mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-14 06:27:45 +08:00
324 lines
30 KiB
HTML
324 lines
30 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.quantization.mode — 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=19f00094" />
|
|
|
|
|
|
<!--[if lt IE 9]>
|
|
<script src="../../../_static/js/html5shiv.min.js"></script>
|
|
<![endif]-->
|
|
|
|
<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/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">Contents:</span></p>
|
|
<ul>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../architecture.html">TensorRT-LLM Architecture</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../gpt_runtime.html">C++ GPT Runtime</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../batch_manager.html">The Batch Manager in TensorRT-LLM</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../gpt_attention.html">Multi-head, Multi-query and Group-query Attention</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../precision.html">Numerical Precision</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../installation.html">Build TensorRT-LLM</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../performance.html">Performance of TensorRT-LLM</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../2023-05-19-how-to-debug.html">How to debug</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../2023-05-17-how-to-add-a-new-model.html">How to add a new model</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../graph-rewriting.html">Graph Rewriting Module</a></li>
|
|
<li class="toctree-l1"><a class="reference internal" href="../../../memory.html">Memory Usage of TensorRT-LLM</a></li>
|
|
</ul>
|
|
<p class="caption" role="heading"><span class="caption-text">Python 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/runtime.html">Runtime</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>
|
|
</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.quantization.mode</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.quantization.mode</h1><div class="highlight"><pre>
|
|
<span></span><span class="c1"># SPDX-FileCopyrightText: Copyright (c) 2022-2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.</span>
|
|
<span class="c1"># SPDX-License-Identifier: Apache-2.0</span>
|
|
<span class="c1">#</span>
|
|
<span class="c1"># Licensed under the Apache License, Version 2.0 (the "License");</span>
|
|
<span class="c1"># you may not use this file except in compliance with the License.</span>
|
|
<span class="c1"># You may obtain a copy of the License at</span>
|
|
<span class="c1">#</span>
|
|
<span class="c1"># http://www.apache.org/licenses/LICENSE-2.0</span>
|
|
<span class="c1">#</span>
|
|
<span class="c1"># Unless required by applicable law or agreed to in writing, software</span>
|
|
<span class="c1"># distributed under the License is distributed on an "AS IS" BASIS,</span>
|
|
<span class="c1"># WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.</span>
|
|
<span class="c1"># See the License for the specific language governing permissions and</span>
|
|
<span class="c1"># limitations under the License.</span>
|
|
<span class="kn">from</span> <span class="nn">enum</span> <span class="kn">import</span> <span class="n">IntFlag</span><span class="p">,</span> <span class="n">auto</span>
|
|
|
|
|
|
<div class="viewcode-block" id="QuantMode">
|
|
<a class="viewcode-back" href="../../../python-api/tensorrt_llm.quantization.html#tensorrt_llm.quantization.QuantMode">[docs]</a>
|
|
<span class="k">class</span> <span class="nc">QuantMode</span><span class="p">(</span><span class="n">IntFlag</span><span class="p">):</span>
|
|
<span class="c1"># [WARNING] KEEP BELOW DEFINITION IN SYNC WITH cpp/tensorrt_llm/common/quantization.h</span>
|
|
|
|
<span class="c1"># The weights are quantized to 4 bits.</span>
|
|
<span class="n">INT4_WEIGHTS</span> <span class="o">=</span> <span class="n">auto</span><span class="p">()</span>
|
|
<span class="c1"># The weights are quantized to 8 bits.</span>
|
|
<span class="n">INT8_WEIGHTS</span> <span class="o">=</span> <span class="n">auto</span><span class="p">()</span>
|
|
<span class="c1"># The activations are quantized.</span>
|
|
<span class="n">ACTIVATIONS</span> <span class="o">=</span> <span class="n">auto</span><span class="p">()</span>
|
|
<span class="c1"># The method uses one scaling factor per channel. It's pre-computed (static) from the weights.</span>
|
|
<span class="n">PER_CHANNEL</span> <span class="o">=</span> <span class="n">auto</span><span class="p">()</span>
|
|
<span class="c1"># The method uses one scaling factor per token. It's computed on-the-fly.</span>
|
|
<span class="n">PER_TOKEN</span> <span class="o">=</span> <span class="n">auto</span><span class="p">()</span>
|
|
<span class="c1"># The method uses one scaling factor per group. It's pre-computed (static) from the weights.</span>
|
|
<span class="n">PER_GROUP</span> <span class="o">=</span> <span class="n">auto</span><span class="p">()</span>
|
|
<span class="c1"># The KV cache is quantized in INT8.</span>
|
|
<span class="n">INT8_KV_CACHE</span> <span class="o">=</span> <span class="n">auto</span><span class="p">()</span>
