mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-14 06:27:45 +08:00
614 lines
90 KiB
HTML
614 lines
90 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.models.quantized.quant — 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.models.quantized.quant</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.models.quantized.quant</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">typing</span> <span class="kn">import</span> <span class="n">Any</span><span class="p">,</span> <span class="n">Union</span>
|
|
|
|
<span class="kn">import</span> <span class="nn">numpy</span> <span class="k">as</span> <span class="nn">np</span>
|
|
|
|
<span class="kn">from</span> <span class="nn">...layers</span> <span class="kn">import</span> <span class="n">ColumnLinear</span><span class="p">,</span> <span class="n">RowLinear</span>
|
|
<span class="kn">from</span> <span class="nn">...models</span> <span class="kn">import</span> <span class="p">(</span><span class="n">BaichuanForCausalLM</span><span class="p">,</span> <span class="n">BloomForCausalLM</span><span class="p">,</span> <span class="n">FalconForCausalLM</span><span class="p">,</span>
|
|
<span class="n">GPTJForCausalLM</span><span class="p">,</span> <span class="n">GPTLMHeadModel</span><span class="p">,</span> <span class="n">LLaMAForCausalLM</span><span class="p">,</span>
|
|
<span class="n">QWenForCausalLM</span><span class="p">)</span>
|
|
<span class="kn">from</span> <span class="nn">...module</span> <span class="kn">import</span> <span class="n">Module</span>
|
|
<span class="kn">from</span> <span class="nn">...quantization</span> <span class="kn">import</span> <span class="n">QuantMode</span>
|
|
<span class="kn">from</span> <span class="nn">...quantization.layers</span> <span class="kn">import</span> <span class="n">FP8Linear</span><span class="p">,</span> <span class="n">FP8RowLinear</span>
|
|
|
|
<span class="c1"># isort: off</span>
|
|
<span class="kn">from</span> <span class="nn">...quantization.layers</span> <span class="kn">import</span> <span class="p">(</span>
|
|
<span class="n">SmoothQuantAttention</span><span class="p">,</span> <span class="n">SmoothQuantGatedMLP</span><span class="p">,</span> <span class="n">SmoothQuantLayerNorm</span><span class="p">,</span>
|
|
<span class="n">SmoothQuantMLP</span><span class="p">,</span> <span class="n">SmoothQuantRmsNorm</span><span class="p">,</span> <span class="n">WeightOnlyGroupwiseQuantColumnLinear</span><span class="p">,</span>
|
|
<span class="n">WeightOnlyGroupwiseQuantRowLinear</span><span class="p">,</span> <span class="n">WeightOnlyQuantColumnLinear</span><span class="p">,</span>
|
|
<span class="n">WeightOnlyQuantRowLinear</span><span class="p">)</span>
|
|
<span class="c1"># isort: on</span>
|
|
|
|
|
|
<span class="k">def</span> <span class="nf">_smooth_quantize_gpt</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">):</span>
|
|
<span class="k">assert</span> <span class="n">quant_mode</span><span class="o">.</span><span class="n">has_act_and_weight_quant</span><span class="p">()</span>
|
|
<span class="k">for</span> <span class="n">layer</span> <span class="ow">in</span> <span class="n">model</span><span class="o">.</span><span class="n">layers</span><span class="p">:</span>
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span>
|
|
<span class="s2">"input_layernorm"</span><span class="p">),</span> <span class="s2">"The layer has no input_layernorm"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">input_layernorm</span> <span class="o">=</span> <span class="n">SmoothQuantLayerNorm</span><span class="p">(</span>
|
|
<span class="n">normalized_shape</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="s2">"attention"</span><span class="p">),</span> <span class="s2">"The layer has no attention"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">attention</span> <span class="o">=</span> <span class="n">SmoothQuantAttention</span><span class="p">(</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">num_attention_heads</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">num_attention_heads</span><span class="p">,</span>
|
|
<span class="n">max_position_embeddings</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">max_position_embeddings</span><span class="p">,</span>
|
|
<span class="n">num_layers</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">num_layers</span><span class="p">,</span>
|
|
<span class="n">apply_query_key_layer_scaling</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">apply_query_key_layer_scaling</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">attention_mask_type</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">attention_mask_type</span><span class="p">,</span>
|
|
<span class="n">position_embedding_type</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">position_embedding_type</span><span class="p">,</span>
|
|
<span class="n">tp_group</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
|
|
<span class="n">tp_size</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="s2">"mlp"</span><span class="p">),</span> <span class="s2">"The layer has no mlp"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">mlp</span> <span class="o">=</span> <span class="n">SmoothQuantMLP</span><span class="p">(</span><span class="n">hidden_size</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">ffn_hidden_size</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span> <span class="o">*</span> <span class="mi">4</span><span class="p">,</span>
|
|
<span class="n">hidden_act</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_act</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">tp_group</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
|
|
<span class="n">tp_size</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span>
|
|
<span class="s2">"post_layernorm"</span><span class="p">),</span> <span class="s2">"The layer has no post_layernorm"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">post_layernorm</span> <span class="o">=</span> <span class="n">SmoothQuantLayerNorm</span><span class="p">(</span>
|
|
<span class="n">normalized_shape</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">)</span>
|
|
|
|
<span class="k">return</span> <span class="n">model</span>
|
|
|
|
|
|
<span class="k">def</span> <span class="nf">_smooth_quantize_llama</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">):</span>
|
|
<span class="k">assert</span> <span class="n">quant_mode</span><span class="o">.</span><span class="n">has_act_and_weight_quant</span><span class="p">()</span>
|
|
