sd"> background_error_handler (Callable, optional): The error handler to process the errors from the background threads/processes. Defaults to None.</span>
|
|
<span class="sd"> '''</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span>
|
|
<span class="n">generation_request</span><span class="p">:</span> <span class="n">GenerationRequest</span><span class="p">,</span>
|
|
<span class="n">background_error_handler</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Callable</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_done</span> <span class="o">=</span> <span class="kc">False</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_cancelled</span> <span class="o">=</span> <span class="kc">False</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_generation_request</span> <span class="o">=</span> <span class="n">generation_request</span>
|
|
|
|
<span class="k">if</span> <span class="n">has_event_loop</span><span class="p">():</span>
|
|
<span class="n">aqueue</span> <span class="o">=</span> <span class="n">AsyncQueue</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">queue</span> <span class="o">=</span> <span class="n">aqueue</span><span class="o">.</span><span class="n">sync_q</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">aqueue</span> <span class="o">=</span> <span class="n">aqueue</span><span class="o">.</span><span class="n">async_q</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">queue</span> <span class="o">=</span> <span class="n">Queue</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">aqueue</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="c1"># In Sampling mode, the Executor runtime will return best_of sequences</span>
|
|
<span class="c1"># in total, which the LLM API will select the n-best sequences among</span>
|
|
<span class="c1"># them based on their cumulative log probabilities.</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_outputs</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="n">CompletionOutput</span><span class="p">]</span> <span class="o">=</span> <span class="p">[</span>
|
|
<span class="n">CompletionOutput</span><span class="p">(</span><span class="n">i</span><span class="p">)</span>
|
|
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_generation_request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">best_of</span><span class="p">)</span>
|
|
<span class="p">]</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">context_logits</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_background_error_handler</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="k">if</span> <span class="n">background_error_handler</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_background_error_handler</span> <span class="o">=</span> <span class="n">WeakMethod</span><span class="p">(</span>
|
|
<span class="n">background_error_handler</span><span class="p">)</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span> <span class="nf">request_id</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="nb">int</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_generation_request</span><span class="o">.</span><span class="n">id</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span> <span class="nf">prompt_token_ids</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]:</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_generation_request</span><span class="o">.</span><span class="n">prompt_token_ids</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span> <span class="nf">finished</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_done</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span> <span class="nf">streaming</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_generation_request</span><span class="o">.</span><span class="n">streaming</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span> <span class="nf">outputs</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="n">List</span><span class="p">[</span><span class="n">CompletionOutput</span><span class="p">]:</span>
|
|
<span class="n">sampling_param</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_generation_request</span><span class="o">.</span><span class="n">sampling_params</span>
|
|
<span class="k">if</span> <span class="p">(</span><span class="n">sampling_param</span><span class="o">.</span><span class="n">use_beam_search</span>
|
|
<span class="ow">or</span> <span class="n">sampling_param</span><span class="o">.</span><span class="n">n</span> <span class="o">==</span> <span class="n">sampling_param</span><span class="o">.</span><span class="n">best_of</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_outputs</span><span class="p">[:</span><span class="n">sampling_param</span><span class="o">.</span><span class="n">n</span><span class="p">]</span>
|
|
<span class="c1"># Pick the top-n outputs, sorted by cumulative log probs.</span>
|
|
<span class="n">sorted_outputs</span> <span class="o">=</span> <span class="nb">sorted</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_outputs</span><span class="p">,</span>
|
|
<span class="n">key</span><span class="o">=</span><span class="k">lambda</span> <span class="n">x</span><span class="p">:</span>
|
|
<span class="p">(</span><span class="n">x</span><span class="o">.</span><span class="n">cumulative_logprob</span>
|
|
<span class="k">if</span> <span class="n">x</span><span class="o">.</span><span class="n">cumulative_logprob</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="k">else</span> <span class="nb">float</span><span class="p">(</span><span class="s1">'-inf'</span><span class="p">)),</span>
|
|
<span class="n">reverse</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
|
|
<span class="c1"># Reindex the sequence.</span>
|
|
<span class="k">for</span> <span class="n">i</span><span class="p">,</span> <span class="n">sorted_out</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">sorted_outputs</span><span class="p">):</span>
|
|
<span class="n">sorted_out</span><span class="o">.</span><span class="n">index</span> <span class="o">=</span> <span class="n">i</span>
|
|
<span class="k">return</span> <span class="n">sorted_outputs</span><span class="p">[:</span><span class="n">sampling_param</span><span class="o">.</span><span class="n">n</span><span class="p">]</span>
|
|
|
|
<span class="k">def</span> <span class="nf">handle_sequence</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">response</span><span class="p">:</span> <span class="s2">"GenerationExecutor.Response"</span><span class="p">,</span>
|
|
<span class="n">sequence_index</span><span class="p">:</span> <span class="nb">int</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">""" Handle a single sequence in the response. """</span>
|
|
|
|
<span class="n">tensors</span> <span class="o">=</span> <span class="n">response</span><span class="o">.</span><span class="n">tensors</span>
|
|
<span class="k">assert</span> <span class="n">tensors</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
|
|
|
|
<span class="n">beam_search</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_generation_request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">use_beam_search</span>
|
|
<span class="n">seq_idx</span> <span class="o">=</span> <span class="n">sequence_index</span>
|
|
<span class="n">src_idx</span> <span class="o">=</span> <span class="n">sequence_index</span> <span class="k">if</span> <span class="n">beam_search</span> <span class="k">else</span> <span class="mi">0</span>
|
|
|
|
<span class="n">output</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_outputs</span><span class="p">[</span><span class="n">seq_idx</span><span class="p">]</span>
|
|
|
|
<span class="n">output</span><span class="o">.</span><span class="n">_last_token_ids_len</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="n">output</span><span class="o">.</span><span class="n">token_ids</span><span class="p">)</span>
|
|
<span class="n">output</span><span class="o">.</span><span class="n">token_ids</span><span class="o">.</span><span class="n">extend</span><span class="p">(</span><span class="n">tensors</span><span class="o">.</span><span class="n">output_token_ids</span><span class="p">[</span><span class="n">src_idx</span><span class="p">])</span>
|
|
<span class="k">if</span> <span class="n">tensors</span><span class="o">.</span><span class="n">cum_log_probs</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="n">output</span><span class="o">.</span><span class="n">cumulative_logprob</span> <span class="o">=</span> <span class="n">tensors</span><span class="o">.</span><span class="n">cum_log_probs</span><span class="p">[</span><span class="n">src_idx</span><span class="p">]</span>
|
|
<span class="k">if</span> <span class="n">tensors</span><span class="o">.</span><span class="n">log_probs</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="n">output</span><span class="o">.</span><span class="n">_last_logprobs_len</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="n">output</span><span class="o">.</span><span class="n">logprobs</span><span class="p">)</span>
|
|
<span class="n">output</span><span class="o">.</span><span class="n">logprobs</span> <span class="o">=</span> <span class="n">tensors</span><span class="o">.</span><span class="n">log_probs</span><span class="p">[</span><span class="n">src_idx</span><span class="p">]</span>
|
|
<span class="k">assert</span> <span class="nb">len</span><span class="p">(</span><span class="n">output</span><span class="o">.</span><span class="n">logprobs</span><span class="p">)</span> <span class="o">==</span> <span class="n">output</span><span class="o">.</span><span class="n">length</span>
|
|
<span class="k">if</span> <span class="n">tensors</span><span class="o">.</span><span class="n">generation_logits</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="n">output</span><span class="o">.</span><span class="n">generation_logits</span> <span class="o">=</span> <span class="n">tensors</span><span class="o">.</span><span class="n">generation_logits</span><span class="p">[</span>
|
|
<span class="n">src_idx</span><span class="p">,</span> <span class="p">:</span><span class="n">output</span><span class="o">.</span><span class="n">length</span><span class="p">]</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">finished</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="n">response</span><span class="o">.</span><span class="n">finish_reasons</span><span class="p">[</span><span class="n">src_idx</span><span class="p">]</span> <span class="o">==</span> <span class="n">tllm</span><span class="o">.</span><span class="n">FinishReason</span><span class="o">.</span><span class="n">END_ID</span><span class="p">:</span>
|
|
<span class="n">output</span><span class="o">.</span><span class="n">finish_reason</span> <span class="o">=</span> <span class="s1">'stop'</span>
|
|
<span class="k">elif</span> <span class="n">response</span><span class="o">.</span><span class="n">finish_reasons</span><span class="p">[</span>
|
|
<span class="n">src_idx</span><span class="p">]</span> <span class="o">==</span> <span class="n">tllm</span><span class="o">.</span><span class="n">FinishReason</span><span class="o">.</span><span class="n">STOP_WORDS</span><span class="p">:</span>
|
|
<span class="n">output</span><span class="o">.</span><span class="n">finish_reason</span> <span class="o">=</span> <span class="s1">'stop'</span>
|
|
<span class="n">sampling_params</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_generation_request</span><span class="o">.</span><span class="n">sampling_params</span>
|
|
<span class="k">for</span> <span class="n">stop_reason</span><span class="p">,</span> <span class="n">stop_ids</span> <span class="ow">in</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">_get_stop_reasons_and_words</span><span class="p">(</span>
|
|
<span class="p">):</span>
|
|
<span class="k">if</span> <span class="n">output</span><span class="o">.</span><span class="n">token_ids</span><span class="p">[</span><span class="o">-</span><span class="nb">len</span><span class="p">(</span><span class="n">stop_ids</span><span class="p">):]</span> <span class="o">==</span> <span class="n">stop_ids</span><span class="p">:</span>
|
|
<span class="n">output</span><span class="o">.</span><span class="n">stop_reason</span> <span class="o">=</span> <span class="n">stop_reason</span>
|
|
<span class="k">if</span> <span class="ow">not</span> <span class="n">sampling_params</span><span class="o">.</span><span class="n">include_stop_str_in_output</span><span class="p">:</span>
|
|
<span class="n">output</span><span class="o">.</span><span class="n">token_ids</span> <span class="o">=</span> <span class="n">output</span><span class="o">.</span><span class="n">token_ids</span><span class="p">[:</span><span class="o">-</span><span class="nb">len</span><span class="p">(</span><span class="n">stop_ids</span><span class="p">)]</span>
|
|
<span class="k">break</span>
|
|
<span class="k">elif</span> <span class="n">response</span><span class="o">.</span><span class="n">finish_reasons</span><span class="p">[</span><span class="n">src_idx</span><span class="p">]</span> <span class="o">==</span> <span class="n">tllm</span><span class="o">.</span><span class="n">FinishReason</span><span class="o">.</span><span class="n">LENGTH</span><span class="p">:</span>
|
|
<span class="n">output</span><span class="o">.</span><span class="n">finish_reason</span> <span class="o">=</span> <span class="s1">'length'</span>
|
|
|
|
<span class="k">def</span> <span class="nf">handle_response</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">response</span><span class="p">:</span> <span class="s2">"GenerationExecutor.Response"</span><span class="p">):</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_done</span> <span class="o">=</span> <span class="n">response</span><span class="o">.</span><span class="n">is_final</span>
|
|
|
|
<span class="k">if</span> <span class="n">response</span><span class="o">.</span><span class="n">error</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="n">handler</span> <span class="o">:=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_background_error_handler</span><span class="p">():</span>
|
|
<span class="n">handler</span><span class="p">(</span><span class="n">response</span><span class="o">.</span><span class="n">error</span><span class="p">)</span>
|
|
|
|
<span class="n">tensors</span> <span class="o">=</span> <span class="n">response</span><span class="o">.</span><span class="n">tensors</span>
|
|
|
|
<span class="c1"># output_token_ids = (beams, tokens)</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_generation_request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">use_beam_search</span><span class="p">:</span>
|
|
<span class="k">for</span> <span class="n">beam_idx</span><span class="p">,</span> <span class="n">_</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">tensors</span><span class="o">.</span><span class="n">output_token_ids</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">handle_sequence</span><span class="p">(</span><span class="n">response</span><span class="p">,</span> <span class="n">beam_idx</span><span class="p">)</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">handle_sequence</span><span class="p">(</span><span class="n">response</span><span class="p">,</span> <span class="n">response</span><span class="o">.</span><span class="n">sequence_index</span><span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="n">tensors</span><span class="o">.</span><span class="n">context_logits</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">context_logits</span> <span class="o">=</span> <span class="n">tensors</span><span class="o">.</span><span class="n">context_logits</span>
|
|
|
|
<span class="c1"># Processing background errors here ASAF during generation.</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_background_error_handler</span> <span class="ow">and</span> <span class="p">(</span>
|
|
<span class="n">handler</span> <span class="o">:=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_background_error_handler</span><span class="p">()):</span>
|
|
<span class="n">handler</span><span class="p">()</span>
|
|
|
|
<span class="k">def</span> <span class="nf">result_step</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">timeout</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">float</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span>
|
|
<span class="n">response</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">queue</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="n">timeout</span><span class="o">=</span><span class="n">timeout</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">handle_response</span><span class="p">(</span><span class="n">response</span><span class="p">)</span>
|
|
|
|
<span class="k">async</span> <span class="k">def</span> <span class="nf">aresult_step</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">aqueue</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">,</span> <span class="s2">"The asyncio event loop was not present during initialization, so async operations are not available."</span>
|
|
<span class="n">response</span> <span class="o">=</span> <span class="k">await</span> <span class="bp">self</span><span class="o">.</span><span class="n">aqueue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
|
|
<span class="n">global_tracer</span><span class="p">()</span><span class="o">.</span><span class="n">log_instant</span><span class="p">(</span><span class="s2">"result_step.get"</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">handle_response</span><span class="p">(</span><span class="n">response</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">result</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">timeout</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">float</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">)</span> <span class="o">-></span> <span class="s2">"GenerationResult"</span><span class="p">:</span>
|
|
<span class="k">while</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">_done</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">result_step</span><span class="p">(</span><span class="n">timeout</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="bp">self</span>
|
|
|
|
<span class="k">async</span> <span class="k">def</span> <span class="nf">aresult</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="s2">"GenerationResult"</span><span class="p">:</span>
|
|
<span class="k">while</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">_done</span><span class="p">:</span>
|
|
<span class="k">await</span> <span class="bp">self</span><span class="o">.</span><span class="n">aresult_step</span><span class="p">()</span>
|
|
<span class="k">return</span> <span class="bp">self</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__await__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">aresult</span><span class="p">()</span><span class="o">.</span><span class="fm">__await__</span><span class="p">()</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__iter__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="bp">self</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__next__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_done</span><span class="p">:</span>
|
|
<span class="k">raise</span> <span class="ne">StopIteration</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">result_step</span><span class="p">()</span>
|
|
<span class="k">return</span> <span class="bp">self</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__aiter__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="bp">self</span>
|
|
|
|
<span class="k">async</span> <span class="k">def</span> <span class="fm">__anext__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_done</span><span class="p">:</span>
|
|
<span class="k">raise</span> <span class="ne">StopAsyncIteration</span>
|
|
|
|