|
|
<span class="c1"># The KV cache is quantized in FP8.</span>
|
|
<span class="n">FP8_KV_CACHE</span> <span class="o">=</span> <span class="n">auto</span><span class="p">()</span>
|
|
<span class="c1"># FP8 QDQ</span>
|
|
<span class="n">FP8_QDQ</span> <span class="o">=</span> <span class="n">auto</span><span class="p">()</span>
|
|
|
|
<span class="c1"># The smallest power-of-two that is not used by a flag. Do not call auto() after that line.</span>
|
|
<span class="n">COUNT</span> <span class="o">=</span> <span class="n">auto</span><span class="p">()</span>
|
|
|
|
<span class="c1"># Bitmask to detect if weights, activations or both are quantized.</span>
|
|
<span class="n">WEIGHTS_AND_ACTIVATIONS</span> <span class="o">=</span> <span class="n">INT4_WEIGHTS</span> <span class="o">|</span> <span class="n">INT8_WEIGHTS</span> <span class="o">|</span> <span class="n">ACTIVATIONS</span>
|
|
<span class="c1"># The mask of all valid flags.</span>
|
|
<span class="n">VALID_FLAGS</span> <span class="o">=</span> <span class="n">COUNT</span> <span class="o">-</span> <span class="mi">1</span>
|
|
|
|
<span class="c1"># All the bits set? You can restrict the test to the bits indicated by "mask".</span>
|
|
<span class="k">def</span> <span class="nf">_all</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">bits</span><span class="p">,</span> <span class="n">mask</span><span class="o">=</span><span class="n">VALID_FLAGS</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">mask</span><span class="p">)</span> <span class="o">==</span> <span class="n">bits</span>
|
|
|
|
<span class="c1"># Is one of the bits of the mask set?</span>
|
|
<span class="k">def</span> <span class="nf">_any</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">bits</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">bits</span><span class="p">)</span> <span class="o">!=</span> <span class="mi">0</span>
|
|
|
|
<span class="k">def</span> <span class="nf">is_int8_weight_only</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">_all</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">INT8_WEIGHTS</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">WEIGHTS_AND_ACTIVATIONS</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">is_int4_weight_only</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">_all</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">INT4_WEIGHTS</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">WEIGHTS_AND_ACTIVATIONS</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">is_weight_only</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">is_int4_weight_only</span><span class="p">()</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">is_int8_weight_only</span><span class="p">()</span>
|
|
|
|
<span class="k">def</span> <span class="nf">is_int4_weight_only_per_group</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">is_int4_weight_only</span><span class="p">()</span> <span class="ow">and</span> <span class="bp">self</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">PER_GROUP</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">has_act_and_weight_quant</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">_all</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">INT8_WEIGHTS</span> <span class="o">|</span> <span class="bp">self</span><span class="o">.</span><span class="n">ACTIVATIONS</span><span class="p">,</span>
|
|
<span class="bp">self</span><span class="o">.</span><span class="n">WEIGHTS_AND_ACTIVATIONS</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">has_act_or_weight_quant</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">_any</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">INT4_WEIGHTS</span> <span class="o">|</span> <span class="bp">self</span><span class="o">.</span><span class="n">INT8_WEIGHTS</span>
|
|
<span class="o">|</span> <span class="bp">self</span><span class="o">.</span><span class="n">ACTIVATIONS</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">has_per_token_dynamic_scaling</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">_any</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">PER_TOKEN</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">has_act_static_scaling</span><span class="p">(</span><span class="bp">self</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">has_per_token_dynamic_scaling</span><span class="p">()</span>
|
|
|
|
<span class="k">def</span> <span class="nf">has_per_channel_scaling</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">_any</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">PER_CHANNEL</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">has_per_group_scaling</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">_any</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">PER_GROUP</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">has_int8_kv_cache</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">_any</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">INT8_KV_CACHE</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">has_fp8_kv_cache</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">_any</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">FP8_KV_CACHE</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">has_kv_cache_quant</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">has_int8_kv_cache</span><span class="p">()</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">has_fp8_kv_cache</span><span class="p">()</span>