<span class="k">for</span> <span class="n">layer</span> <span class="ow">in</span> <span class="n">model</span><span class="o">.</span><span class="n">layers</span><span class="p">:</span>
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span>
|
|
<span class="s2">"input_layernorm"</span><span class="p">),</span> <span class="s2">"The layer has no input_layernorm"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">input_layernorm</span> <span class="o">=</span> <span class="n">SmoothQuantRmsNorm</span><span class="p">(</span>
|
|
<span class="n">normalized_shape</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="s2">"attention"</span><span class="p">),</span> <span class="s2">"The layer has no attention"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">attention</span> <span class="o">=</span> <span class="n">SmoothQuantAttention</span><span class="p">(</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">num_attention_heads</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">num_attention_heads</span><span class="p">,</span>
|
|
<span class="n">num_kv_heads</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">num_kv_heads</span><span class="p">,</span>
|
|
<span class="n">max_position_embeddings</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">max_position_embeddings</span><span class="p">,</span>
|
|
<span class="n">num_layers</span><span class="o">=</span><span class="n">model</span><span class="o">.</span><span class="n">num_layers</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">attention_mask_type</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">attention_mask_type</span><span class="p">,</span>
|
|
<span class="n">position_embedding_type</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">position_embedding_type</span><span class="p">,</span>
|
|
<span class="n">tp_group</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
|
|
<span class="n">tp_size</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">,</span>
|
|
<span class="n">bias</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
|
|
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="s2">"mlp"</span><span class="p">),</span> <span class="s2">"The layer has no mlp"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">mlp</span> <span class="o">=</span> <span class="n">SmoothQuantGatedMLP</span><span class="p">(</span><span class="n">hidden_size</span><span class="o">=</span><span class="n">model</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">ffn_hidden_size</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">mlp_hidden_size</span><span class="p">,</span>
|
|
<span class="n">hidden_act</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_act</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">tp_group</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
|
|
<span class="n">tp_size</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">,</span>
|
|
<span class="n">bias</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span>
|
|
<span class="n">layer</span><span class="p">,</span>
|
|
<span class="s2">"post_layernorm"</span><span class="p">),</span> <span class="s2">"The layer has no post_rmspost_layernormnorm"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">post_layernorm</span> <span class="o">=</span> <span class="n">SmoothQuantRmsNorm</span><span class="p">(</span>
|
|
<span class="n">normalized_shape</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">)</span>
|
|
|
|
<span class="k">return</span> <span class="n">model</span>
|
|
|
|
|
|
<span class="k">def</span> <span class="nf">_smooth_quantize_bloom</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">):</span>
|
|
<span class="k">assert</span> <span class="n">quant_mode</span><span class="o">.</span><span class="n">has_act_and_weight_quant</span><span class="p">()</span>
|
|
<span class="k">for</span> <span class="n">layer</span> <span class="ow">in</span> <span class="n">model</span><span class="o">.</span><span class="n">layers</span><span class="p">:</span>
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span>
|
|
<span class="s2">"input_layernorm"</span><span class="p">),</span> <span class="s2">"The layer has no input_layernorm"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">input_layernorm</span> <span class="o">=</span> <span class="n">SmoothQuantLayerNorm</span><span class="p">(</span>
|
|
<span class="n">normalized_shape</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="s2">"attention"</span><span class="p">),</span> <span class="s2">"The layer has no attention"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">attention</span> <span class="o">=</span> <span class="n">SmoothQuantAttention</span><span class="p">(</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">num_attention_heads</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">num_attention_heads</span><span class="p">,</span>
|
|
<span class="n">max_position_embeddings</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">max_position_embeddings</span><span class="p">,</span>
|
|
<span class="n">num_layers</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">num_layers</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">attention_mask_type</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">attention_mask_type</span><span class="p">,</span>
|
|
<span class="n">position_embedding_type</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">position_embedding_type</span><span class="p">,</span>
|
|
<span class="n">tp_group</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
|
|
<span class="n">tp_size</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">tp_rank</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_rank</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">)</span>
|
|
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="s2">"mlp"</span><span class="p">),</span> <span class="s2">"The layer has no mlp"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">mlp</span> <span class="o">=</span> <span class="n">SmoothQuantMLP</span><span class="p">(</span><span class="n">hidden_size</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">ffn_hidden_size</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span> <span class="o">*</span> <span class="mi">4</span><span class="p">,</span>
|
|
<span class="n">hidden_act</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_act</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">tp_group</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
|
|
<span class="n">tp_size</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span>
|
|
<span class="n">layer</span><span class="p">,</span>
|
|