<span class="k">await</span> <span class="bp">self</span><span class="o">.</span><span class="n">aresult_step</span><span class="p">()</span>
|
|
<span class="k">return</span> <span class="bp">self</span>
|
|
|
|
<span class="k">def</span> <span class="nf">running</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">_done</span>
|
|
|
|
<span class="k">def</span> <span class="nf">cancelled</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_cancelled</span>
|
|
|
|
<span class="k">def</span> <span class="nf">cancel</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">raise</span> <span class="ne">NotImplementedError</span>
|
|
|
|
<span class="k">def</span> <span class="nf">done</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_done</span>
|
|
|
|
<span class="k">def</span> <span class="nf">exception</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">timeout</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">float</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span>
|
|
<span class="k">try</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">result</span><span class="p">(</span><span class="n">timeout</span><span class="p">)</span>
|
|
<span class="k">except</span> <span class="ne">RuntimeError</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="n">e</span>
|
|
|
|
<span class="k">def</span> <span class="nf">_repr_fields</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="p">[</span>
|
|
<span class="s1">'request_id'</span><span class="p">,</span> <span class="s1">'prompt_token_ids'</span><span class="p">,</span> <span class="s1">'outputs'</span><span class="p">,</span> <span class="s1">'finished'</span><span class="p">,</span>
|
|
<span class="s2">"context_logits"</span>
|
|
<span class="p">]</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__repr__</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="nb">str</span><span class="p">:</span>
|
|
<span class="nb">repr</span> <span class="o">=</span> <span class="p">[]</span>
|
|
<span class="k">for</span> <span class="n">field</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">_repr_fields</span><span class="p">():</span>
|
|
<span class="n">value</span> <span class="o">=</span> <span class="nb">getattr</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">field</span><span class="p">)</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">value</span><span class="p">,</span> <span class="nb">str</span><span class="p">):</span>
|
|
<span class="nb">repr</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="n">field</span><span class="si">}</span><span class="s2">=</span><span class="si">{</span><span class="n">value</span><span class="si">!r}</span><span class="s2">"</span><span class="p">)</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="nb">repr</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="n">field</span><span class="si">}</span><span class="s2">=</span><span class="si">{</span><span class="n">value</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span>
|
|
<span class="nb">repr</span> <span class="o">=</span> <span class="s2">", "</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="nb">repr</span><span class="p">)</span>
|
|
<span class="nb">repr</span> <span class="o">=</span> <span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="vm">__class__</span><span class="o">.</span><span class="vm">__name__</span><span class="si">}</span><span class="s2">(</span><span class="si">{</span><span class="nb">repr</span><span class="si">}</span><span class="s2">)"</span>
|
|
<span class="k">return</span> <span class="nb">repr</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__hash__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="nb">hash</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">request_id</span><span class="p">)</span>
|
|
|
|
|
|
<div class="viewcode-block" id="NoStatsAvailable">
|
|
<a class="viewcode-back" href="../../llm-api/reference.html#tensorrt_llm.llmapi.NoStatsAvailable">[docs]</a>
|
|
<span class="k">class</span> <span class="nc">NoStatsAvailable</span><span class="p">(</span><span class="ne">Exception</span><span class="p">):</span>
|
|
<span class="k">pass</span></div>
|
|
|
|
|
|
|
|
<span class="k">class</span> <span class="nc">GenerationExecutor</span><span class="p">(</span><span class="n">ABC</span><span class="p">):</span>
|
|
|
|
<span class="k">class</span> <span class="nc">ResponseTensors</span><span class="p">(</span><span class="n">NamedTuple</span><span class="p">):</span>
|
|
<span class="n">output_token_ids</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]]</span>
|
|
<span class="c1"># context_logits is a tensor or a string denoting the path to the shared memory.</span>
|
|
<span class="n">context_logits</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span> <span class="o">|</span> <span class="nb">str</span><span class="p">]</span>
|
|
<span class="c1"># generation_logits is a tensor or a string denoting the path to the shared memory.</span>
|
|
<span class="n">generation_logits</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span> <span class="o">|</span> <span class="nb">str</span><span class="p">]</span>
|
|
<span class="n">log_probs</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">list</span><span class="p">]</span>
|
|
<span class="n">cum_log_probs</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">list</span><span class="p">]</span>
|
|
|
|
<span class="k">class</span> <span class="nc">Response</span><span class="p">(</span><span class="n">NamedTuple</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">""" The response from the cpp-executor to the Python main thread. """</span>
|
|
<span class="n">client_id</span><span class="p">:</span> <span class="nb">int</span>
|
|
<span class="n">tensors</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="s2">"GenerationExecutor.ResponseTensors"</span><span class="p">]</span>
|
|
<span class="n">finish_reasons</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="n">tllm</span><span class="o">.</span><span class="n">FinishReason</span><span class="p">]]</span>
|
|
<span class="n">is_final</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">bool</span><span class="p">]</span>
|
|
<span class="n">sequence_index</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span>
|
|
<span class="c1"># There are two types of errors:</span>
|
|
<span class="c1"># 1. str for the errors from the cpp-executor.await_responses, this will be dispatched to the user's</span>
|
|
<span class="c1"># generate_async as a per-request error, and won't stop the whole service.</span>
|
|
<span class="c1"># 2. Exception for the errors from the background threads/processes, this will be processed in the main thread,</span>
|
|
<span class="c1"># and stop the whole service.</span>
|
|
<span class="n">error</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span> <span class="o">|</span> <span class="ne">Exception</span><span class="p">]</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_stats</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="n">atexit</span><span class="o">.</span><span class="n">register</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">shutdown</span><span class="p">)</span>
|
|
|
|
<span class="c1"># This is used to capture the exceptions from the threads.</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span> <span class="o">=</span> <span class="n">Queue</span><span class="p">()</span>
|
|
|
|
<span class="c1"># A flag to avoid calling shutdown() recursively. This happens when the background threads raise errors.</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">doing_shutdown</span> <span class="o">=</span> <span class="kc">False</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_last_client_id</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">1</span>
|
|
|
|
<span class="nd">@abstractmethod</span>
|
|
<span class="k">def</span> <span class="nf">submit</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">request</span><span class="p">:</span> <span class="n">GenerationRequest</span><span class="p">)</span> <span class="o">-></span> <span class="n">GenerationResult</span><span class="p">:</span>
|
|
<span class="k">pass</span>
|
|
|
|
<span class="k">def</span> <span class="nf">generate_async</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="p">,</span>
|
|
<span class="n">prompt_token_ids</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">],</span>
|
|
<span class="n">sampling_params</span><span class="p">:</span> <span class="n">SamplingParams</span><span class="p">,</span>
|
|
<span class="n">query_token_ids</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">np</span><span class="o">.</span><span class="n">ndarray</span><span class="p">,</span> <span class="nb">list</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">lora_request</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">LoRARequest</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">prompt_adapter_request</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">PromptAdapterRequest</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">streaming</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">prompt_tuning_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">list</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="p">)</span> <span class="o">-></span> <span class="n">GenerationResult</span><span class="p">:</span>
|
|
<span class="w"> </span><span class="sd">"""Generate output for the given prompt token ids in the asynchronous mode.</span>
|
|
<span class="sd"> Asynchronous generation accepts single prompt only.</span>
|
|
<span class="sd"> """</span>
|
|
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">prompt_token_ids</span><span class="p">[</span><span class="mi">0</span><span class="p">],</span> <span class="nb">int</span><span class="p">)</span>
|
|
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">sampling_params</span><span class="p">,</span> <span class="n">SamplingParams</span><span class="p">)</span>
|
|
<span class="n">result</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">submit</span><span class="p">(</span>
|
|
<span class="n">GenerationRequest</span><span class="p">(</span><span class="n">prompt_token_ids</span><span class="p">,</span>
|
|
<span class="n">sampling_params</span><span class="o">=</span><span class="n">sampling_params</span><span class="p">,</span>
|
|
<span class="n">query_token_ids</span><span class="o">=</span><span class="n">query_token_ids</span><span class="p">,</span>
|
|
<span class="n">lora_request</span><span class="o">=</span><span class="n">lora_request</span><span class="p">,</span>
|
|
<span class="n">prompt_adapter_request</span><span class="o">=</span><span class="n">prompt_adapter_request</span><span class="p">,</span>
|
|
<span class="n">streaming</span><span class="o">=</span><span class="n">streaming</span><span class="p">,</span>
|
|
<span class="n">prompt_tuning_config</span><span class="o">=</span><span class="n">prompt_tuning_config</span><span class="p">))</span>
|
|
<span class="k">return</span> <span class="n">result</span>
|
|
|
|
<span class="k">def</span> <span class="nf">generate</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="p">,</span>
|
|
<span class="n">prompt_token_ids</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">],</span> <span class="n">List</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]]],</span>
|
|
<span class="n">sampling_params</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">SamplingParams</span><span class="p">,</span> <span class="n">List</span><span class="p">[</span><span class="n">SamplingParams</span><span class="p">]],</span>
|
|
<span class="n">query_token_ids</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">np</span><span class="o">.</span><span class="n">ndarray</span><span class="p">,</span> <span class="nb">list</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">lora_request</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span><span class="n">LoRARequest</span><span class="p">,</span> <span class="n">List</span><span class="p">[</span><span class="n">LoRARequest</span><span class="p">]]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">prompt_adapter_request</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Union</span><span class="p">[</span>
|
|
<span class="n">PromptAdapterRequest</span><span class="p">,</span> <span class="n">List</span><span class="p">[</span><span class="n">PromptAdapterRequest</span><span class="p">]]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="p">)</span> <span class="o">-></span> <span class="n">Union</span><span class="p">[</span><span class="n">GenerationResult</span><span class="p">,</span> <span class="n">List</span><span class="p">[</span><span class="n">GenerationResult</span><span class="p">]]:</span>
|
|
<span class="w"> </span><span class="sd">"""Generate output for the given prompt token ids in the synchronous mode.</span>
|
|
<span class="sd"> Synchronous generation accepts either single prompt or batched prompts.</span>
|
|
<span class="sd"> """</span>
|
|
<span class="n">unbatched</span> <span class="o">=</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">prompt_token_ids</span><span class="p">[</span><span class="mi">0</span><span class="p">],</span> <span class="nb">int</span><span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="n">unbatched</span><span class="p">:</span>
|
|
<span class="n">prompt_token_ids</span> <span class="o">=</span> <span class="p">[</span><span class="n">prompt_token_ids</span><span class="p">]</span>
|
|
<span class="k">if</span> <span class="n">query_token_ids</span><span class="p">:</span>
|
|
<span class="n">query_token_ids</span> <span class="o">=</span> <span class="p">[</span><span class="n">query_token_ids</span><span class="p">]</span>
|
|
|
|
<span class="n">futures</span> <span class="o">=</span> <span class="p">[]</span>
|
|
<span class="k">for</span> <span class="n">i</span><span class="p">,</span> <span class="n">p</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">prompt_token_ids</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">sampling_params</span><span class="p">,</span> <span class="nb">list</span><span class="p">):</span>
|
|
<span class="n">sp</span> <span class="o">=</span> <span class="n">sampling_params</span><span class="p">[</span><span class="n">i</span><span class="p">]</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="n">sp</span> <span class="o">=</span> <span class="n">sampling_params</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">lora_request</span><span class="p">,</span> <span class="nb">list</span><span class="p">):</span>
|
|
<span class="n">lora_req</span> <span class="o">=</span> <span class="n">lora_request</span><span class="p">[</span><span class="n">i</span><span class="p">]</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="n">lora_req</span> <span class="o">=</span> <span class="n">lora_request</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">prompt_adapter_request</span><span class="p">,</span> <span class="nb">list</span><span class="p">):</span>
|
|
<span class="n">pa_req</span> <span class="o">=</span> <span class="n">prompt_adapter_request</span><span class="p">[</span><span class="n">i</span><span class="p">]</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="n">pa_req</span> <span class="o">=</span> <span class="n">prompt_adapter_request</span>
|
|
<span class="n">future</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">generate_async</span><span class="p">(</span><span class="n">p</span><span class="p">,</span>
|
|
<span class="n">sampling_params</span><span class="o">=</span><span class="n">sp</span><span class="p">,</span>
|
|
<span class="n">query_token_ids</span><span class="o">=</span><span class="n">query_token_ids</span><span class="p">,</span>
|
|
<span class="n">lora_request</span><span class="o">=</span><span class="n">lora_req</span><span class="p">,</span>
|
|
<span class="n">prompt_adapter_request</span><span class="o">=</span><span class="n">pa_req</span><span class="p">,</span>
|
|
<span class="n">streaming</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
|
|
<span class="n">futures</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">future</span><span class="p">)</span>
|
|
|
|
<span class="k">for</span> <span class="n">future</span> <span class="ow">in</span> <span class="n">futures</span><span class="p">:</span>
|
|
<span class="n">future</span><span class="o">.</span><span class="n">result</span><span class="p">()</span>
|
|
|
|
<span class="k">if</span> <span class="n">unbatched</span><span class="p">:</span>
|
|
<span class="n">futures</span> <span class="o">=</span> <span class="n">futures</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
|
|
|
|
<span class="k">return</span> <span class="n">futures</span>
|
|
|
|
<span class="k">def</span> <span class="nf">_get_next_client_id</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="c1"># (self._last_client_id + 1) % UINT64_MAX</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_last_client_id</span> <span class="o">=</span> <span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_last_client_id</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)</span> <span class="o">&</span> <span class="p">((</span><span class="mi">1</span> <span class="o"><<</span> <span class="mi">64</span><span class="p">)</span> <span class="o">-</span> <span class="mi">1</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_last_client_id</span>
|
|
|
|
<span class="k">def</span> <span class="nf">_handle_background_error</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">error</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="ne">Exception</span> <span class="o">|</span> <span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">""" Process the errors from the threads or processes.</span>
|
|
<span class="sd"> NOTE: This should be called in the main thread.</span>
|
|
<span class="sd"> """</span>
|
|
<span class="k">if</span> <span class="n">error</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="c1"># For details please refer to the comment of `GenerationResult.error`</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">error</span><span class="p">,</span> <span class="ne">Exception</span><span class="p">):</span>
|
|
<span class="c1"># Serious error from background thread or process</span>
|
|
<span class="k">if</span> <span class="n">enable_llm_debug</span><span class="p">():</span>
|
|
<span class="n">print_colored</span><span class="p">(</span>
|
|
<span class="sa">f</span><span class="s2">"Got background error: </span><span class="si">{</span><span class="nb">repr</span><span class="p">(</span><span class="n">error</span><span class="p">)</span><span class="si">}</span><span class="s2">, will shutdown the LLM instance</span><span class="se">\n</span><span class="s2">"</span><span class="p">,</span>
|
|
<span class="s2">"red"</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
|
|
<span class="k">raise</span> <span class="n">error</span>
|
|
<span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">error</span><span class="p">,</span> <span class="nb">str</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="n">enable_llm_debug</span><span class="p">():</span>
|
|
<span class="n">print_colored</span><span class="p">(</span><span class="sa">f</span><span class="s2">"Got per-request error: </span><span class="si">{</span><span class="nb">repr</span><span class="p">(</span><span class="n">error</span><span class="p">)</span><span class="si">}</span><span class="se">\n</span><span class="s2">"</span><span class="p">,</span>
|
|
<span class="s2">"red"</span><span class="p">)</span>
|
|
<span class="n">print_colored</span><span class="p">(</span><span class="nb">str</span><span class="p">(</span><span class="n">traceback</span><span class="o">.</span><span class="n">extract_stack</span><span class="p">())</span> <span class="o">+</span> <span class="s2">"</span><span class="se">\n</span><span class="s2">"</span><span class="p">,</span> <span class="s2">"red"</span><span class="p">)</span>