|
|
|
|
<span class="k">def</span> <span class="nf">has_fp8_qdq</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">_any</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">FP8_QDQ</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">has_any_quant</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">_any</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">INT8_WEIGHTS</span> <span class="o">|</span> <span class="bp">self</span><span class="o">.</span><span class="n">ACTIVATIONS</span>
|
|
<span class="o">|</span> <span class="bp">self</span><span class="o">.</span><span class="n">INT8_KV_CACHE</span> <span class="o">|</span> <span class="bp">self</span><span class="o">.</span><span class="n">FP8_KV_CACHE</span>
|
|
<span class="o">|</span> <span class="bp">self</span><span class="o">.</span><span class="n">FP8_QDQ</span><span class="p">)</span>
|
|
|
|
<span class="k">def</span> <span class="nf">set_int8_kv_cache</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="bp">self</span><span class="o">.</span><span class="n">INT8_KV_CACHE</span>
|
|
|
|
<span class="k">def</span> <span class="nf">set_fp8_kv_cache</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="bp">self</span><span class="o">.</span><span class="n">FP8_KV_CACHE</span>
|
|
|
|
<span class="k">def</span> <span class="nf">set_fp8_qdq</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="bp">self</span><span class="o">.</span><span class="n">FP8_QDQ</span>
|
|
|
|
<span class="nd">@staticmethod</span>
|
|
<span class="k">def</span> <span class="nf">from_description</span><span class="p">(</span><span class="n">quantize_weights</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">quantize_activations</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">per_token</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">per_channel</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">per_group</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">use_int4_weights</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">use_int8_kv_cache</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">use_fp8_kv_cache</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">use_fp8_qdq</span><span class="o">=</span><span class="kc">False</span><span class="p">):</span>
|
|
|
|
<span class="k">def</span> <span class="nf">raise_error</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">"Unsupported combination of QuantMode args: "</span>
|
|
<span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="n">quantize_weights</span><span class="si">=}</span><span class="s2">, "</span>
|
|
<span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="n">quantize_activations</span><span class="si">=}</span><span class="s2">, "</span>
|
|
<span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="n">per_token</span><span class="si">=}</span><span class="s2">, "</span>
|
|
<span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="n">per_channel</span><span class="si">=}</span><span class="s2">, "</span>
|
|
<span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="n">per_group</span><span class="si">=}</span><span class="s2">, "</span>
|
|
<span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="n">use_int4_weights</span><span class="si">=}</span><span class="s2">"</span>
|
|
<span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="n">use_int8_kv_cache</span><span class="si">=}</span><span class="s2">"</span>
|
|
<span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="n">use_fp8_kv_cache</span><span class="si">=}</span><span class="s2">"</span>
|
|
<span class="sa">f</span><span class="s2">"</span><span class="si">{</span><span class="n">use_fp8_qdq</span><span class="si">=}</span><span class="s2">"</span><span class="p">)</span>
|
|
|
|
<span class="c1"># We must quantize weights when we quantize activations.</span>
|
|
<span class="k">if</span> <span class="n">quantize_activations</span> <span class="ow">and</span> <span class="ow">not</span> <span class="n">quantize_weights</span><span class="p">:</span>
|
|
<span class="n">raise_error</span><span class="p">()</span>
|
|
|
|
<span class="c1"># If we set per_token or per_channel, we must quantize both weights and activations.</span>
|
|
<span class="k">if</span> <span class="p">(</span><span class="n">per_token</span> <span class="ow">or</span> <span class="n">per_channel</span><span class="p">)</span> <span class="ow">and</span> <span class="ow">not</span> <span class="p">(</span><span class="n">quantize_weights</span>
|
|
<span class="ow">and</span> <span class="n">quantize_activations</span><span class="p">):</span>
|
|
<span class="n">raise_error</span><span class="p">()</span>
|
|
|
|
<span class="n">mode</span> <span class="o">=</span> <span class="n">QuantMode</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span>
|
|
|
|