<span class="s2">"post_layernorm"</span><span class="p">),</span> <span class="s2">"The layer has no post_rmspost_layernormnorm"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">post_layernorm</span> <span class="o">=</span> <span class="n">SmoothQuantLayerNorm</span><span class="p">(</span>
|
|
<span class="n">normalized_shape</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">)</span>
|
|
|
|
<span class="nb">setattr</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="s1">'quant_mode'</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="n">model</span>
|
|
|
|
|
|
<span class="k">def</span> <span class="nf">_smooth_quantize_baichuan</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">):</span>
|
|
<span class="c1"># Baichuan models' structures are similar to LLaMA so we can reuse the impl</span>
|
|
<span class="k">return</span> <span class="n">_smooth_quantize_llama</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">)</span>
|
|
|
|
|
|
<span class="k">def</span> <span class="nf">_smooth_quantize_internlm</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">):</span>
|
|
<span class="k">assert</span> <span class="n">quant_mode</span><span class="o">.</span><span class="n">has_act_and_weight_quant</span><span class="p">()</span>
|
|
<span class="k">for</span> <span class="n">layer</span> <span class="ow">in</span> <span class="n">model</span><span class="o">.</span><span class="n">layers</span><span class="p">:</span>
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span>
|
|
<span class="s2">"input_layernorm"</span><span class="p">),</span> <span class="s2">"The layer has no input_layernorm"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">input_layernorm</span> <span class="o">=</span> <span class="n">SmoothQuantRmsNorm</span><span class="p">(</span>
|
|
<span class="n">normalized_shape</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="s2">"attention"</span><span class="p">),</span> <span class="s2">"The layer has no attention"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">attention</span> <span class="o">=</span> <span class="n">SmoothQuantAttention</span><span class="p">(</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">num_attention_heads</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">num_attention_heads</span><span class="p">,</span>
|
|
<span class="n">num_kv_heads</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">num_kv_heads</span><span class="p">,</span>
|
|
<span class="n">max_position_embeddings</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">max_position_embeddings</span><span class="p">,</span>
|
|
<span class="n">num_layers</span><span class="o">=</span><span class="n">model</span><span class="o">.</span><span class="n">num_layers</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">attention_mask_type</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">attention_mask_type</span><span class="p">,</span>
|
|
<span class="n">position_embedding_type</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">position_embedding_type</span><span class="p">,</span>
|
|
<span class="n">tp_group</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
|
|
<span class="n">tp_size</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">,</span>
|
|
<span class="n">bias</span><span class="o">=</span><span class="n">model</span><span class="o">.</span><span class="n">attn_bias</span><span class="p">)</span>
|
|
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="s2">"mlp"</span><span class="p">),</span> <span class="s2">"The layer has no mlp"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">mlp</span> <span class="o">=</span> <span class="n">SmoothQuantGatedMLP</span><span class="p">(</span><span class="n">hidden_size</span><span class="o">=</span><span class="n">model</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">ffn_hidden_size</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">mlp_hidden_size</span><span class="p">,</span>
|
|
<span class="n">hidden_act</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_act</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">tp_group</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
|
|
<span class="n">tp_size</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">,</span>
|
|
<span class="n">bias</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span>
|
|
<span class="n">layer</span><span class="p">,</span>
|
|
<span class="s2">"post_layernorm"</span><span class="p">),</span> <span class="s2">"The layer has no post_rmspost_layernormnorm"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">post_layernorm</span> <span class="o">=</span> <span class="n">SmoothQuantRmsNorm</span><span class="p">(</span>
|
|
<span class="n">normalized_shape</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">)</span>
|
|
|
|
<span class="nb">setattr</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="s1">'quant_mode'</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="n">model</span>
|
|
|
|
|
|
<span class="k">def</span> <span class="nf">_smooth_quantize_qwen</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">):</span>
|
|
<span class="k">assert</span> <span class="n">quant_mode</span><span class="o">.</span><span class="n">has_act_and_weight_quant</span><span class="p">()</span>
|
|
<span class="k">for</span> <span class="n">layer</span> <span class="ow">in</span> <span class="n">model</span><span class="o">.</span><span class="n">layers</span><span class="p">:</span>
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="s2">"ln_1"</span><span class="p">),</span> <span class="s2">"The layer has no ln_1"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">ln_1</span> <span class="o">=</span> <span class="n">SmoothQuantRmsNorm</span><span class="p">(</span><span class="n">normalized_shape</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="s2">"attention"</span><span class="p">),</span> <span class="s2">"The layer has no attention"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">attention</span> <span class="o">=</span> <span class="n">SmoothQuantAttention</span><span class="p">(</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">num_attention_heads</span><span class="p">,</span>
|
|
<span class="n">max_position_embeddings</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">max_position_embeddings</span><span class="p">,</span>
|
|
<span class="n">num_layers</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">num_layers</span><span class="p">,</span>
|
|
<span class="n">apply_query_key_layer_scaling</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">apply_query_key_layer_scaling</span><span class="p">,</span>