|
|
<span class="c1"># A per-request error, can be captured and ignored</span>
|
|
<span class="k">raise</span> <span class="n">RequestError</span><span class="p">(</span><span class="n">error</span><span class="p">)</span>
|
|
|
|
<span class="c1"># Here we raise the first error in the queue. This method will be called repeatedly and user can choose to catch</span>
|
|
<span class="c1"># more than one error.</span>
|
|
<span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="o">.</span><span class="n">empty</span><span class="p">():</span>
|
|
<span class="n">e</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="o">.</span><span class="n">task_done</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
|
|
<span class="c1"># We can catch some exceptions here.</span>
|
|
<span class="k">raise</span> <span class="n">e</span>
|
|
|
|
<span class="nd">@abstractmethod</span>
|
|
<span class="k">def</span> <span class="nf">shutdown</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">pass</span>
|
|
|
|
<span class="k">def</span> <span class="nf">create_stats_queue</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="c1"># Stats queue is created during first submission to ensure event loop exists if it is needed.</span>
|
|
<span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">_stats</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="n">has_event_loop</span><span class="p">():</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_stats</span> <span class="o">=</span> <span class="n">AsyncQueue</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_stats</span><span class="o">.</span><span class="n">sync_q</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">stats_aqueue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_stats</span><span class="o">.</span><span class="n">async_q</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_stats</span> <span class="o">=</span> <span class="n">Queue</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_stats</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">stats_aqueue</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="k">def</span> <span class="nf">get_stats</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">timeout</span><span class="o">=</span><span class="kc">None</span><span class="p">)</span> <span class="o">-></span> <span class="nb">str</span><span class="p">:</span>
|
|
<span class="w"> </span><span class="sd">''' Get the stats from the runtime.</span>
|
|
|
|
<span class="sd"> Exceptions:</span>
|
|
<span class="sd"> NoStatsAvailable: If the stats are not available.</span>
|
|
|
|
<span class="sd"> Returns:</span>
|
|
<span class="sd"> str: The stats in JSON format.</span>
|
|
|
|
<span class="sd"> Known issue:</span>
|
|
<span class="sd"> The `get_stats` cannot mix with `aget_stats` in the same Executor instance.</span>
|
|
<span class="sd"> '''</span>
|
|
<span class="k">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span><span class="p">,</span> <span class="s2">"The stats queue is not created. It is likely that `get_stats` and `aget_stats` methods"</span> \
|
|
<span class="s2">" are mixed."</span>
|
|
<span class="k">try</span><span class="p">:</span>
|
|
<span class="n">res</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="n">timeout</span><span class="o">=</span><span class="n">timeout</span><span class="p">)</span>
|
|
<span class="k">except</span> <span class="n">Empty</span><span class="p">:</span>
|
|
<span class="k">raise</span> <span class="n">NoStatsAvailable</span>
|
|
<span class="k">return</span> <span class="n">res</span>
|
|
|
|
<span class="k">async</span> <span class="k">def</span> <span class="nf">aget_stats</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">timeout</span><span class="o">=</span><span class="kc">None</span><span class="p">)</span> <span class="o">-></span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]:</span>
|
|
<span class="w"> </span><span class="sd">''' Get the stats from the runtime.</span>
|
|
|
|
<span class="sd"> Exceptions:</span>
|
|
<span class="sd"> NoStatsAvailable: If the stats are not available.</span>
|
|
|
|
<span class="sd"> Returns:</span>
|
|
<span class="sd"> str: The stats in JSON format.</span>
|
|
|
|
<span class="sd"> Known issue:</span>
|
|
<span class="sd"> The `aget_stats` cannot mix with `get_stats` in the same Executor instance.</span>
|
|
<span class="sd"> '''</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">create_stats_queue</span><span class="p">()</span>
|
|
<span class="k">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">stats_aqueue</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
|
|
|
|
<span class="k">if</span> <span class="ow">not</span> <span class="n">has_event_loop</span><span class="p">():</span>
|
|
<span class="k">raise</span> <span class="n">NoStatsAvailable</span>
|
|
|
|
<span class="k">try</span><span class="p">:</span>
|
|
<span class="n">res</span> <span class="o">=</span> <span class="k">await</span> <span class="bp">self</span><span class="o">.</span><span class="n">stats_aqueue</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="n">timeout</span><span class="o">=</span><span class="n">timeout</span><span class="p">)</span>
|
|
<span class="k">except</span> <span class="n">asyncio</span><span class="o">.</span><span class="n">TimeoutError</span><span class="p">:</span>
|
|
<span class="k">raise</span> <span class="n">NoStatsAvailable</span>
|
|
<span class="k">return</span> <span class="n">res</span>
|
|
|
|
<span class="nd">@staticmethod</span>
|
|
<span class="k">def</span> <span class="nf">create</span><span class="p">(</span>
|
|
<span class="n">engine</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">Path</span><span class="p">,</span> <span class="n">Engine</span><span class="p">],</span>
|
|
<span class="n">executor_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">tllm</span><span class="o">.</span><span class="n">ExecutorConfig</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">logits_post_processor_map</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Callable</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">model_world_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">1</span><span class="p">,</span>
|
|
<span class="n">world_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">0</span><span class="p">,</span>
|
|
<span class="n">mpi_session</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">MpiSession</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">reuse_mpi_comm</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
|
|
<span class="p">)</span> <span class="o">-></span> <span class="n">Union</span><span class="p">[</span><span class="s2">"ExecutorBindingsProxy"</span><span class="p">,</span> <span class="s2">"ExecutorBindingsWorker"</span><span class="p">]:</span>
|
|
|
|
<span class="k">if</span> <span class="n">world_size</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="n">world_size</span> <span class="o">=</span> <span class="n">mpi_world_size</span><span class="p">()</span>
|
|
|
|
<span class="k">if</span> <span class="n">world_size</span> <span class="o">></span> <span class="mi">1</span> <span class="ow">and</span> <span class="n">world_size</span> <span class="o"><</span> <span class="n">model_world_size</span><span class="p">:</span>
|
|
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span>
|
|
<span class="s2">"Cannot instantiate Generator for engine built "</span>
|
|
<span class="sa">f</span><span class="s2">"for </span><span class="si">{</span><span class="n">model_world_size</span><span class="si">}</span><span class="s2"> ranks, while currently running "</span>
|
|
<span class="sa">f</span><span class="s2">"on </span><span class="si">{</span><span class="n">world_size</span><span class="si">}</span><span class="s2"> ranks."</span><span class="p">)</span>
|
|
|
|
<span class="n">worker_kwargs</span> <span class="o">=</span> <span class="p">{</span>
|
|
<span class="s2">"engine"</span><span class="p">:</span> <span class="n">engine</span><span class="p">,</span>
|
|
<span class="s2">"executor_config"</span><span class="p">:</span> <span class="n">executor_config</span><span class="p">,</span>
|
|
<span class="s2">"logits_post_processor_map"</span><span class="p">:</span> <span class="n">logits_post_processor_map</span><span class="p">,</span>
|
|
<span class="p">}</span>
|
|
|
|
<span class="c1"># The case where the Python main process is launched by mpirun</span>
|
|
<span class="n">mpirun_launch</span> <span class="o">=</span> <span class="n">external_mpi_comm_available</span><span class="p">(</span><span class="n">model_world_size</span><span class="p">)</span>
|
|
<span class="c1"># The case where the Python main process utilizes mpi4py to spawn MPI workers</span>
|
|
<span class="n">spawn_workers</span> <span class="o">=</span> <span class="n">need_spawn_mpi_workers</span><span class="p">(</span><span class="n">model_world_size</span><span class="p">)</span>
|
|
<span class="k">if</span> <span class="n">spawn_workers</span> <span class="ow">or</span> <span class="p">(</span><span class="n">mpirun_launch</span> <span class="ow">and</span> <span class="n">reuse_mpi_comm</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="n">reuse_mpi_comm</span><span class="p">:</span>
|
|
<span class="k">assert</span> <span class="n">mpi_session</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">,</span> <span class="s2">"reuse_mpi_comm requires an external MPI session"</span>
|
|
<span class="k">return</span> <span class="n">ExecutorBindingsProxy</span><span class="p">(</span><span class="n">worker_kwargs</span><span class="p">,</span>
|
|
<span class="n">model_world_size</span><span class="o">=</span><span class="n">model_world_size</span><span class="p">,</span>
|
|
<span class="n">mpi_session</span><span class="o">=</span><span class="n">mpi_session</span><span class="p">)</span>
|
|
|
|
<span class="c1"># For single-gpu case:</span>
|
|
<span class="c1"># Partition the workload to multiple process for performance. While this requires uses to protect their entrypoint</span>
|
|
<span class="c1"># to `if __name__ == "__main__":`.</span>
|
|
<span class="c1">#</span>
|
|
<span class="c1"># The LogitsPostProcessorConfig cannot pickle, which cannot work with multi-process now</span>
|
|
<span class="c1"># TODO: Make logits_post_processor work with multi-process</span>
|
|
<span class="k">if</span> <span class="n">executor_config</span> <span class="ow">is</span> <span class="kc">None</span> <span class="ow">or</span> <span class="n">executor_config</span><span class="o">.</span><span class="n">logits_post_processor_config</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="n">ctx</span> <span class="o">=</span> <span class="n">multiprocessing</span><span class="o">.</span><span class="n">get_context</span><span class="p">(</span><span class="s2">"spawn"</span><span class="p">)</span>
|
|
<span class="c1"># The ProcessPoolExecutorSession is used to support Windows, as mpi4py cannot.</span>
|
|
<span class="n">mpi_session</span> <span class="o">=</span> <span class="n">ProcessPoolExecutorSession</span><span class="p">(</span><span class="n">n_workers</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span>
|
|
<span class="n">mp_context</span><span class="o">=</span><span class="n">ctx</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="n">ExecutorBindingsProxy</span><span class="p">(</span><span class="n">worker_kwargs</span><span class="p">,</span>
|
|
<span class="n">model_world_size</span><span class="o">=</span><span class="n">model_world_size</span><span class="p">,</span>
|
|
<span class="n">mpi_session</span><span class="o">=</span><span class="n">mpi_session</span><span class="p">)</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="n">ExecutorBindingsWorker</span><span class="p">(</span><span class="o">**</span><span class="n">worker_kwargs</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">wait_first_completed</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="p">,</span> <span class="n">futures</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="n">GenerationResult</span><span class="p">]</span>
|
|
<span class="p">)</span> <span class="o">-></span> <span class="n">Generator</span><span class="p">[</span><span class="n">GenerationResult</span><span class="p">,</span> <span class="kc">None</span><span class="p">,</span> <span class="kc">None</span><span class="p">]:</span>
|
|
<span class="n">wait_set</span> <span class="o">=</span> <span class="nb">set</span><span class="p">(</span><span class="n">futures</span><span class="p">)</span>
|
|
|
|
<span class="c1"># clear already-finished requests</span>
|
|
<span class="k">for</span> <span class="n">f</span> <span class="ow">in</span> <span class="n">futures</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="n">f</span><span class="o">.</span><span class="n">_done</span><span class="p">:</span>
|
|
<span class="n">wait_set</span><span class="o">.</span><span class="n">pop</span><span class="p">(</span><span class="n">f</span><span class="p">)</span>
|
|
<span class="k">yield</span> <span class="n">f</span>
|
|
|
|
<span class="c1"># wait remaining active requests</span>
|
|
<span class="k">while</span> <span class="nb">len</span><span class="p">(</span><span class="n">wait_set</span><span class="p">)</span> <span class="o">></span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="n">fut</span> <span class="o">=</span> <span class="n">wait_set</span><span class="o">.</span><span class="n">pop</span><span class="p">()</span>
|
|
<span class="k">if</span> <span class="n">fut</span><span class="o">.</span><span class="n">request_id</span> <span class="ow">not</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="p">:</span>
|
|
<span class="k">yield</span> <span class="n">fut</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="n">wait_set</span><span class="o">.</span><span class="n">add</span><span class="p">(</span><span class="n">fut</span><span class="p">)</span>
|
|
|
|
|
|
<span class="k">class</span> <span class="nc">ProcessPoolExecutorSession</span><span class="p">(</span><span class="n">MpiSession</span><span class="p">):</span>
|
|
<span class="c1"># This process pool is introduced for better recoverable exceptions handling.</span>
|
|
<span class="c1"># It replaces MpiPoolExecutor for single-gpu case.</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">n_workers</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">n_workers</span> <span class="o">=</span> <span class="n">n_workers</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_pool</span> <span class="o">=</span> <span class="n">ProcessPoolExecutor</span><span class="p">(</span><span class="n">max_workers</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">n_workers</span><span class="p">,</span>
|
|
<span class="o">**</span><span class="n">kwargs</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">submit</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">task</span><span class="p">:</span> <span class="n">Callable</span><span class="p">,</span> <span class="o">*</span><span class="n">args</span><span class="p">,</span>
|
|
<span class="o">**</span><span class="n">kwargs</span><span class="p">)</span> <span class="o">-></span> <span class="n">List</span><span class="p">[</span><span class="n">concurrent</span><span class="o">.</span><span class="n">futures</span><span class="o">.</span><span class="n">Future</span><span class="p">]:</span>
|
|
<span class="k">return</span> <span class="p">[</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_pool</span><span class="o">.</span><span class="n">submit</span><span class="p">(</span><span class="n">task</span><span class="p">,</span> <span class="o">*</span><span class="n">args</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">)</span>
|
|
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">n_workers</span><span class="p">)</span>
|
|
<span class="p">]</span>
|
|
|
|
<span class="k">def</span> <span class="nf">submit_sync</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">task</span><span class="p">:</span> <span class="n">Callable</span><span class="p">,</span> <span class="o">*</span><span class="n">args</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">)</span> <span class="o">-></span> <span class="n">List</span><span class="p">[</span><span class="n">Any</span><span class="p">]:</span>
|
|
<span class="n">futures</span> <span class="o">=</span> <span class="p">[</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_pool</span><span class="o">.</span><span class="n">submit</span><span class="p">(</span><span class="n">task</span><span class="p">,</span> <span class="o">*</span><span class="n">args</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">)</span>
|
|
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">n_workers</span><span class="p">)</span>
|
|
<span class="p">]</span>
|
|
<span class="k">return</span> <span class="p">[</span><span class="n">future</span><span class="o">.</span><span class="n">result</span><span class="p">()</span> <span class="k">for</span> <span class="n">future</span> <span class="ow">in</span> <span class="n">futures</span><span class="p">]</span>
|
|
|
|
<span class="k">def</span> <span class="nf">shutdown</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_pool</span><span class="o">.</span><span class="n">shutdown</span><span class="p">(</span><span class="n">wait</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
|
|
|
|
|
|
<span class="k">class</span> <span class="nc">ExecutorBindingsWorker</span><span class="p">(</span><span class="n">GenerationExecutor</span><span class="p">):</span>
|
|
|
|
<span class="k">class</span> <span class="nc">WorkerExit</span><span class="p">(</span><span class="ne">GeneratorExit</span><span class="p">):</span>
|
|
<span class="k">pass</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="p">,</span>
|
|
<span class="n">engine</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">Path</span><span class="p">,</span> <span class="n">Engine</span><span class="p">],</span>
|
|
<span class="n">executor_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">tllm</span><span class="o">.</span><span class="n">ExecutorConfig</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">logits_post_processor_map</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Callable</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">engine</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">rank</span> <span class="o">=</span> <span class="n">mpi_rank</span><span class="p">()</span>
|
|
<span class="c1"># mapping: client_id -> GenerationResult</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">int</span><span class="p">,</span> <span class="n">GenerationResult</span><span class="p">]</span> <span class="o">=</span> <span class="p">{}</span>
|
|
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">engine</span><span class="p">,</span> <span class="nb">list</span><span class="p">):</span>
|
|
<span class="n">engine</span> <span class="o">=</span> <span class="n">engine</span><span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">rank</span><span class="p">]</span>
|
|
|
|
<span class="k">if</span> <span class="n">executor_config</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="n">executor_config</span> <span class="o">=</span> <span class="n">tllm</span><span class="o">.</span><span class="n">ExecutorConfig</span><span class="p">(</span><span class="mi">1</span><span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="n">logits_post_processor_map</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="n">executor_config</span><span class="o">.</span><span class="n">logits_post_processor_config</span> <span class="o">=</span> <span class="n">tllm</span><span class="o">.</span><span class="n">LogitsPostProcessorConfig</span><span class="p">(</span>