<span class="c1"># Do we quantize the weights - if so, do we use INT4 or INT8?</span>
|
|
<span class="k">if</span> <span class="n">quantize_weights</span> <span class="ow">and</span> <span class="n">use_int4_weights</span><span class="p">:</span>
|
|
<span class="n">mode</span> <span class="o">=</span> <span class="n">mode</span> <span class="o">|</span> <span class="n">QuantMode</span><span class="o">.</span><span class="n">INT4_WEIGHTS</span>
|
|
<span class="k">elif</span> <span class="n">quantize_weights</span><span class="p">:</span>
|
|
<span class="n">mode</span> <span class="o">=</span> <span class="n">mode</span> <span class="o">|</span> <span class="n">QuantMode</span><span class="o">.</span><span class="n">INT8_WEIGHTS</span>
|
|
|
|
<span class="c1"># Do we quantize the activations?</span>
|
|
<span class="k">if</span> <span class="n">quantize_activations</span><span class="p">:</span>
|
|
<span class="n">mode</span> <span class="o">=</span> <span class="n">mode</span> <span class="o">|</span> <span class="n">QuantMode</span><span class="o">.</span><span class="n">ACTIVATIONS</span>
|
|
|
|
<span class="c1"># Per-channel/per-token/per-group additional flags.</span>
|
|
<span class="k">if</span> <span class="n">per_channel</span><span class="p">:</span>
|
|
<span class="n">mode</span> <span class="o">=</span> <span class="n">mode</span> <span class="o">|</span> <span class="n">QuantMode</span><span class="o">.</span><span class="n">PER_CHANNEL</span>
|
|
<span class="k">if</span> <span class="n">per_token</span><span class="p">:</span>
|
|
<span class="n">mode</span> <span class="o">=</span> <span class="n">mode</span> <span class="o">|</span> <span class="n">QuantMode</span><span class="o">.</span><span class="n">PER_TOKEN</span>
|
|
<span class="k">if</span> <span class="n">per_group</span><span class="p">:</span>
|
|
<span class="n">mode</span> <span class="o">=</span> <span class="n">mode</span> <span class="o">|</span> <span class="n">QuantMode</span><span class="o">.</span><span class="n">PER_GROUP</span>
|
|
|
|
<span class="c1"># Int8 KV cache</span>
|
|
<span class="k">if</span> <span class="n">use_int8_kv_cache</span><span class="p">:</span>
|
|
<span class="n">mode</span> <span class="o">=</span> <span class="n">mode</span> <span class="o">|</span> <span class="n">QuantMode</span><span class="o">.</span><span class="n">INT8_KV_CACHE</span>
|
|
|
|
<span class="c1"># FP8 KV cache</span>
|
|
<span class="k">if</span> <span class="n">use_fp8_kv_cache</span><span class="p">:</span>
|
|
<span class="n">mode</span> <span class="o">=</span> <span class="n">mode</span> <span class="o">|</span> <span class="n">QuantMode</span><span class="o">.</span><span class="n">FP8_KV_CACHE</span>
|
|
|
|
<span class="k">if</span> <span class="n">use_fp8_qdq</span><span class="p">:</span>
|
|
<span class="n">mode</span> <span class="o">=</span> <span class="n">mode</span> <span class="o">|</span> <span class="n">QuantMode</span><span class="o">.</span><span class="n">FP8_QDQ</span>
|
|
|
|
<span class="k">return</span> <span class="n">mode</span>
|
|
|
|
<span class="nd">@staticmethod</span>
|
|
<span class="k">def</span> <span class="nf">use_smooth_quant</span><span class="p">(</span><span class="n">per_token</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">per_channel</span><span class="o">=</span><span class="kc">False</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="n">QuantMode</span><span class="o">.</span><span class="n">from_description</span><span class="p">(</span><span class="kc">True</span><span class="p">,</span> <span class="kc">True</span><span class="p">,</span> <span class="n">per_token</span><span class="p">,</span> <span class="n">per_channel</span><span class="p">)</span>
|
|
|
|
<span class="nd">@staticmethod</span>
|
|
<span class="k">def</span> <span class="nf">use_weight_only</span><span class="p">(</span><span class="n">use_int4_weights</span><span class="o">=</span><span class="kc">False</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="n">QuantMode</span><span class="o">.</span><span class="n">from_description</span><span class="p">(</span><span class="n">quantize_weights</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
|
<span class="n">quantize_activations</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">per_token</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">per_channel</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">per_group</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">use_int4_weights</span><span class="o">=</span><span class="n">use_int4_weights</span><span class="p">)</span></div>
|
|
|
|
</pre></div>
|
|
|
|
</div>
|
|
</div>
|
|
<footer>
|
|
|
|
<hr/>
|
|
|
|
<div role="contentinfo">
|
|
<p>© Copyright 2023, NVidia.</p>
|
|
</div>
|
|
|
|
Built with <a href="https://www.sphinx-doc.org/">Sphinx</a> using a
|
|
<a href="https://github.com/readthedocs/sphinx_rtd_theme">theme</a>
|
|
provided by <a href="https://readthedocs.org">Read the Docs</a>.
|
|
|
|
|
|
</footer>
|
|
</div>
|
|
</div>
|
|
</section>
|
|
</div>
|
|
<script>
|
|
jQuery(function () {
|
|
SphinxRtdTheme.Navigation.enable(true);
|
|
});
|
|
</script>
|
|
|
|
</body>
|
|
</html> |