|
|
<span class="n">attention_mask_type</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">attention_mask_type</span><span class="p">,</span>
|
|
<span class="n">bias</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">bias</span><span class="p">,</span>
|
|
<span class="n">qkv_bias_only</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">position_embedding_type</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">position_embedding_type</span><span class="p">,</span>
|
|
<span class="n">tp_group</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
|
|
<span class="n">tp_size</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="s2">"mlp"</span><span class="p">),</span> <span class="s2">"The layer has no mlp"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">mlp</span> <span class="o">=</span> <span class="n">SmoothQuantGatedMLP</span><span class="p">(</span><span class="n">hidden_size</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">ffn_hidden_size</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">mlp_hidden_size</span> <span class="o">//</span>
|
|
<span class="mi">2</span><span class="p">,</span>
|
|
<span class="n">hidden_act</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_act</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">bias</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">bias</span><span class="p">,</span>
|
|
<span class="n">tp_group</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
|
|
<span class="n">tp_size</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="s2">"ln_2"</span><span class="p">),</span> <span class="s2">"The layer has no ln_2"</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">ln_2</span> <span class="o">=</span> <span class="n">SmoothQuantRmsNorm</span><span class="p">(</span><span class="n">normalized_shape</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">layer</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">)</span>
|
|
|
|
<span class="nb">setattr</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="s1">'quant_mode'</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="n">model</span>
|
|
|
|
|
|
<span class="k">def</span> <span class="nf">_smooth_quantize</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">):</span>
|
|
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">GPTLMHeadModel</span><span class="p">)</span> <span class="ow">or</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">LLaMAForCausalLM</span><span class="p">)</span> \
|
|
<span class="ow">or</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">BloomForCausalLM</span><span class="p">)</span> <span class="ow">or</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">BaichuanForCausalLM</span><span class="p">)</span> \
|
|
<span class="ow">or</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">QWenForCausalLM</span><span class="p">),</span> \
|
|
<span class="s2">"Only GPTLMHeadModel, LLaMAForCausalLM BloomForCausalLM and BaichuanForCausalLM are well tested now"</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">GPTLMHeadModel</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="n">_smooth_quantize_gpt</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">LLaMAForCausalLM</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="n">_smooth_quantize_llama</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">BloomForCausalLM</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="n">_smooth_quantize_bloom</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">BaichuanForCausalLM</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="n">_smooth_quantize_baichuan</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">QWenForCausalLM</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="n">_smooth_quantize_qwen</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="k">assert</span> <span class="kc">False</span><span class="p">,</span> <span class="sa">f</span><span class="s2">"Model </span><span class="si">{</span><span class="nb">type</span><span class="p">(</span><span class="n">model</span><span class="p">)</span><span class="o">.</span><span class="vm">__name__</span><span class="si">}</span><span class="s2"> is not supported by SmoothQuant yet"</span>
|
|
|
|
|
|
<span class="k">def</span> <span class="nf">_weight_only_quantize</span><span class="p">(</span><span class="n">model</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="p">,</span>
|
|
<span class="n">exclude_modules</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">current_key_name</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
|
|
<span class="k">assert</span> <span class="n">quant_mode</span><span class="o">.</span><span class="n">is_weight_only</span><span class="p">()</span>
|
|
|
|
<span class="n">exclude_modules</span> <span class="o">=</span> <span class="p">[</span><span class="s1">'lm_head'</span>
|
|
<span class="p">]</span> <span class="k">if</span> <span class="n">exclude_modules</span> <span class="ow">is</span> <span class="kc">None</span> <span class="k">else</span> <span class="n">exclude_modules</span>
|
|
|
|
<span class="k">for</span> <span class="n">name</span><span class="p">,</span> <span class="n">module</span> <span class="ow">in</span> <span class="n">model</span><span class="o">.</span><span class="n">named_children</span><span class="p">():</span>
|
|
<span class="k">if</span> <span class="n">current_key_name</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="n">current_key_name</span> <span class="o">=</span> <span class="p">[]</span>
|
|
<span class="n">current_key_name</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">name</span><span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="nb">len</span><span class="p">(</span><span class="nb">list</span><span class="p">(</span><span class="n">module</span><span class="o">.</span><span class="n">children</span><span class="p">()))</span> <span class="o">></span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="n">_weight_only_quantize</span><span class="p">(</span><span class="n">module</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">,</span> <span class="n">exclude_modules</span><span class="p">,</span>
|
|
<span class="n">current_key_name</span><span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">module</span><span class="p">,</span> <span class="n">ColumnLinear</span><span class="p">)</span> <span class="ow">and</span> <span class="n">name</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">exclude_modules</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="ow">not</span> <span class="nb">any</span><span class="p">(</span><span class="n">key</span> <span class="ow">in</span> <span class="s1">'.'</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">current_key_name</span><span class="p">)</span>