|
|
<span class="n">processor_map</span><span class="o">=</span><span class="n">logits_post_processor_map</span><span class="p">,</span> <span class="n">replicate</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">_create_engine</span><span class="p">():</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">engine</span><span class="p">,</span> <span class="n">Engine</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="n">tllm</span><span class="o">.</span><span class="n">Executor</span><span class="p">(</span><span class="n">engine</span><span class="o">.</span><span class="n">engine</span><span class="p">,</span>
|
|
<span class="n">json</span><span class="o">.</span><span class="n">dumps</span><span class="p">(</span><span class="n">engine</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">to_dict</span><span class="p">(),</span>
|
|
<span class="bp">cls</span><span class="o">=</span><span class="n">ConfigEncoder</span><span class="p">),</span>
|
|
<span class="n">tllm</span><span class="o">.</span><span class="n">ModelType</span><span class="o">.</span><span class="n">DECODER_ONLY</span><span class="p">,</span>
|
|
<span class="n">executor_config</span><span class="o">=</span><span class="n">executor_config</span><span class="p">,</span>
|
|
<span class="n">managed_weights</span><span class="o">=</span><span class="n">engine</span><span class="o">.</span><span class="n">managed_weights</span><span class="p">)</span>
|
|
|
|
<span class="n">use_default_executor</span> <span class="o">=</span> <span class="kc">True</span>
|
|
<span class="c1"># PIVOT_TO_PYTHON_START</span>
|
|
<span class="n">use_default_executor</span> <span class="o">=</span> <span class="ow">not</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">executor_config</span><span class="p">,</span> <span class="s2">"backend"</span><span class="p">)</span>
|
|
<span class="c1"># PIVOT_TO_PYTHON_END</span>
|
|
<span class="k">if</span> <span class="n">use_default_executor</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="n">tllm</span><span class="o">.</span><span class="n">Executor</span><span class="p">(</span><span class="n">engine</span><span class="p">,</span> <span class="n">tllm</span><span class="o">.</span><span class="n">ModelType</span><span class="o">.</span><span class="n">DECODER_ONLY</span><span class="p">,</span>
|
|
<span class="n">executor_config</span><span class="p">)</span>
|
|
|
|
<span class="c1"># PIVOT_TO_PYTHON_START</span>
|
|
<span class="kn">from</span> <span class="nn">tensorrt_llm.pyexecutor.backend_registries.backend_registry</span> <span class="kn">import</span> \
|
|
<span class="n">unique_create_executor</span>
|
|
<span class="k">return</span> <span class="n">unique_create_executor</span><span class="p">(</span><span class="n">engine</span><span class="p">,</span>
|
|
<span class="n">tllm</span><span class="o">.</span><span class="n">ModelType</span><span class="o">.</span><span class="n">DECODER_ONLY</span><span class="p">,</span>
|
|
<span class="n">executor_config</span><span class="o">=</span><span class="n">executor_config</span><span class="p">,</span>
|
|
<span class="n">device_id</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">rank</span> <span class="o">%</span>
|
|
<span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">device_count</span><span class="p">())</span>
|
|
<span class="c1"># PIVOT_TO_PYTHON_END</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">engine</span> <span class="o">=</span> <span class="n">_create_engine</span><span class="p">()</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_lora_manager</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">LoraManager</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_prompt_adapter_manager</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">PromptAdapterManager</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_runtime_model_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">ModelConfig</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">rank</span> <span class="o">==</span> <span class="mi">0</span> <span class="ow">and</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="p">,</span> <span class="n">tllm</span><span class="o">.</span><span class="n">Executor</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">engine</span><span class="p">,</span> <span class="n">Engine</span><span class="p">):</span>
|
|
<span class="n">engine_config</span> <span class="o">=</span> <span class="n">engine</span><span class="o">.</span><span class="n">config</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="n">engine_config</span> <span class="o">=</span> <span class="n">EngineConfig</span><span class="o">.</span><span class="n">from_json_file</span><span class="p">(</span>
|
|
<span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="n">engine</span><span class="si">}</span><span class="s2">/config.json"</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_runtime_model_config</span> <span class="o">=</span> <span class="n">_engine_config_to_model_config</span><span class="p">(</span>
|
|
<span class="n">engine_config</span><span class="p">)</span>
|
|
<span class="k">if</span> <span class="n">engine_config</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">plugin_config</span><span class="o">.</span><span class="n">lora_plugin</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_lora_manager</span> <span class="o">=</span> <span class="n">LoraManager</span><span class="p">()</span>
|
|
<span class="k">if</span> <span class="n">engine_config</span><span class="o">.</span><span class="n">build_config</span><span class="o">.</span><span class="n">max_prompt_embedding_table_size</span> <span class="o">></span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_prompt_adapter_manager</span> <span class="o">=</span> <span class="n">PromptAdapterManager</span><span class="p">()</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">await_response_thread</span> <span class="o">=</span> <span class="n">ManagedThread</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">await_response_task</span><span class="p">,</span>
|
|
<span class="n">error_queue</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="p">,</span>
|
|
<span class="n">name</span><span class="o">=</span><span class="s2">"await_response_thread"</span><span class="p">)</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span> <span class="o">=</span> <span class="n">ManagedThread</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_task</span><span class="p">,</span>
|
|
<span class="n">error_queue</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="p">,</span>
|
|
<span class="n">name</span><span class="o">=</span><span class="s2">"dispatch_stats_thread"</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">create_stats_queue</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="c1"># Stats queue is created during first submission to ensure event loop exists if it is needed.</span>
|
|
<span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">_stats</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="n">has_event_loop</span><span class="p">():</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_stats</span> <span class="o">=</span> <span class="n">AsyncQueue</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_stats</span><span class="o">.</span><span class="n">sync_q</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">stats_aqueue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_stats</span><span class="o">.</span><span class="n">async_q</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_stats</span> <span class="o">=</span> <span class="n">Queue</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_stats</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">stats_aqueue</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="k">def</span> <span class="nf">set_result_queue</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">queue</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">"""In multi-gpu mode, result_queue will be set here to communicate between the proxy and the worker 0 process."""</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span> <span class="o">=</span> <span class="n">queue</span>
|
|
|
|
<span class="k">def</span> <span class="nf">set_stats_queue</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">queue</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">"""In multi-gpu mode, stats_queue will be set here to communicate between the proxy and the worker 0 process."""</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_stats</span> <span class="o">=</span> <span class="n">queue</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_stats</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">stats_aqueue</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="k">def</span> <span class="nf">return_queue</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">client_id</span><span class="p">:</span> <span class="nb">int</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">""" If a centralized result queue is registered (used for communication with the proxy)</span>
|
|
<span class="sd"> send the message there.</span>
|
|
<span class="sd"> Otherwise, push the result directly in the GenerationResult queue.</span>
|
|
<span class="sd"> """</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="p">[</span><span class="n">client_id</span><span class="p">]</span><span class="o">.</span><span class="n">queue</span>
|
|
|
|
<span class="k">def</span> <span class="nf">start_awaiter_thread</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">can_enqueue_requests</span><span class="p">(</span>
|
|
<span class="p">)</span> <span class="ow">and</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">await_response_thread</span><span class="o">.</span><span class="n">is_alive</span><span class="p">():</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">await_response_thread</span><span class="o">.</span><span class="n">start</span><span class="p">()</span>
|
|
|
|
<span class="k">def</span> <span class="nf">start_stats_thread</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">can_enqueue_requests</span><span class="p">(</span>
|
|
<span class="p">)</span> <span class="ow">and</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span><span class="o">.</span><span class="n">is_alive</span><span class="p">():</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span><span class="o">.</span><span class="n">start</span><span class="p">()</span>
|
|
|
|
<span class="k">def</span> <span class="nf">_engine_response_callback</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">response</span><span class="p">:</span> <span class="n">tllm</span><span class="o">.</span><span class="n">Response</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="n">response</span>
|
|
|
|
<span class="k">def</span> <span class="nf">await_response_task</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
|
|
<span class="c1"># Get responses and place in queue.</span>
|
|
|
|
<span class="n">async_queues</span> <span class="o">=</span> <span class="p">[]</span>
|
|
<span class="n">event_loop</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="k">for</span> <span class="n">response</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">await_responses</span><span class="p">(</span><span class="n">timeout</span><span class="o">=</span><span class="n">datetime</span><span class="o">.</span><span class="n">timedelta</span><span class="p">(</span>
|
|
<span class="n">milliseconds</span><span class="o">=</span><span class="mi">100</span><span class="p">)):</span>
|
|
<span class="n">response</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_engine_response_callback</span><span class="p">(</span><span class="n">response</span><span class="p">)</span>
|
|
<span class="k">if</span> <span class="n">response</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="k">continue</span>
|
|
|
|
<span class="n">client_id</span> <span class="o">=</span> <span class="n">response</span><span class="o">.</span><span class="n">client_id</span>
|
|
<span class="k">assert</span> <span class="n">client_id</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
|
|
<span class="k">if</span> <span class="n">response</span><span class="o">.</span><span class="n">has_error</span><span class="p">():</span>
|
|
<span class="c1"># This error will be dispatched to the user's generate_async for the corresponding request. It won't</span>
|
|
<span class="c1"># stop the whole service.</span>
|
|
<span class="n">rsp</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">Response</span><span class="p">(</span>
|
|
<span class="n">client_id</span><span class="p">,</span>
|
|
<span class="n">tensors</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="c1"># Note: error Response only has one finish reason.</span>
|
|
<span class="c1"># Since the error will be raised in the main thread, so the finish reason is not actually used.</span>
|
|
<span class="n">finish_reasons</span><span class="o">=</span><span class="p">[</span><span class="n">tllm</span><span class="o">.</span><span class="n">FinishReason</span><span class="o">.</span><span class="n">NOT_FINISHED</span><span class="p">],</span>
|
|
<span class="n">is_final</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
|
<span class="n">sequence_index</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">error</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">error_msg</span><span class="p">)</span>
|
|
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="n">tensors</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">ResponseTensors</span><span class="p">(</span>
|
|
<span class="n">output_token_ids</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">output_token_ids</span><span class="p">,</span>
|
|
<span class="n">context_logits</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">context_logits</span><span class="p">,</span>
|
|
<span class="n">generation_logits</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">generation_logits</span><span class="p">,</span>
|
|
<span class="n">log_probs</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">log_probs</span><span class="p">,</span>
|
|
<span class="n">cum_log_probs</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">cum_log_probs</span><span class="p">,</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="n">rsp</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">Response</span><span class="p">(</span>
|
|
<span class="n">client_id</span><span class="p">,</span>
|
|
<span class="n">tensors</span><span class="p">,</span>
|
|
<span class="n">finish_reasons</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">finish_reasons</span><span class="p">,</span>
|
|
<span class="n">is_final</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">is_final</span><span class="p">,</span>
|
|
<span class="n">sequence_index</span><span class="o">=</span><span class="n">response</span><span class="o">.</span><span class="n">result</span><span class="o">.</span><span class="n">sequence_index</span><span class="p">,</span>
|
|
<span class="n">error</span><span class="o">=</span><span class="kc">None</span><span class="p">)</span>
|
|
|
|
<span class="n">queue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">return_queue</span><span class="p">(</span><span class="n">client_id</span><span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_has_background_error</span><span class="p">():</span>
|
|
<span class="n">rsp</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_create_error_response</span><span class="p">(</span><span class="n">client_id</span><span class="p">)</span>
|
|
|
|
<span class="c1"># For AsyncQueue.sync_q, we will batch the events to avoid too many event notifications, thus put without</span>
|
|
<span class="c1"># wait here.</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">queue</span><span class="p">,</span> <span class="n">_SyncQueue</span><span class="p">):</span>
|
|
<span class="n">global_tracer</span><span class="p">()</span><span class="o">.</span><span class="n">log_instant</span><span class="p">(</span><span class="s2">"worker-rsp.put"</span><span class="p">)</span>
|
|
<span class="n">queue</span><span class="o">.</span><span class="n">put_nowait</span><span class="p">(</span><span class="n">rsp</span><span class="p">)</span>
|
|
<span class="n">async_queues</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">queue</span><span class="p">)</span>
|
|
<span class="c1"># all the loops are identical</span>
|
|
<span class="n">event_loop</span> <span class="o">=</span> <span class="n">event_loop</span> <span class="ow">or</span> <span class="n">queue</span><span class="o">.</span><span class="n">loop</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="n">global_tracer</span><span class="p">()</span><span class="o">.</span><span class="n">log_instant</span><span class="p">(</span><span class="s2">"worker-rsp.put"</span><span class="p">)</span>
|
|
<span class="n">queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">rsp</span><span class="p">)</span> <span class="c1"># This could be IPC</span>
|
|
|
|
<span class="c1"># Eliminate the finished GenerationRequest instances timely, which may take considerable memory.</span>
|
|
<span class="k">if</span> <span class="n">rsp</span><span class="o">.</span><span class="n">is_final</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="o">.</span><span class="n">pop</span><span class="p">(</span><span class="n">client_id</span><span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="n">async_queues</span><span class="p">:</span>
|
|
<span class="n">_SyncQueue</span><span class="o">.</span><span class="n">notify_queues</span><span class="p">(</span><span class="n">event_loop</span><span class="p">,</span> <span class="n">async_queues</span><span class="p">)</span>
|
|
|
|
<span class="k">return</span> <span class="kc">True</span> <span class="c1"># success</span>
|
|
|
|
<span class="k">def</span> <span class="nf">_has_background_error</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="o">.</span><span class="n">empty</span><span class="p">()</span>
|
|
|
|
<span class="k">def</span> <span class="nf">_create_error_response</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">client_id</span><span class="p">)</span> <span class="o">-></span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">:</span>
|
|
<span class="n">bck_error</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="o">.</span><span class="n">get_nowait</span><span class="p">()</span>
|
|
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">bck_error</span><span class="p">,</span> <span class="ne">Exception</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">(</span><span class="n">client_id</span><span class="p">,</span>
|
|
<span class="n">tensors</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">finish_reasons</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">is_final</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">sequence_index</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">error</span><span class="o">=</span><span class="n">bck_error</span><span class="p">)</span>
|
|
|
|
<span class="n">stats_count</span> <span class="o">=</span> <span class="mi">0</span>
|
|
|
|
<span class="k">def</span> <span class="nf">dispatch_stats_task</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
|
|
<span class="n">time</span><span class="o">.</span><span class="n">sleep</span><span class="p">(</span><span class="mf">0.1</span><span class="p">)</span>
|
|
<span class="c1"># Get stats and place in queue.</span>
|
|
<span class="k">for</span> <span class="n">stats</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">get_latest_iteration_stats</span><span class="p">():</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">stats_count</span> <span class="o">+=</span> <span class="mi">1</span>