|
|
<span class="k">for</span> <span class="n">key</span> <span class="ow">in</span> <span class="n">exclude_modules</span><span class="p">):</span>
|
|
<span class="n">model</span><span class="o">.</span><span class="n">_modules</span><span class="p">[</span><span class="n">name</span><span class="p">]</span> <span class="o">=</span> <span class="n">WeightOnlyQuantColumnLinear</span><span class="p">(</span>
|
|
<span class="n">in_features</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">in_features</span><span class="p">,</span>
|
|
<span class="n">out_features</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">out_features</span> <span class="o">*</span> <span class="n">module</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">bias</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">bias</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">tp_group</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
|
|
<span class="n">tp_size</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">gather_output</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">gather_output</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">module</span><span class="p">,</span> <span class="n">RowLinear</span><span class="p">)</span> <span class="ow">and</span> <span class="n">name</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">exclude_modules</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="ow">not</span> <span class="nb">any</span><span class="p">(</span><span class="n">key</span> <span class="ow">in</span> <span class="s1">'.'</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">current_key_name</span><span class="p">)</span>
|
|
<span class="k">for</span> <span class="n">key</span> <span class="ow">in</span> <span class="n">exclude_modules</span><span class="p">):</span>
|
|
<span class="n">model</span><span class="o">.</span><span class="n">_modules</span><span class="p">[</span><span class="n">name</span><span class="p">]</span> <span class="o">=</span> <span class="n">WeightOnlyQuantRowLinear</span><span class="p">(</span>
|
|
<span class="n">in_features</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">in_features</span> <span class="o">*</span> <span class="n">module</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">out_features</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">out_features</span><span class="p">,</span>
|
|
<span class="n">bias</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">bias</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">tp_group</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
|
|
<span class="n">tp_size</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="o">=</span><span class="n">quant_mode</span><span class="p">)</span>
|
|
|
|
<span class="n">current_key_name</span><span class="o">.</span><span class="n">pop</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="n">model</span>
|
|
|
|
|
|
<span class="k">def</span> <span class="nf">_weight_only_groupwise_quantize</span><span class="p">(</span><span class="n">model</span><span class="p">,</span>
|
|
<span class="n">quant_mode</span><span class="p">,</span>
|
|
<span class="n">group_size</span><span class="o">=</span><span class="mi">128</span><span class="p">,</span>
|
|
<span class="n">pre_quant_scale</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">zero</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
|
<span class="n">exclude_modules</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">current_key_name</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
|
|
<span class="n">exclude_modules</span> <span class="o">=</span> <span class="p">[</span><span class="s1">'lm_head'</span>
|
|
<span class="p">]</span> <span class="k">if</span> <span class="n">exclude_modules</span> <span class="ow">is</span> <span class="kc">None</span> <span class="k">else</span> <span class="n">exclude_modules</span>
|
|
|
|
<span class="k">for</span> <span class="n">name</span><span class="p">,</span> <span class="n">module</span> <span class="ow">in</span> <span class="n">model</span><span class="o">.</span><span class="n">named_children</span><span class="p">():</span>
|
|
<span class="k">if</span> <span class="n">current_key_name</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="n">current_key_name</span> <span class="o">=</span> <span class="p">[]</span>
|
|
<span class="n">current_key_name</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">name</span><span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="nb">len</span><span class="p">(</span><span class="nb">list</span><span class="p">(</span><span class="n">module</span><span class="o">.</span><span class="n">children</span><span class="p">()))</span> <span class="o">></span> <span class="mi">0</span><span class="p">:</span>
|
|
<span class="n">_weight_only_groupwise_quantize</span><span class="p">(</span><span class="n">module</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">,</span> <span class="n">group_size</span><span class="p">,</span>
|
|
<span class="n">pre_quant_scale</span><span class="p">,</span> <span class="n">zero</span><span class="p">,</span>
|
|
<span class="n">exclude_modules</span><span class="p">,</span> <span class="n">current_key_name</span><span class="p">)</span>
|
|
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">module</span><span class="p">,</span> <span class="n">ColumnLinear</span><span class="p">)</span> <span class="ow">and</span> <span class="n">name</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">exclude_modules</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="ow">not</span> <span class="nb">any</span><span class="p">(</span><span class="n">key</span> <span class="ow">in</span> <span class="s1">'.'</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">current_key_name</span><span class="p">)</span>
|
|
<span class="k">for</span> <span class="n">key</span> <span class="ow">in</span> <span class="n">exclude_modules</span><span class="p">):</span>
|
|
<span class="n">model</span><span class="o">.</span><span class="n">_modules</span><span class="p">[</span><span class="n">name</span><span class="p">]</span> <span class="o">=</span> <span class="n">WeightOnlyGroupwiseQuantColumnLinear</span><span class="p">(</span>
|
|
<span class="n">in_features</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">in_features</span><span class="p">,</span>
|
|