|
|
<span class="k">while</span> <span class="nb">hasattr</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span><span class="p">,</span> <span class="s2">"full"</span><span class="p">)</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span><span class="o">.</span><span class="n">full</span><span class="p">():</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
|
|
|
|
<span class="k">try</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">stats</span><span class="o">.</span><span class="n">to_json_str</span><span class="p">())</span>
|
|
<span class="k">except</span> <span class="n">AsyncQueue</span><span class="o">.</span><span class="n">EventLoopShutdownError</span><span class="p">:</span>
|
|
<span class="c1"># This happens in the last stats loop while the generate workflow is stopped.</span>
|
|
<span class="k">pass</span>
|
|
<span class="k">except</span> <span class="ne">Exception</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
|
|
<span class="k">raise</span> <span class="n">e</span>
|
|
|
|
<span class="k">return</span> <span class="kc">True</span> <span class="c1"># success</span>
|
|
|
|
<span class="k">def</span> <span class="nf">start</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">create_stats_queue</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">start_awaiter_thread</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">start_stats_thread</span><span class="p">()</span>
|
|
|
|
<span class="k">def</span> <span class="nf">_load_lora_adapter</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">lora_request</span><span class="p">:</span> <span class="n">LoRARequest</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_lora_manager</span><span class="o">.</span><span class="n">load_from_ckpt</span><span class="p">(</span>
|
|
<span class="p">[</span><span class="n">lora_request</span><span class="o">.</span><span class="n">path</span><span class="p">],</span>
|
|
<span class="n">model_config</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_runtime_model_config</span><span class="p">,</span>
|
|
<span class="n">runtime_mapping</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">uids</span><span class="o">=</span><span class="p">[</span><span class="nb">str</span><span class="p">(</span><span class="n">lora_request</span><span class="o">.</span><span class="n">adapter_id</span><span class="p">)])</span>
|
|
|
|
<span class="k">def</span> <span class="nf">_load_prompt_adapter</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span>
|
|
<span class="n">prompt_adapter_request</span><span class="p">:</span> <span class="n">PromptAdapterRequest</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_prompt_adapter_manager</span><span class="o">.</span><span class="n">load_from_ckpt</span><span class="p">(</span>
|
|
<span class="p">[</span><span class="n">prompt_adapter_request</span><span class="o">.</span><span class="n">local_path</span><span class="p">],</span>
|
|
<span class="n">model_config</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_runtime_model_config</span><span class="p">,</span>
|
|
<span class="n">uids</span><span class="o">=</span><span class="p">[</span><span class="nb">str</span><span class="p">(</span><span class="n">prompt_adapter_request</span><span class="o">.</span><span class="n">adapter_id</span><span class="p">)])</span>
|
|
|
|
<span class="k">def</span> <span class="nf">_enqueue_request</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">request</span><span class="p">:</span> <span class="n">GenerationRequest</span><span class="p">)</span> <span class="o">-></span> <span class="nb">int</span><span class="p">:</span>
|
|
<span class="k">assert</span> <span class="n">request</span><span class="o">.</span><span class="n">id</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_lora_manager</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="n">request</span><span class="o">.</span><span class="n">lora_request</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_load_lora_adapter</span><span class="p">(</span><span class="n">request</span><span class="o">.</span><span class="n">lora_request</span><span class="p">)</span>
|
|
<span class="n">uid</span> <span class="o">=</span> <span class="nb">str</span><span class="p">(</span><span class="n">request</span><span class="o">.</span><span class="n">lora_request</span><span class="o">.</span><span class="n">adapter_id</span><span class="p">)</span>
|
|
<span class="n">lora_config</span> <span class="o">=</span> <span class="n">tllm</span><span class="o">.</span><span class="n">LoraConfig</span><span class="p">(</span>
|
|
<span class="n">task_id</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">lora_request</span><span class="o">.</span><span class="n">adapter_id</span><span class="p">,</span>
|
|
<span class="n">weights</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_lora_manager</span><span class="o">.</span><span class="n">cpp_lora_weights</span><span class="p">[</span><span class="n">uid</span><span class="p">],</span>
|
|
<span class="n">config</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_lora_manager</span><span class="o">.</span><span class="n">cpp_lora_config</span><span class="p">[</span><span class="n">uid</span><span class="p">])</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="n">lora_config</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="n">prompt_token_ids</span> <span class="o">=</span> <span class="n">copy</span><span class="o">.</span><span class="n">deepcopy</span><span class="p">(</span><span class="n">request</span><span class="o">.</span><span class="n">prompt_token_ids</span><span class="p">)</span>
|
|
<span class="n">prompt_tuning_config</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="k">if</span> <span class="n">request</span><span class="o">.</span><span class="n">prompt_adapter_request</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="k">assert</span> <span class="n">request</span><span class="o">.</span><span class="n">prompt_tuning_config</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">,</span> \
|
|
<span class="s2">"cannot accept both prompt_adapter_request and prompt_tuning_config in one request"</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_load_prompt_adapter</span><span class="p">(</span><span class="n">request</span><span class="o">.</span><span class="n">prompt_adapter_request</span><span class="p">)</span>
|
|
<span class="n">uid</span> <span class="o">=</span> <span class="nb">str</span><span class="p">(</span><span class="n">request</span><span class="o">.</span><span class="n">prompt_adapter_request</span><span class="o">.</span><span class="n">adapter_id</span><span class="p">)</span>
|
|
<span class="n">prompt_tuning_config</span> <span class="o">=</span> <span class="n">tllm</span><span class="o">.</span><span class="n">PromptTuningConfig</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_prompt_adapter_manager</span><span class="o">.</span><span class="n">uid_to_weights</span><span class="p">[</span><span class="n">uid</span><span class="p">])</span>
|
|
<span class="n">vocab_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_runtime_model_config</span><span class="o">.</span><span class="n">vocab_size</span>
|
|
<span class="n">pa_length</span> <span class="o">=</span> <span class="n">prompt_tuning_config</span><span class="o">.</span><span class="n">embedding_table</span><span class="o">.</span><span class="n">size</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span>
|
|
<span class="n">prompt_token_ids</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="nb">range</span><span class="p">(</span>
|
|
<span class="n">vocab_size</span><span class="p">,</span> <span class="n">vocab_size</span> <span class="o">+</span> <span class="n">pa_length</span><span class="p">))</span> <span class="o">+</span> <span class="n">prompt_token_ids</span>
|
|
<span class="k">elif</span> <span class="n">request</span><span class="o">.</span><span class="n">prompt_tuning_config</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="n">prompt_tuning_config</span> <span class="o">=</span> <span class="n">tllm</span><span class="o">.</span><span class="n">PromptTuningConfig</span><span class="p">(</span>
|
|
<span class="n">request</span><span class="o">.</span><span class="n">prompt_tuning_config</span><span class="p">[</span><span class="mi">0</span><span class="p">])</span>
|
|
|
|
<span class="k">assert</span> <span class="n">request</span><span class="o">.</span><span class="n">id</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
|
|
<span class="k">try</span><span class="p">:</span>
|
|
<span class="n">executor_request</span> <span class="o">=</span> <span class="n">tllm</span><span class="o">.</span><span class="n">Request</span><span class="p">(</span>
|
|
<span class="n">client_id</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">id</span><span class="p">,</span>
|
|
<span class="n">input_token_ids</span><span class="o">=</span><span class="n">prompt_token_ids</span><span class="p">,</span>
|
|
<span class="n">max_tokens</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">max_tokens</span><span class="p">,</span>
|
|
<span class="n">max_new_tokens</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">max_new_tokens</span><span class="p">,</span>
|
|
<span class="n">streaming</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">streaming</span><span class="p">,</span>
|
|
<span class="n">sampling_config</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">_get_sampling_config</span><span class="p">(),</span>
|
|
<span class="n">end_id</span><span class="o">=-</span><span class="mi">1</span> <span class="k">if</span> <span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">ignore_eos</span> <span class="k">else</span>
|
|
<span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">end_id</span><span class="p">,</span>
|
|
<span class="n">pad_id</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">pad_id</span><span class="p">,</span>
|
|
<span class="n">output_config</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">_get_output_config</span><span class="p">(),</span>
|
|
<span class="n">lookahead_config</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">lookahead_config</span><span class="p">,</span>
|
|
<span class="n">guided_decoding_params</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span>
|
|
<span class="n">_get_guided_decoding_params</span><span class="p">(),</span>
|
|
<span class="n">bad_words</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">_get_bad_words</span><span class="p">(),</span>
|
|
<span class="n">stop_words</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">_get_stop_words</span><span class="p">(),</span>
|
|
<span class="n">embedding_bias</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span><span class="n">embedding_bias</span><span class="p">,</span>
|
|
<span class="n">external_draft_tokens_config</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span>
|
|
<span class="n">external_draft_tokens_config</span><span class="p">,</span>
|
|
<span class="n">lora_config</span><span class="o">=</span><span class="n">lora_config</span><span class="p">,</span>
|
|
<span class="n">prompt_tuning_config</span><span class="o">=</span><span class="n">prompt_tuning_config</span><span class="p">,</span>
|
|
<span class="n">logits_post_processor_name</span><span class="o">=</span><span class="n">request</span><span class="o">.</span><span class="n">sampling_params</span><span class="o">.</span>
|
|
<span class="n">logits_post_processor_name</span><span class="p">,</span>
|
|
<span class="p">)</span>
|
|
<span class="k">if</span> <span class="n">request</span><span class="o">.</span><span class="n">query_token_ids</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="c1"># pytorch star attention workflow</span>
|
|
<span class="c1"># a workaround to avoid public interface update</span>
|
|
<span class="n">req_id</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">enqueue_request</span><span class="p">(</span><span class="n">executor_request</span><span class="p">,</span>
|
|
<span class="n">request</span><span class="o">.</span><span class="n">query_token_ids</span><span class="p">)</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="n">req_id</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">enqueue_request</span><span class="p">(</span><span class="n">executor_request</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="n">req_id</span>
|
|
<span class="k">except</span> <span class="ne">Exception</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
|
|
<span class="k">raise</span> <span class="n">RequestError</span><span class="p">(</span><span class="nb">str</span><span class="p">(</span><span class="n">e</span><span class="p">))</span>
|
|
|
|
<span class="k">def</span> <span class="nf">submit</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">request</span><span class="p">:</span> <span class="n">GenerationRequest</span><span class="p">)</span> <span class="o">-></span> <span class="n">GenerationResult</span><span class="p">:</span>
|
|
<span class="w"> </span><span class="sd">""" Low-level API to the executor. Return a "future" GenerationResult which can be waited. """</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">start</span><span class="p">()</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">rank</span> <span class="o">!=</span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span>
|
|
<span class="s2">"Only rank 0 can submit requests.</span><span class="se">\n</span><span class="s2">"</span>
|
|
<span class="s2">"To fix this, ensure that the llm.generate(...) method is "</span>
|
|
<span class="s2">"guarded with the `if __name__ == '__main__':` block."</span><span class="p">)</span>
|
|
|
|
<span class="n">client_id</span> <span class="o">=</span> <span class="n">request</span><span class="o">.</span><span class="n">id</span> <span class="k">if</span> <span class="n">request</span><span class="o">.</span><span class="n">id</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="k">else</span> <span class="bp">self</span><span class="o">.</span><span class="n">_get_next_client_id</span><span class="p">(</span>
|
|
<span class="p">)</span>
|
|
<span class="k">if</span> <span class="n">request</span><span class="o">.</span><span class="n">id</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="n">request</span><span class="o">.</span><span class="n">set_id</span><span class="p">(</span><span class="n">client_id</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_enqueue_request</span><span class="p">(</span><span class="n">request</span><span class="p">)</span>
|
|
|
|
<span class="n">result</span> <span class="o">=</span> <span class="n">GenerationResult</span><span class="p">(</span>
|
|
<span class="n">request</span><span class="p">,</span> <span class="n">background_error_handler</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_handle_background_error</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="p">[</span><span class="n">client_id</span><span class="p">]</span> <span class="o">=</span> <span class="n">result</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_handle_background_error</span><span class="p">()</span>
|
|
|
|
<span class="k">return</span> <span class="n">result</span>
|
|
|
|
<span class="k">def</span> <span class="nf">shutdown</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="n">enable_llm_debug</span><span class="p">():</span>
|
|
<span class="k">try</span><span class="p">:</span>
|
|
<span class="n">print_colored</span><span class="p">(</span><span class="s1">'Proxy.shutdown...</span><span class="se">\n</span><span class="s1">'</span><span class="p">,</span> <span class="s2">"yellow"</span><span class="p">)</span>
|
|
<span class="nb">print</span><span class="p">(</span><span class="n">traceback</span><span class="o">.</span><span class="n">extract_stack</span><span class="p">())</span>
|
|
<span class="k">except</span> <span class="ne">ValueError</span><span class="p">:</span>
|
|
<span class="k">pass</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">doing_shutdown</span><span class="p">:</span>
|
|
<span class="k">return</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">doing_shutdown</span> <span class="o">=</span> <span class="kc">True</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">engine</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">can_enqueue_requests</span><span class="p">():</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">await_response_thread</span><span class="o">.</span><span class="n">is_alive</span><span class="p">():</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">await_response_thread</span><span class="o">.</span><span class="n">stop</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">await_response_thread</span><span class="o">.</span><span class="n">join</span><span class="p">()</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span><span class="o">.</span><span class="n">is_alive</span><span class="p">():</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span><span class="o">.</span><span class="n">stop</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span><span class="o">.</span><span class="n">join</span><span class="p">()</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">engine</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="c1"># Check if there are any errors from the threads before shutdown.</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_handle_background_error</span><span class="p">()</span>
|
|
|
|
<span class="k">def</span> <span class="nf">block_subordinates</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">rank</span> <span class="o">!=</span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="p">,</span> <span class="n">tllm</span><span class="o">.</span><span class="n">Executor</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
|
|
<span class="k">raise</span> <span class="bp">self</span><span class="o">.</span><span class="n">WorkerExit</span><span class="p">(</span>
|
|
<span class="s2">"block_subordinates() should be used in a `with ExecutorBindingsWorker() as ...:` block"</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="c1"># PIVOT_TO_PYTHON_START</span>
|
|
<span class="kn">from</span> <span class="nn">tensorrt_llm.pyexecutor.py_executor</span> <span class="kn">import</span> <span class="n">PyExecutor</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="p">,</span> <span class="n">PyExecutor</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">engine</span><span class="o">.</span><span class="n">wait_shutdown</span><span class="p">()</span>
|
|
<span class="c1"># PIVOT_TO_PYTHON_END</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__enter__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="bp">self</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__exit__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">exc_type</span><span class="p">,</span> <span class="n">exc_value</span><span class="p">,</span> <span class="n">traceback</span><span class="p">)</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
|
|
<span class="k">return</span> <span class="n">exc_type</span> <span class="ow">is</span> <span class="kc">None</span> <span class="ow">or</span> <span class="n">exc_type</span> <span class="o">==</span> <span class="n">ExecutorBindingsWorker</span><span class="o">.</span><span class="n">WorkerExit</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__del__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
|
|
|
|
|
|
<span class="k">class</span> <span class="nc">ZeroMqQueue</span><span class="p">:</span>
|
|
<span class="w"> </span><span class="sd">''' A Queue-like container for IPC using ZeroMQ. '''</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">address</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span> <span class="o">*</span><span class="p">,</span> <span class="n">is_server</span><span class="p">:</span> <span class="nb">bool</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">'''</span>
|
|
<span class="sd"> Parameters:</span>
|
|
<span class="sd"> address (Tuple[str, str], optional): The address (tcp-ip_port, authkey) for the IPC. Defaults to None.</span>
|
|
<span class="sd"> is_server (bool): Whether the current process is the server or the client.</span>
|
|
<span class="sd"> '''</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">address</span> <span class="o">=</span> <span class="n">address</span> <span class="ow">or</span> <span class="s2">"tcp://127.0.0.1:*"</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">is_server</span> <span class="o">=</span> <span class="n">is_server</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">context</span> <span class="o">=</span> <span class="n">zmq</span><span class="o">.</span><span class="n">Context</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">poller</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">socket</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_setup_done</span> <span class="o">=</span> <span class="kc">False</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">socket</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">context</span><span class="o">.</span><span class="n">socket</span><span class="p">(</span><span class="n">zmq</span><span class="o">.</span><span class="n">PAIR</span><span class="p">)</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">is_server</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">socket</span><span class="o">.</span><span class="n">bind</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">address</span>
|
|
<span class="p">)</span> <span class="c1"># Binds to the address and occupy a port immediately</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">address</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">socket</span><span class="o">.</span><span class="n">getsockopt</span><span class="p">(</span><span class="n">zmq</span><span class="o">.</span><span class="n">LAST_ENDPOINT</span><span class="p">)</span><span class="o">.</span><span class="n">decode</span><span class="p">()</span>
|
|
|
|
<span class="k">def</span> <span class="nf">setup_lazily</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_setup_done</span><span class="p">:</span>
|
|
<span class="k">return</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_setup_done</span> <span class="o">=</span> <span class="kc">True</span>
|
|
|
|
<span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">is_server</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">socket</span><span class="o">.</span><span class="n">connect</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">address</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">poller</span> <span class="o">=</span> <span class="n">zmq</span><span class="o">.</span><span class="n">Poller</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">poller</span><span class="o">.</span><span class="n">register</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">socket</span><span class="p">,</span> <span class="n">zmq</span><span class="o">.</span><span class="n">POLLIN</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">poll</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">timeout</span><span class="p">:</span> <span class="nb">int</span><span class="p">)</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
|
|
<span class="w"> </span><span class="sd">"""</span>
|
|
<span class="sd"> Parameters:</span>
|
|
<span class="sd"> timeout (int): Timeout in seconds</span>
|
|
<span class="sd"> """</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">setup_lazily</span><span class="p">()</span>
|
|
|
|
<span class="n">events</span> <span class="o">=</span> <span class="nb">dict</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">poller</span><span class="o">.</span><span class="n">poll</span><span class="p">(</span><span class="n">timeout</span><span class="o">=</span><span class="n">timeout</span> <span class="o">*</span> <span class="mi">1000</span><span class="p">))</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">socket</span> <span class="ow">in</span> <span class="n">events</span> <span class="ow">and</span> <span class="n">events</span><span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">socket</span><span class="p">]</span> <span class="o">==</span> <span class="n">zmq</span><span class="o">.</span><span class="n">POLLIN</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="kc">True</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="kc">False</span>
|
|
|
|
<span class="k">def</span> <span class="nf">put</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">obj</span><span class="p">:</span> <span class="n">Any</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">setup_lazily</span><span class="p">()</span>
|
|
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">obj</span><span class="p">,</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">):</span>
|
|
<span class="n">tensors</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_store_tensors_in_shmm</span><span class="p">(</span><span class="n">obj</span><span class="o">.</span><span class="n">tensors</span><span class="p">)</span>
|
|
<span class="n">obj</span> <span class="o">=</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">(</span><span class="n">client_id</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">client_id</span><span class="p">,</span>
|
|
<span class="n">tensors</span><span class="o">=</span><span class="n">tensors</span><span class="p">,</span>
|
|
<span class="n">finish_reasons</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">finish_reasons</span><span class="p">,</span>
|
|
<span class="n">is_final</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">is_final</span><span class="p">,</span>
|
|
<span class="n">error</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">error</span><span class="p">)</span>
|
|
|
|
<span class="n">message</span> <span class="o">=</span> <span class="n">pickle</span><span class="o">.</span><span class="n">dumps</span><span class="p">(</span><span class="n">obj</span><span class="p">)</span> <span class="c1"># nosec B301</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">socket</span><span class="o">.</span><span class="n">send</span><span class="p">(</span><span class="n">message</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">get</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="n">Any</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">setup_lazily</span><span class="p">()</span>
|
|
|
|
<span class="n">message</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">socket</span><span class="o">.</span><span class="n">recv</span><span class="p">()</span>
|
|
<span class="n">obj</span> <span class="o">=</span> <span class="n">pickle</span><span class="o">.</span><span class="n">loads</span><span class="p">(</span><span class="n">message</span><span class="p">)</span> <span class="c1"># nosec B301</span>
|
|
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">obj</span><span class="p">,</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">):</span>
|
|
<span class="n">tensors</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_load_tensors_from_shmm</span><span class="p">(</span><span class="n">obj</span><span class="o">.</span><span class="n">tensors</span><span class="p">)</span>
|
|
<span class="n">obj</span> <span class="o">=</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">(</span><span class="n">client_id</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">client_id</span><span class="p">,</span>
|
|
<span class="n">tensors</span><span class="o">=</span><span class="n">tensors</span><span class="p">,</span>
|
|
<span class="n">finish_reasons</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">finish_reasons</span><span class="p">,</span>
|
|
<span class="n">is_final</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">is_final</span><span class="p">,</span>
|
|
<span class="n">error</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">error</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="n">obj</span>
|
|
|
|
<span class="k">def</span> <span class="nf">close</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">socket</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">socket</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">socket</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">context</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">context</span><span class="o">.</span><span class="n">term</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">context</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="k">def</span> <span class="nf">_store_tensors_in_shmm</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="p">,</span> <span class="n">tensors</span><span class="p">:</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">ResponseTensors</span>
|
|
<span class="p">)</span> <span class="o">-></span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">ResponseTensors</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="n">tensors</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="n">tensors</span>
|
|
|
|
<span class="c1"># The tensors are huge and cannot be transferred through socket directly. We need to store them in shared memory,</span>
|
|
<span class="c1"># and replace the tensors with the shared memory path.</span>
|
|
<span class="k">def</span> <span class="nf">store_tensor</span><span class="p">(</span><span class="n">tensor</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">])</span> <span class="o">-></span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]:</span>
|
|
<span class="k">if</span> <span class="n">tensor</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="kc">None</span>
|
|
<span class="c1"># NOTE: We create random shmm here rather than two specific shmm for context and generation logit, since the</span>
|
|
<span class="c1"># shmm may not be read timely by the IpcQueue.get() in the other side, so there might be multiple alive shmm</span>
|
|
<span class="c1"># for logits.</span>
|
|
<span class="c1"># A known issue: the shmm instance may leak if the IpcQueue.get() thread is stopped before the IpcQueue.put()</span>
|
|
<span class="c1"># thread. This is not a big issue since the shmm will be automatically cleaned up when the process exits.</span>
|
|
<span class="n">shm</span> <span class="o">=</span> <span class="n">SharedMemory</span><span class="p">(</span><span class="n">create</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">size</span><span class="o">=</span><span class="n">tensor</span><span class="o">.</span><span class="n">nbytes</span> <span class="o">+</span> <span class="mi">2048</span><span class="p">)</span>
|
|
<span class="n">torch</span><span class="o">.</span><span class="n">save</span><span class="p">(</span><span class="n">tensor</span><span class="p">,</span> <span class="n">shm</span><span class="o">.</span><span class="n">_mmap</span><span class="p">)</span>
|
|
<span class="n">shm</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
|
|
<span class="k">return</span> <span class="n">shm</span><span class="o">.</span><span class="n">name</span>
|
|
|
|
<span class="k">return</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">ResponseTensors</span><span class="p">(</span>
|
|
<span class="n">output_token_ids</span><span class="o">=</span><span class="n">tensors</span><span class="o">.</span><span class="n">output_token_ids</span><span class="p">,</span>
|
|
<span class="n">context_logits</span><span class="o">=</span><span class="n">store_tensor</span><span class="p">(</span><span class="n">tensors</span><span class="o">.</span><span class="n">context_logits</span><span class="p">),</span>
|
|
<span class="n">generation_logits</span><span class="o">=</span><span class="n">store_tensor</span><span class="p">(</span><span class="n">tensors</span><span class="o">.</span><span class="n">generation_logits</span><span class="p">),</span>
|
|
<span class="n">log_probs</span><span class="o">=</span><span class="n">tensors</span><span class="o">.</span><span class="n">log_probs</span><span class="p">,</span>
|
|
<span class="n">cum_log_probs</span><span class="o">=</span><span class="n">tensors</span><span class="o">.</span><span class="n">cum_log_probs</span><span class="p">,</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">_load_tensors_from_shmm</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="p">,</span> <span class="n">tensors</span><span class="p">:</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">ResponseTensors</span>
|
|
<span class="p">)</span> <span class="o">-></span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">ResponseTensors</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="n">tensors</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="n">tensors</span>
|
|
|
|
<span class="k">def</span> <span class="nf">load_tensor</span><span class="p">(</span><span class="n">tensor</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">])</span> <span class="o">-></span> <span class="n">Optional</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">]:</span>
|
|
<span class="k">if</span> <span class="n">tensor</span> <span class="ow">is</span> <span class="kc">None</span> <span class="ow">or</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">tensor</span><span class="p">,</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="n">tensor</span>
|
|
|
|
<span class="n">shm</span> <span class="o">=</span> <span class="n">SharedMemory</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="n">tensor</span><span class="p">,</span> <span class="n">create</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
|
|
<span class="n">tensor</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="n">io</span><span class="o">.</span><span class="n">BytesIO</span><span class="p">(</span><span class="n">shm</span><span class="o">.</span><span class="n">buf</span><span class="p">))</span>
|
|
<span class="n">shm</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
|
|
<span class="n">shm</span><span class="o">.</span><span class="n">unlink</span><span class="p">()</span>
|
|
<span class="k">return</span> <span class="n">tensor</span>
|
|
|
|
<span class="k">return</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">ResponseTensors</span><span class="p">(</span>
|
|
<span class="n">output_token_ids</span><span class="o">=</span><span class="n">tensors</span><span class="o">.</span><span class="n">output_token_ids</span><span class="p">,</span>
|
|
<span class="n">context_logits</span><span class="o">=</span><span class="n">load_tensor</span><span class="p">(</span><span class="n">tensors</span><span class="o">.</span><span class="n">context_logits</span><span class="p">),</span>
|
|
<span class="n">generation_logits</span><span class="o">=</span><span class="n">load_tensor</span><span class="p">(</span><span class="n">tensors</span><span class="o">.</span><span class="n">generation_logits</span><span class="p">),</span>
|
|
<span class="n">log_probs</span><span class="o">=</span><span class="n">tensors</span><span class="o">.</span><span class="n">log_probs</span><span class="p">,</span>
|
|
<span class="n">cum_log_probs</span><span class="o">=</span><span class="n">tensors</span><span class="o">.</span><span class="n">cum_log_probs</span><span class="p">,</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__del__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
|
|
|
|
|
|
<span class="n">IpcQueue</span> <span class="o">=</span> <span class="n">ZeroMqQueue</span>
|
|
|
|
|
|
<span class="k">class</span> <span class="nc">FusedIpcQueue</span><span class="p">:</span>
|
|
<span class="w"> </span><span class="sd">''' A Queue-like container for IPC with optional message batched. '''</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span>
|
|
<span class="n">address</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="o">*</span><span class="p">,</span>
|
|
<span class="n">is_server</span><span class="p">:</span> <span class="nb">bool</span><span class="p">,</span>
|
|
<span class="n">fuse_message</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">fuse_size</span><span class="o">=</span><span class="mi">100000</span><span class="p">,</span>
|
|
<span class="n">error_queue</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">queue_cls</span><span class="o">=</span><span class="n">ZeroMqQueue</span><span class="p">):</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">queue</span> <span class="o">=</span> <span class="n">queue_cls</span><span class="p">(</span><span class="n">address</span><span class="o">=</span><span class="n">address</span><span class="p">,</span> <span class="n">is_server</span><span class="o">=</span><span class="n">is_server</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">fuse_message</span> <span class="o">=</span> <span class="n">fuse_message</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">error_queue</span> <span class="o">=</span> <span class="n">error_queue</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">fuse_size</span> <span class="o">=</span> <span class="n">fuse_size</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_message_counter</span> <span class="o">=</span> <span class="mi">0</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_obj_counter</span> <span class="o">=</span> <span class="mi">0</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_send_thread</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">sending_queue</span> <span class="o">=</span> <span class="n">Queue</span><span class="p">()</span> <span class="k">if</span> <span class="n">fuse_message</span> <span class="k">else</span> <span class="kc">None</span>
|
|
|
|
<span class="k">def</span> <span class="nf">setup_sender</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">fuse_message</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">_send_thread</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="k">return</span>
|
|
|
|
<span class="k">def</span> <span class="nf">send_task</span><span class="p">():</span>
|
|
<span class="k">while</span> <span class="kc">True</span><span class="p">:</span>
|
|
<span class="n">qsize</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">sending_queue</span><span class="o">.</span><span class="n">qsize</span><span class="p">()</span>
|
|
<span class="k">if</span> <span class="n">qsize</span> <span class="o">></span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="n">qsize</span> <span class="o">=</span> <span class="nb">min</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">fuse_size</span><span class="p">,</span> <span class="n">qsize</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_obj_counter</span> <span class="o">+=</span> <span class="n">qsize</span>
|
|
<span class="n">message</span> <span class="o">=</span> <span class="p">[</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">sending_queue</span><span class="o">.</span><span class="n">get_nowait</span><span class="p">()</span> <span class="k">for</span> <span class="n">_</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="n">qsize</span><span class="p">)</span>
|
|
<span class="p">]</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">message</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_message_counter</span> <span class="o">+=</span> <span class="mi">1</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="n">time</span><span class="o">.</span><span class="n">sleep</span><span class="p">(</span><span class="mf">0.001</span><span class="p">)</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_send_thread</span> <span class="o">=</span> <span class="n">ManagedThread</span><span class="p">(</span><span class="n">send_task</span><span class="p">,</span>