<span class="n">out_features</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">out_features</span> <span class="o">*</span> <span class="n">module</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">group_size</span><span class="o">=</span><span class="n">group_size</span><span class="p">,</span>
|
|
<span class="n">pre_quant_scale</span><span class="o">=</span><span class="n">pre_quant_scale</span><span class="p">,</span>
|
|
<span class="n">zero</span><span class="o">=</span><span class="n">zero</span><span class="p">,</span>
|
|
<span class="n">bias</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">bias</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">tp_group</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
|
|
<span class="n">tp_size</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">gather_output</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">gather_output</span><span class="p">)</span>
|
|
<span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">module</span><span class="p">,</span> <span class="n">RowLinear</span><span class="p">)</span> <span class="ow">and</span> <span class="n">name</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">exclude_modules</span><span class="p">:</span>
|
|
<span class="k">if</span> <span class="ow">not</span> <span class="nb">any</span><span class="p">(</span><span class="n">key</span> <span class="ow">in</span> <span class="s1">'.'</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">current_key_name</span><span class="p">)</span>
|
|
<span class="k">for</span> <span class="n">key</span> <span class="ow">in</span> <span class="n">exclude_modules</span><span class="p">):</span>
|
|
<span class="n">model</span><span class="o">.</span><span class="n">_modules</span><span class="p">[</span><span class="n">name</span><span class="p">]</span> <span class="o">=</span> <span class="n">WeightOnlyGroupwiseQuantRowLinear</span><span class="p">(</span>
|
|
<span class="n">in_features</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">in_features</span> <span class="o">*</span> <span class="n">module</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
|
|
<span class="n">out_features</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">out_features</span><span class="p">,</span>
|
|
<span class="n">group_size</span><span class="o">=</span><span class="n">group_size</span><span class="p">,</span>
|
|
<span class="n">pre_quant_scale</span><span class="o">=</span><span class="n">pre_quant_scale</span><span class="p">,</span>
|
|
<span class="n">zero</span><span class="o">=</span><span class="n">zero</span><span class="p">,</span>
|
|
<span class="n">bias</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">bias</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">,</span>
|
|
<span class="n">dtype</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
|
<span class="n">tp_group</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
|
|
<span class="n">tp_size</span><span class="o">=</span><span class="n">module</span><span class="o">.</span><span class="n">tp_size</span><span class="p">)</span>
|
|
|
|
<span class="n">current_key_name</span><span class="o">.</span><span class="n">pop</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="n">model</span>
|
|
|
|
|
|
<div class="viewcode-block" id="quantize_model">
|
|
<a class="viewcode-back" href="../../../../python-api/tensorrt_llm.models.html#tensorrt_llm.models.quantize_model">[docs]</a>
|
|
<span class="k">def</span> <span class="nf">quantize_model</span><span class="p">(</span><span class="n">model</span><span class="p">:</span> <span class="n">Module</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">:</span> <span class="n">QuantMode</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">:</span> <span class="n">Any</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="n">quant_mode</span><span class="o">.</span><span class="n">has_fp8_qdq</span><span class="p">()</span> <span class="ow">or</span> <span class="n">quant_mode</span><span class="o">.</span><span class="n">has_fp8_kv_cache</span><span class="p">():</span>
|
|
<span class="n">model</span> <span class="o">=</span> <span class="n">_fp8_quantize</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">)</span>
|
|
<span class="k">elif</span> <span class="n">quant_mode</span><span class="o">.</span><span class="n">has_act_and_weight_quant</span><span class="p">():</span>
|
|
<span class="n">model</span> <span class="o">=</span> <span class="n">_smooth_quantize</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">elif</span> <span class="n">quant_mode</span><span class="o">.</span><span class="n">is_weight_only</span><span class="p">():</span>
|
|
<span class="k">if</span> <span class="n">quant_mode</span><span class="o">.</span><span class="n">has_per_group_scaling</span><span class="p">():</span>
|
|
<span class="n">model</span> <span class="o">=</span> <span class="n">_weight_only_groupwise_quantize</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">)</span>
|
|
<span class="k">else</span><span class="p">:</span>
|
|
<span class="n">model</span> <span class="o">=</span> <span class="n">_weight_only_quantize</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">)</span>
|
|
|
|
<span class="nb">setattr</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="s2">"quant_mode"</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">)</span>
|
|
<span class="k">return</span> <span class="n">model</span></div>
|
|
|
|
|
|
|
|
<span class="k">def</span> <span class="nf">get_dummy_quant_scales</span><span class="p">(</span><span class="n">num_layers</span><span class="p">):</span>
|
|
<span class="k">return</span> <span class="p">{</span>
|
|
<span class="s1">'lm_head_act'</span><span class="p">:</span> <span class="mf">0.99</span><span class="p">,</span>
|
|
<span class="s1">'lm_head_weights'</span><span class="p">:</span> <span class="mf">0.99</span><span class="p">,</span>
|
|
<span class="s1">'fc_act'</span><span class="p">:</span> <span class="p">[</span><span class="mf">0.99</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">num_layers</span><span class="p">)],</span>
|
|
<span class="s1">'fc_weights'</span><span class="p">:</span> <span class="p">[</span><span class="mf">0.99</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">num_layers</span><span class="p">)],</span>
|
|
<span class="s1">'gate_act'</span><span class="p">:</span> <span class="p">[</span><span class="mf">0.99</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">num_layers</span><span class="p">)],</span>
|
|