|
|
<span class="n">name</span><span class="o">=</span><span class="s2">"fused_send_thread"</span><span class="p">,</span>
|
|
<span class="n">error_queue</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">error_queue</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_send_thread</span><span class="o">.</span><span class="n">start</span><span class="p">()</span>
|
|
|
|
<span class="k">def</span> <span class="nf">put</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">obj</span><span class="p">:</span> <span class="n">Any</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">setup_sender</span><span class="p">()</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">fuse_message</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">sending_queue</span><span class="o">.</span><span class="n">put_nowait</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_prepare_message</span><span class="p">(</span><span class="n">obj</span><span class="p">))</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_prepare_message</span><span class="p">(</span><span class="n">obj</span><span class="p">))</span>
|
|
|
|
<span class="k">def</span> <span class="nf">get</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="n">Any</span><span class="p">:</span>
|
|
<span class="n">obj</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">obj</span><span class="p">,</span> <span class="nb">list</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">_process_message</span><span class="p">(</span><span class="n">o</span><span class="p">)</span> <span class="k">for</span> <span class="n">o</span> <span class="ow">in</span> <span class="n">obj</span><span class="p">]</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_process_message</span><span class="p">(</span><span class="n">obj</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">_prepare_message</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">obj</span><span class="p">:</span> <span class="n">Any</span><span class="p">)</span> <span class="o">-></span> <span class="n">Any</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">obj</span><span class="p">,</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">):</span>
|
|
<span class="n">tensors</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">queue</span><span class="o">.</span><span class="n">_store_tensors_in_shmm</span><span class="p">(</span><span class="n">obj</span><span class="o">.</span><span class="n">tensors</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">(</span>
|
|
<span class="n">client_id</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">client_id</span><span class="p">,</span>
|
|
<span class="n">tensors</span><span class="o">=</span><span class="n">tensors</span><span class="p">,</span>
|
|
<span class="n">finish_reasons</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">finish_reasons</span><span class="p">,</span>
|
|
<span class="n">is_final</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">is_final</span><span class="p">,</span>
|
|
<span class="n">sequence_index</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">sequence_index</span><span class="p">,</span>
|
|
<span class="n">error</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">error</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="n">obj</span>
|
|
|
|
<span class="k">def</span> <span class="nf">_process_message</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">obj</span><span class="p">:</span> <span class="n">Any</span><span class="p">)</span> <span class="o">-></span> <span class="n">Any</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">obj</span><span class="p">,</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">):</span>
|
|
<span class="n">tensors</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">queue</span><span class="o">.</span><span class="n">_load_tensors_from_shmm</span><span class="p">(</span><span class="n">obj</span><span class="o">.</span><span class="n">tensors</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="n">GenerationExecutor</span><span class="o">.</span><span class="n">Response</span><span class="p">(</span>
|
|
<span class="n">client_id</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">client_id</span><span class="p">,</span>
|
|
<span class="n">tensors</span><span class="o">=</span><span class="n">tensors</span><span class="p">,</span>
|
|
<span class="n">finish_reasons</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">finish_reasons</span><span class="p">,</span>
|
|
<span class="n">is_final</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">is_final</span><span class="p">,</span>
|
|
<span class="n">sequence_index</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">sequence_index</span><span class="p">,</span>
|
|
<span class="n">error</span><span class="o">=</span><span class="n">obj</span><span class="o">.</span><span class="n">error</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="n">obj</span>
|
|
|
|
<span class="nd">@property</span>
|
|
<span class="k">def</span> <span class="nf">address</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="n">Tuple</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">int</span><span class="p">,</span> <span class="nb">bytes</span><span class="p">]:</span>
|
|
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">queue</span><span class="o">.</span><span class="n">address</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__del__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
|
|
|
|
<span class="k">def</span> <span class="nf">print_fuse_stats</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_message_counter</span> <span class="o">></span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="n">print_colored</span><span class="p">(</span>
|
|
<span class="sa">f</span><span class="s2">"IPCQueue: </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">_message_counter</span><span class="si">}</span><span class="s2"> messages, </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">_obj_counter</span><span class="si">}</span><span class="s2"> objects sent, average: </span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">_obj_counter</span><span class="o">/</span><span class="bp">self</span><span class="o">.</span><span class="n">_message_counter</span><span class="si">}</span><span class="s2">.</span><span class="se">\n</span><span class="s2">"</span><span class="p">,</span>
|
|
<span class="s2">"green"</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">close</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">queue</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">_send_thread</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_send_thread</span><span class="o">.</span><span class="n">stop</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_send_thread</span><span class="o">.</span><span class="n">join</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_send_thread</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="k">if</span> <span class="n">enable_llm_debug</span><span class="p">():</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">print_fuse_stats</span><span class="p">()</span>
|
|
|
|
|
|
<span class="k">class</span> <span class="nc">ExecutorBindingsProxy</span><span class="p">(</span><span class="n">GenerationExecutor</span><span class="p">):</span>
|
|
<span class="n">READY_SIGNAL</span> <span class="o">=</span> <span class="sa">b</span><span class="s2">"READY"</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span>
|
|
<span class="n">workers_kwargs</span><span class="p">:</span> <span class="nb">dict</span><span class="p">,</span>
|
|
<span class="n">model_world_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">1</span><span class="p">,</span>
|
|
<span class="n">mpi_session</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">MpiSession</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="o">*</span><span class="p">,</span>
|
|
<span class="n">worker_cls</span><span class="p">:</span> <span class="nb">type</span> <span class="o">=</span> <span class="n">ExecutorBindingsWorker</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">workers_started</span> <span class="o">=</span> <span class="kc">False</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">worker_cls</span> <span class="o">=</span> <span class="n">worker_cls</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">request_queue</span> <span class="o">=</span> <span class="n">IpcQueue</span><span class="p">(</span><span class="n">is_server</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">request_error_queue</span> <span class="o">=</span> <span class="n">IpcQueue</span><span class="p">(</span><span class="n">is_server</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span> <span class="o">=</span> <span class="n">FusedIpcQueue</span><span class="p">(</span><span class="n">is_server</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">fuse_message</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">mp_stats_queue</span> <span class="o">=</span> <span class="n">FusedIpcQueue</span><span class="p">(</span><span class="n">is_server</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">fuse_message</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">int</span><span class="p">,</span> <span class="n">GenerationResult</span><span class="p">]</span> <span class="o">=</span> <span class="p">{}</span>
|
|
|
|
<span class="k">if</span> <span class="n">mpi_session</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span> <span class="o">=</span> <span class="n">MpiPoolSession</span><span class="p">(</span><span class="n">n_workers</span><span class="o">=</span><span class="n">model_world_size</span><span class="p">)</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span> <span class="o">=</span> <span class="n">mpi_session</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">model_world_size</span> <span class="o">=</span> <span class="n">model_world_size</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">workers_kwargs</span> <span class="o">=</span> <span class="nb">dict</span><span class="p">(</span>
|
|
<span class="o">**</span><span class="n">workers_kwargs</span><span class="p">,</span>
|
|
<span class="n">request_queue_addr</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">request_queue</span><span class="o">.</span><span class="n">address</span><span class="p">,</span>
|
|
<span class="n">request_error_queue_addr</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">request_error_queue</span><span class="o">.</span><span class="n">address</span><span class="p">,</span>
|
|
<span class="n">result_queue_addr</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span><span class="o">.</span><span class="n">address</span><span class="p">,</span>
|
|
<span class="n">stats_queue_addr</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">mp_stats_queue</span><span class="o">.</span><span class="n">address</span><span class="p">,</span>
|
|
<span class="p">)</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_result_thread</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">ManagedThread</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">ManagedThread</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_start_executor_workers</span><span class="p">()</span>
|
|
|
|
<span class="nd">@staticmethod</span>
|
|
<span class="k">def</span> <span class="nf">workers_main</span><span class="p">(</span><span class="n">engine</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">Path</span><span class="p">,</span> <span class="n">Engine</span><span class="p">],</span>
|
|
<span class="n">request_queue_addr</span><span class="p">:</span> <span class="n">Tuple</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">int</span><span class="p">,</span> <span class="nb">bytes</span><span class="p">],</span>
|
|
<span class="n">request_error_queue_addr</span><span class="p">:</span> <span class="n">Tuple</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">int</span><span class="p">,</span> <span class="nb">bytes</span><span class="p">],</span>
|
|
<span class="n">result_queue_addr</span><span class="p">:</span> <span class="n">Tuple</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">int</span><span class="p">,</span> <span class="nb">bytes</span><span class="p">],</span>
|
|
<span class="n">stats_queue_addr</span><span class="p">:</span> <span class="n">Tuple</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">int</span><span class="p">,</span> <span class="nb">bytes</span><span class="p">],</span>
|
|
<span class="n">executor_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">tllm</span><span class="o">.</span><span class="n">ExecutorConfig</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">logits_post_processor_map</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span>
|
|
<span class="n">Callable</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">worker_cls</span><span class="p">:</span> <span class="nb">type</span> <span class="o">=</span> <span class="n">ExecutorBindingsWorker</span><span class="p">,</span>
|
|
<span class="n">tracer_init_kwargs</span><span class="o">=</span><span class="kc">None</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="n">result_queue</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="k">if</span> <span class="n">tracer_init_kwargs</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="n">mpi_rank</span><span class="p">()</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="n">tracer</span> <span class="o">=</span> <span class="n">VizTracer</span><span class="p">(</span><span class="o">**</span><span class="n">tracer_init_kwargs</span><span class="p">)</span>
|
|
<span class="n">tracer</span><span class="o">.</span><span class="n">register_exit</span><span class="p">()</span>
|
|
<span class="n">tracer</span><span class="o">.</span><span class="n">start</span><span class="p">()</span>
|
|
<span class="n">set_global_tracer</span><span class="p">(</span><span class="n">tracer</span><span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="n">mpi_rank</span><span class="p">()</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="n">request_queue</span> <span class="o">=</span> <span class="n">IpcQueue</span><span class="p">(</span><span class="n">request_queue_addr</span><span class="p">,</span> <span class="n">is_server</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
|
|
<span class="n">request_error_queue</span> <span class="o">=</span> <span class="n">IpcQueue</span><span class="p">(</span><span class="n">request_error_queue_addr</span><span class="p">,</span>
|
|
<span class="n">is_server</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
|
|
<span class="n">result_queue</span> <span class="o">=</span> <span class="n">FusedIpcQueue</span><span class="p">(</span><span class="n">result_queue_addr</span><span class="p">,</span>
|
|
<span class="n">is_server</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">fuse_message</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
|
|
<span class="n">mp_stats_queue</span> <span class="o">=</span> <span class="n">FusedIpcQueue</span><span class="p">(</span><span class="n">stats_queue_addr</span><span class="p">,</span>
|
|
<span class="n">is_server</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">fuse_message</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">notify_proxy_threads_to_quit</span><span class="p">():</span>
|
|
<span class="c1"># Signal the dispatcher thread in the proxy to quit</span>
|
|
<span class="n">result_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="kc">None</span><span class="p">)</span>
|
|
<span class="c1"># Signal the stats thread in the proxy to quit</span>
|
|
<span class="n">mp_stats_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="kc">None</span><span class="p">)</span>
|
|
|
|
<span class="c1"># Error handling in the Worker/MPI process</span>
|
|
<span class="c1"># 1. During Executor initialization, the errors will be captured and send back via request_error_queue.</span>
|
|
<span class="c1"># 2. During execution, the errors will be captured by ManagedThreads</span>
|
|
<span class="c1"># a) For per-request error, the error will be send back via result_queue, and eventually raised in</span>
|
|
<span class="c1"># handle_response() in the main thread.</span>
|
|
<span class="c1"># b) For system error, the error will be raised in the MPI process and handled by future.done_callback,</span>
|
|
<span class="c1"># that will propagate the error to the error_queue in the main thread.</span>
|
|
|
|
<span class="k">try</span><span class="p">:</span>
|
|
<span class="n">executor</span> <span class="o">=</span> <span class="n">worker_cls</span><span class="p">(</span><span class="n">engine</span><span class="p">,</span> <span class="n">executor_config</span><span class="p">,</span>
|
|
<span class="n">logits_post_processor_map</span><span class="p">)</span>
|
|
<span class="k">except</span> <span class="ne">Exception</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="n">mpi_rank</span><span class="p">()</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="n">request_error_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">e</span><span class="p">)</span>
|
|
<span class="k">return</span>
|
|
|
|
<span class="k">with</span> <span class="n">executor</span><span class="p">:</span>
|
|
<span class="k">try</span><span class="p">:</span>
|
|
<span class="n">executor</span><span class="o">.</span><span class="n">block_subordinates</span><span class="p">()</span>
|
|
|
|
<span class="k">if</span> <span class="n">mpi_rank</span><span class="p">()</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="n">executor</span><span class="o">.</span><span class="n">set_result_queue</span><span class="p">(</span><span class="n">result_queue</span><span class="p">)</span>
|
|
<span class="n">executor</span><span class="o">.</span><span class="n">set_stats_queue</span><span class="p">(</span><span class="n">mp_stats_queue</span><span class="p">)</span>
|
|
<span class="n">request_error_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">ExecutorBindingsProxy</span><span class="o">.</span><span class="n">READY_SIGNAL</span><span class="p">)</span>
|
|
<span class="k">while</span> <span class="p">(</span><span class="n">req</span> <span class="o">:=</span> <span class="n">request_queue</span><span class="o">.</span><span class="n">get</span><span class="p">())</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="k">try</span><span class="p">:</span>
|
|
<span class="n">result</span> <span class="o">=</span> <span class="n">executor</span><span class="o">.</span><span class="n">submit</span><span class="p">(</span><span class="n">req</span><span class="p">)</span>
|
|
<span class="n">request_error_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="kc">None</span><span class="p">)</span> <span class="c1"># None means success</span>
|
|
<span class="k">except</span> <span class="n">RequestError</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
|
|
<span class="n">request_error_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">e</span><span class="p">)</span>
|
|
|
|
<span class="n">notify_proxy_threads_to_quit</span><span class="p">()</span>
|
|
|
|
<span class="k">except</span> <span class="n">ExecutorBindingsWorker</span><span class="o">.</span><span class="n">WorkerExit</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
|
|
<span class="c1"># This will capture by the with-statement and exit normally.</span>
|
|
<span class="k">raise</span> <span class="n">e</span>
|
|
|
|
<span class="k">except</span> <span class="ne">Exception</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span> <span class="c1"># other critical errors</span>
|
|
<span class="k">if</span> <span class="n">mpi_rank</span><span class="p">()</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="n">notify_proxy_threads_to_quit</span><span class="p">()</span>
|
|
<span class="n">err</span> <span class="o">=</span> <span class="n">CppExecutorError</span><span class="p">(</span><span class="sa">f</span><span class="s2">"Failed during generation: </span><span class="si">{</span><span class="n">e</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span>
|
|
<span class="k">if</span> <span class="n">mpi_rank</span><span class="p">()</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="n">request_error_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">err</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">dispatch_result_task</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="p">(</span><span class="n">res</span> <span class="o">:=</span> <span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span><span class="o">.</span><span class="n">get</span><span class="p">())</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="kc">False</span> <span class="c1"># shutdown the thread</span>
|
|
|
|
<span class="n">async_queues</span> <span class="o">=</span> <span class="p">[]</span>
|
|
<span class="n">event_loop</span> <span class="o">=</span> <span class="kc">None</span>
|
|
|
|
<span class="k">def</span> <span class="nf">process_res</span><span class="p">(</span><span class="n">res</span><span class="p">):</span>
|
|
<span class="n">client_id</span> <span class="o">=</span> <span class="n">res</span><span class="o">.</span><span class="n">client_id</span>
|
|
<span class="k">nonlocal</span> <span class="n">event_loop</span>
|
|
<span class="k">nonlocal</span> <span class="n">async_queues</span>
|
|
|
|
<span class="n">queue</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="p">[</span><span class="n">client_id</span><span class="p">]</span><span class="o">.</span><span class="n">queue</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">queue</span><span class="p">,</span> <span class="n">_SyncQueue</span><span class="p">):</span>
|
|
<span class="n">queue</span><span class="o">.</span><span class="n">put_nowait</span><span class="p">(</span><span class="n">res</span><span class="p">)</span>
|
|
<span class="n">async_queues</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">queue</span><span class="p">)</span>
|
|
<span class="c1"># all the loops are identical</span>
|
|
<span class="n">event_loop</span> <span class="o">=</span> <span class="n">event_loop</span> <span class="ow">or</span> <span class="n">queue</span><span class="o">.</span><span class="n">loop</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="n">queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">res</span><span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="n">res</span><span class="o">.</span><span class="n">is_final</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="o">.</span><span class="n">pop</span><span class="p">(</span><span class="n">client_id</span><span class="p">)</span>
|
|
|
|
<span class="n">res</span> <span class="o">=</span> <span class="n">res</span> <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">res</span><span class="p">,</span> <span class="nb">list</span><span class="p">)</span> <span class="k">else</span> <span class="p">[</span><span class="n">res</span><span class="p">]</span>
|
|
|
|
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="n">res</span><span class="p">:</span>
|
|
<span class="n">global_tracer</span><span class="p">()</span><span class="o">.</span><span class="n">log_instant</span><span class="p">(</span><span class="s2">"IPC.get"</span><span class="p">)</span>
|
|
<span class="k">if</span> <span class="n">i</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="kc">False</span>
|
|
<span class="n">process_res</span><span class="p">(</span><span class="n">i</span><span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="n">async_queues</span><span class="p">:</span>
|
|
<span class="n">_SyncQueue</span><span class="o">.</span><span class="n">notify_queues</span><span class="p">(</span><span class="n">event_loop</span><span class="p">,</span> <span class="n">async_queues</span><span class="p">)</span>
|
|
|
|
<span class="k">return</span> <span class="kc">True</span> <span class="c1"># success</span>
|
|
|
|
<span class="k">def</span> <span class="nf">dispatch_stats_task</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
|
|
<span class="c1"># get-stats is not urgent, so we can sleep a bit</span>
|
|
|
|
<span class="n">time</span><span class="o">.</span><span class="n">sleep</span><span class="p">(</span><span class="mf">0.1</span><span class="p">)</span>
|
|
|
|
<span class="k">try</span><span class="p">:</span>
|
|
<span class="n">stats</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">mp_stats_queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
|
|
<span class="k">except</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="kc">False</span>
|
|
|
|
<span class="k">if</span> <span class="n">stats</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="kc">False</span>
|
|
|
|
<span class="n">stats</span> <span class="o">=</span> <span class="n">stats</span> <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">stats</span><span class="p">,</span> <span class="nb">list</span><span class="p">)</span> <span class="k">else</span> <span class="p">[</span><span class="n">stats</span><span class="p">]</span>
|
|
|
|
<span class="k">while</span> <span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span><span class="o">.</span><span class="n">full</span><span class="p">():</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
|
|
|
|
<span class="k">try</span><span class="p">:</span>
|
|
<span class="k">for</span> <span class="n">s</span> <span class="ow">in</span> <span class="n">stats</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="n">s</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="kc">False</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">stats_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">s</span><span class="p">)</span>
|
|
<span class="k">except</span> <span class="n">AsyncQueue</span><span class="o">.</span><span class="n">EventLoopShutdownError</span><span class="p">:</span>
|
|
<span class="c1"># This happens in the last stats loop while the generate workflow is stopped.</span>
|
|
<span class="k">pass</span>
|
|
<span class="k">except</span> <span class="ne">Exception</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
|
|
<span class="k">raise</span> <span class="n">e</span>
|
|
|
|
<span class="k">return</span> <span class="kc">True</span> <span class="c1"># success</span>
|
|
|
|
<span class="k">def</span> <span class="nf">_start_dispatch_threads</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">dispatch_result_thread</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_result_thread</span> <span class="o">=</span> <span class="n">ManagedThread</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_result_task</span><span class="p">,</span>
|
|
<span class="n">error_queue</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="p">,</span>
|
|
<span class="n">name</span><span class="o">=</span><span class="s2">"proxy_dispatch_result_thread"</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span> <span class="o">=</span> <span class="n">ManagedThread</span><span class="p">(</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_task</span><span class="p">,</span>
|
|
<span class="n">error_queue</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="p">,</span>
|
|
<span class="n">name</span><span class="o">=</span><span class="s2">"proxy_dispatch_stats_thread"</span><span class="p">)</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_result_thread</span><span class="o">.</span><span class="n">start</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">create_stats_queue</span><span class="p">()</span>
|
|
<span class="c1"># TODO: clean up the stats thread, and replace with a decent get_stats API</span>
|
|
<span class="c1">#self.dispatch_stats_thread.start()</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_handle_background_error</span><span class="p">()</span>
|
|
|
|
<span class="k">def</span> <span class="nf">_start_executor_workers</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
|
|
<span class="k">def</span> <span class="nf">mpi_done_callback</span><span class="p">(</span><span class="n">future</span><span class="p">:</span> <span class="n">concurrent</span><span class="o">.</span><span class="n">futures</span><span class="o">.</span><span class="n">Future</span><span class="p">):</span>
|
|
<span class="c1"># This is called when the MPI worker is done, so future.exception() will not block.</span>
|
|
<span class="k">if</span> <span class="n">future</span><span class="o">.</span><span class="n">exception</span><span class="p">()</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_error_queue</span><span class="o">.</span><span class="n">put_nowait</span><span class="p">(</span><span class="n">future</span><span class="o">.</span><span class="n">exception</span><span class="p">())</span>
|
|
|
|
<span class="n">tracer_init_kwargs</span> <span class="o">=</span> <span class="n">get_tracer</span><span class="p">()</span><span class="o">.</span><span class="n">init_kwargs</span> <span class="k">if</span> <span class="n">enable_llm_tracer</span><span class="p">(</span>
|
|
<span class="p">)</span> <span class="k">else</span> <span class="kc">None</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_futures</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span><span class="o">.</span><span class="n">submit</span><span class="p">(</span>
|
|
<span class="n">ExecutorBindingsProxy</span><span class="o">.</span><span class="n">workers_main</span><span class="p">,</span>
|
|
<span class="o">**</span><span class="bp">self</span><span class="o">.</span><span class="n">workers_kwargs</span><span class="p">,</span>
|
|
<span class="n">worker_cls</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">worker_cls</span><span class="p">,</span>
|
|
<span class="n">tracer_init_kwargs</span><span class="o">=</span><span class="n">tracer_init_kwargs</span><span class="p">)</span>
|
|
<span class="k">for</span> <span class="n">fut</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">mpi_futures</span><span class="p">:</span>
|
|
<span class="n">fut</span><span class="o">.</span><span class="n">add_done_callback</span><span class="p">(</span><span class="n">mpi_done_callback</span><span class="p">)</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">workers_started</span> <span class="o">=</span> <span class="kc">True</span>
|
|
|
|
<span class="k">while</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">request_error_queue</span><span class="o">.</span><span class="n">poll</span><span class="p">(</span><span class="mi">1</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_handle_background_error</span><span class="p">()</span>
|
|
|
|
<span class="n">ready_signal</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">request_error_queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
|
|
<span class="k">if</span> <span class="n">ready_signal</span> <span class="o">!=</span> <span class="n">ExecutorBindingsProxy</span><span class="o">.</span><span class="n">READY_SIGNAL</span><span class="p">:</span>
|
|
<span class="k">raise</span> <span class="n">ready_signal</span>
|
|
|
|
<span class="k">def</span> <span class="nf">shutdown</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="n">enable_llm_debug</span><span class="p">():</span>
|
|
<span class="k">try</span><span class="p">:</span>
|
|
<span class="n">print_colored</span><span class="p">(</span><span class="s1">'Proxy.shutdown...</span><span class="se">\n</span><span class="s1">'</span><span class="p">,</span> <span class="s2">"yellow"</span><span class="p">)</span>
|
|
<span class="n">print_colored</span><span class="p">(</span><span class="nb">str</span><span class="p">(</span><span class="n">traceback</span><span class="o">.</span><span class="n">format_exc</span><span class="p">())</span> <span class="o">+</span> <span class="s2">"</span><span class="se">\n</span><span class="s2">"</span><span class="p">,</span> <span class="s2">"yellow"</span><span class="p">)</span>
|
|
<span class="k">except</span> <span class="ne">ValueError</span><span class="p">:</span>
|
|
<span class="k">pass</span>
|
|
<span class="k">if</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">workers_started</span><span class="p">:</span>
|
|
<span class="k">return</span>
|
|
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">doing_shutdown</span><span class="p">:</span>
|
|
<span class="k">return</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">doing_shutdown</span> <span class="o">=</span> <span class="kc">True</span>
|
|
|
|
<span class="c1"># step1: notify the workers to quit</span>
|
|
<span class="k">if</span> <span class="nb">all</span><span class="p">(</span><span class="ow">not</span> <span class="n">f</span><span class="o">.</span><span class="n">done</span><span class="p">()</span> <span class="k">for</span> <span class="n">f</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">mpi_futures</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">request_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="kc">None</span><span class="p">)</span>
|
|
|
|
<span class="k">for</span> <span class="n">f</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">mpi_futures</span><span class="p">:</span>
|
|
<span class="k">try</span><span class="p">:</span>
|
|
<span class="n">f</span><span class="o">.</span><span class="n">result</span><span class="p">()</span>
|
|
<span class="k">except</span><span class="p">:</span>
|
|
<span class="c1"># The errors are already captured in mpi_done_callback, ignored here</span>
|
|
<span class="k">pass</span>
|
|
|
|
<span class="c1"># step2: notify the background threads to quit</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">dispatch_result_thread</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">dispatch_result_thread</span><span class="o">.</span><span class="n">is_alive</span><span class="p">(</span>
|
|
<span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_result_thread</span><span class="o">.</span><span class="n">stop</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_result_thread</span><span class="o">.</span><span class="n">join</span><span class="p">()</span>
|
|
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span><span class="o">.</span><span class="n">is_alive</span><span class="p">(</span>
|
|
<span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span><span class="o">.</span><span class="n">stop</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">dispatch_stats_thread</span><span class="o">.</span><span class="n">join</span><span class="p">()</span>
|
|
|
|
<span class="c1"># step3: finish all remaining work</span>
|
|
|
|
<span class="c1"># close all the sockets</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">request_queue</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">request_error_queue</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">result_queue</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">mp_stats_queue</span><span class="o">.</span><span class="n">close</span><span class="p">()</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">workers_started</span> <span class="o">=</span> <span class="kc">False</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">mpi_session</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
|
|
|
|
<span class="c1"># Process the errors in-case error during shutting down the threads</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_handle_background_error</span><span class="p">()</span>
|
|
|
|
<span class="k">def</span> <span class="nf">submit</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">request</span><span class="p">:</span> <span class="n">GenerationRequest</span><span class="p">)</span> <span class="o">-></span> <span class="n">GenerationResult</span><span class="p">:</span>
|
|
<span class="w"> </span><span class="sd">"""</span>
|
|
<span class="sd"> Low-level API to the executor. Return a "future" GenerationResult which can be waited.</span>
|
|
<span class="sd"> Forwards the request to the workers through the request queue.</span>
|
|
<span class="sd"> """</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_start_dispatch_threads</span><span class="p">()</span>
|
|
|
|
<span class="n">request</span><span class="o">.</span><span class="n">set_id</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_get_next_client_id</span><span class="p">())</span>
|
|
|
|
<span class="n">result</span> <span class="o">=</span> <span class="n">GenerationResult</span><span class="p">(</span>
|
|
<span class="n">request</span><span class="p">,</span> <span class="n">background_error_handler</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_handle_background_error</span><span class="p">)</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_results</span><span class="p">[</span><span class="n">request</span><span class="o">.</span><span class="n">id</span><span class="p">]</span> <span class="o">=</span> <span class="n">result</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">request_queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">request</span><span class="p">)</span>
|
|
|
|
<span class="n">error</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">request_error_queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">error</span><span class="p">,</span> <span class="ne">Exception</span><span class="p">):</span>
|
|
<span class="k">raise</span> <span class="n">error</span>
|
|
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">_handle_background_error</span><span class="p">()</span>
|
|
|
|
<span class="k">return</span> <span class="n">result</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__del__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__enter__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="bp">self</span>
|
|
|
|
<span class="k">def</span> <span class="fm">__exit__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">exc_type</span><span class="p">,</span> <span class="n">exc_value</span><span class="p">,</span> <span class="n">traceback</span><span class="p">):</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
|
|
<span class="k">return</span> <span class="kc">False</span> <span class="c1"># propagate the exception</span>
|
|
|
|
|
|
<span class="k">if</span> <span class="n">enable_llm_debug</span><span class="p">():</span>
|
|
<span class="n">print_colored</span><span class="p">(</span><span class="s2">"LLM debug mode enabled.</span><span class="se">\n</span><span class="s2">"</span><span class="p">,</span> <span class="s2">"yellow"</span><span class="p">)</span>
|
|
<span class="c1"># This will dump all the alive threads when the process is interrupted by SIGINT.</span>
|
|
<span class="n">faulthandler</span><span class="o">.</span><span class="n">register</span><span class="p">(</span><span class="n">signal</span><span class="o">.</span><span class="n">SIGINT</span><span class="p">,</span> <span class="n">all_threads</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
|
|
|
|
|
|
<span class="k">def</span> <span class="nf">has_event_loop</span><span class="p">()</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
|
|
<span class="k">try</span><span class="p">:</span>
|
|
<span class="n">asyncio</span><span class="o">.</span><span class="n">get_running_loop</span><span class="p">()</span>
|
|
<span class="k">except</span> <span class="ne">RuntimeError</span><span class="p">:</span>
|
|
<span class="k">return</span> <span class="kc">False</span>
|
|
<span class="k">return</span> <span class="kc">True</span>
|
|
</pre></div>
|
|
|
|
</div>
|
|
</div>
|
|
<footer>
|
|
|
|
<hr/>
|
|
|
|
<div role="contentinfo">
|
|
<jinja2.runtime.BlockReference object at 0x7fed980d8d10>
|
|
|
|
<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> |