<span class="s1">'gate_weights'</span><span class="p">:</span> <span class="p">[</span><span class="mf">0.99</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">num_layers</span><span class="p">)],</span>
|
|
<span class="s1">'proj_act'</span><span class="p">:</span> <span class="p">[</span><span class="mf">0.99</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">num_layers</span><span class="p">)],</span>
|
|
<span class="s1">'proj_weights'</span><span class="p">:</span> <span class="p">[</span><span class="mf">0.99</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">num_layers</span><span class="p">)],</span>
|
|
<span class="s1">'qkv_act'</span><span class="p">:</span> <span class="p">[</span><span class="mf">0.99</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">num_layers</span><span class="p">)],</span>
|
|
<span class="s1">'qkv_weights'</span><span class="p">:</span> <span class="p">[</span><span class="mf">0.99</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">num_layers</span><span class="p">)],</span>
|
|
<span class="s1">'qkv_output'</span><span class="p">:</span> <span class="p">[</span><span class="mf">5.0</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">num_layers</span><span class="p">)],</span>
|
|
<span class="s1">'dense_act'</span><span class="p">:</span> <span class="p">[</span><span class="mf">0.99</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">num_layers</span><span class="p">)],</span>
|
|
<span class="s1">'dense_weights'</span><span class="p">:</span> <span class="p">[</span><span class="mf">0.99</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">num_layers</span><span class="p">)],</span>
|
|
<span class="p">}</span>
|
|
|
|
|
|
<span class="k">def</span> <span class="nf">_quantize_layer</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="n">layer_idx</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">,</span> <span class="n">quant_scales</span><span class="p">):</span>
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="s2">"mlp"</span><span class="p">),</span> <span class="s2">"The layer has no mlp"</span>
|
|
<span class="n">fake_fp8_sf_dt</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">float32</span>
|
|
|
|
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">mlp</span><span class="o">.</span><span class="n">fc</span><span class="p">,</span> <span class="p">(</span><span class="n">FP8Linear</span><span class="p">,</span> <span class="n">FP8RowLinear</span><span class="p">))</span>
|
|
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">mlp</span><span class="o">.</span><span class="n">proj</span><span class="p">,</span> <span class="p">(</span><span class="n">FP8Linear</span><span class="p">,</span> <span class="n">FP8RowLinear</span><span class="p">))</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">mlp</span><span class="o">.</span><span class="n">fc</span><span class="o">.</span><span class="n">activation_scaling_factor</span><span class="o">.</span><span class="n">value</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span>
|
|
<span class="p">[</span><span class="n">quant_scales</span><span class="p">[</span><span class="s1">'fc_act'</span><span class="p">][</span><span class="n">layer_idx</span><span class="p">]],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">fake_fp8_sf_dt</span><span class="p">)</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">mlp</span><span class="o">.</span><span class="n">fc</span><span class="o">.</span><span class="n">weights_scaling_factor</span><span class="o">.</span><span class="n">value</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span>
|
|
<span class="p">[</span><span class="n">quant_scales</span><span class="p">[</span><span class="s1">'fc_weights'</span><span class="p">][</span><span class="n">layer_idx</span><span class="p">]],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">fake_fp8_sf_dt</span><span class="p">)</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">mlp</span><span class="o">.</span><span class="n">proj</span><span class="o">.</span><span class="n">activation_scaling_factor</span><span class="o">.</span><span class="n">value</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span>
|
|
<span class="p">[</span><span class="n">quant_scales</span><span class="p">[</span><span class="s1">'proj_act'</span><span class="p">][</span><span class="n">layer_idx</span><span class="p">]],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">fake_fp8_sf_dt</span><span class="p">)</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">mlp</span><span class="o">.</span><span class="n">proj</span><span class="o">.</span><span class="n">weights_scaling_factor</span><span class="o">.</span><span class="n">value</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span>
|
|
<span class="p">[</span><span class="n">quant_scales</span><span class="p">[</span><span class="s1">'proj_weights'</span><span class="p">][</span><span class="n">layer_idx</span><span class="p">]],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">fake_fp8_sf_dt</span><span class="p">)</span>
|
|
<span class="k">if</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">mlp</span><span class="p">,</span> <span class="s1">'gate'</span><span class="p">):</span>
|
|
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">mlp</span><span class="o">.</span><span class="n">gate</span><span class="p">,</span> <span class="p">(</span><span class="n">FP8Linear</span><span class="p">,</span> <span class="n">FP8RowLinear</span><span class="p">))</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">mlp</span><span class="o">.</span><span class="n">gate</span><span class="o">.</span><span class="n">activation_scaling_factor</span><span class="o">.</span><span class="n">value</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span>
|
|
<span class="p">[</span><span class="n">quant_scales</span><span class="p">[</span><span class="s1">'gate_act'</span><span class="p">][</span><span class="n">layer_idx</span><span class="p">]],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">fake_fp8_sf_dt</span><span class="p">)</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">mlp</span><span class="o">.</span><span class="n">gate</span><span class="o">.</span><span class="n">weights_scaling_factor</span><span class="o">.</span><span class="n">value</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span>
|
|
<span class="p">[</span><span class="n">quant_scales</span><span class="p">[</span><span class="s1">'gate_weights'</span><span class="p">][</span><span class="n">layer_idx</span><span class="p">]],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">fake_fp8_sf_dt</span><span class="p">)</span>
|
|
|
|
<span class="k">assert</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="s2">"attention"</span><span class="p">),</span> <span class="s2">"The layer has no attention"</span>
|
|
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">attention</span><span class="o">.</span><span class="n">qkv</span><span class="p">,</span> <span class="p">(</span><span class="n">FP8Linear</span><span class="p">,</span> <span class="n">FP8RowLinear</span><span class="p">))</span>
|
|
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="n">attention</span><span class="o">.</span><span class="n">dense</span><span class="p">,</span> <span class="p">(</span><span class="n">FP8Linear</span><span class="p">,</span> <span class="n">FP8RowLinear</span><span class="p">))</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">attention</span><span class="o">.</span><span class="n">qkv</span><span class="o">.</span><span class="n">activation_scaling_factor</span><span class="o">.</span><span class="n">value</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span>
|
|
<span class="p">[</span><span class="n">quant_scales</span><span class="p">[</span><span class="s1">'qkv_act'</span><span class="p">][</span><span class="n">layer_idx</span><span class="p">]],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">fake_fp8_sf_dt</span><span class="p">)</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">attention</span><span class="o">.</span><span class="n">qkv</span><span class="o">.</span><span class="n">weights_scaling_factor</span><span class="o">.</span><span class="n">value</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span>
|
|
<span class="p">[</span><span class="n">quant_scales</span><span class="p">[</span><span class="s1">'qkv_weights'</span><span class="p">][</span><span class="n">layer_idx</span><span class="p">]],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">fake_fp8_sf_dt</span><span class="p">)</span>
|
|
<span class="k">if</span> <span class="n">quant_mode</span><span class="o">.</span><span class="n">has_fp8_kv_cache</span><span class="p">():</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">attention</span><span class="o">.</span><span class="n">kv_orig_quant_scale</span><span class="o">.</span><span class="n">value</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span>
|
|
<span class="p">[</span><span class="n">quant_scales</span><span class="p">[</span><span class="s1">'qkv_output'</span><span class="p">][</span><span class="n">layer_idx</span><span class="p">]],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">fake_fp8_sf_dt</span><span class="p">)</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">attention</span><span class="o">.</span><span class="n">kv_quant_orig_scale</span><span class="o">.</span><span class="n">value</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span>
|
|
<span class="p">[</span><span class="mf">1.0</span> <span class="o">/</span> <span class="n">quant_scales</span><span class="p">[</span><span class="s1">'qkv_output'</span><span class="p">][</span><span class="n">layer_idx</span><span class="p">]],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">fake_fp8_sf_dt</span><span class="p">)</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">attention</span><span class="o">.</span><span class="n">dense</span><span class="o">.</span><span class="n">activation_scaling_factor</span><span class="o">.</span><span class="n">value</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span>
|
|
<span class="p">[</span><span class="n">quant_scales</span><span class="p">[</span><span class="s1">'dense_act'</span><span class="p">][</span><span class="n">layer_idx</span><span class="p">]],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">fake_fp8_sf_dt</span><span class="p">)</span>
|
|
<span class="n">layer</span><span class="o">.</span><span class="n">attention</span><span class="o">.</span><span class="n">dense</span><span class="o">.</span><span class="n">weights_scaling_factor</span><span class="o">.</span><span class="n">value</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span>
|
|
<span class="p">[</span><span class="n">quant_scales</span><span class="p">[</span><span class="s1">'dense_weights'</span><span class="p">][</span><span class="n">layer_idx</span><span class="p">]],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">fake_fp8_sf_dt</span><span class="p">)</span>
|
|
|
|
<span class="k">return</span> <span class="n">layer</span>
|
|
|
|
|
|
<span class="k">def</span> <span class="nf">_default_fp8_quantize</span><span class="p">(</span><span class="n">model</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">GPTLMHeadModel</span><span class="p">,</span> <span class="n">LLaMAForCausalLM</span><span class="p">,</span>
|
|
<span class="n">GPTJForCausalLM</span><span class="p">],</span>
|
|
<span class="n">quant_mode</span><span class="p">:</span> <span class="n">QuantMode</span><span class="p">,</span>
|
|
<span class="n">quant_scales</span><span class="p">:</span> <span class="nb">dict</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span>
|
|
<span class="w"> </span><span class="sd">"""</span>
|
|
<span class="sd"> Quantize all linear layers (i.e., MLP, Attention QKV/Dense) and KV cache IO with dummy scales</span>
|
|
<span class="sd"> This is used by benchmark script and therefore is intentionally decoupled from AMMO toolkit</span>
|
|
<span class="sd"> """</span>
|
|
<span class="k">if</span> <span class="n">quant_scales</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
|
<span class="n">num_layers</span> <span class="o">=</span> <span class="nb">getattr</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="s1">'_num_layers'</span><span class="p">,</span>
|
|
<span class="nb">getattr</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="s1">'num_layers'</span><span class="p">,</span> <span class="kc">None</span><span class="p">))</span>
|
|
<span class="k">assert</span> <span class="n">num_layers</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
|
|
<span class="n">quant_scales</span> <span class="o">=</span> <span class="n">get_dummy_quant_scales</span><span class="p">(</span><span class="n">num_layers</span><span class="p">)</span>
|
|
|
|
<span class="k">assert</span> <span class="n">model</span><span class="o">.</span><span class="n">quant_mode</span> <span class="o">==</span> <span class="n">quant_mode</span><span class="p">,</span> <span class="s2">"Quant setting not consistent with model init setting"</span>
|
|
|
|
<span class="n">use_fp8_qdq</span> <span class="o">=</span> <span class="n">quant_mode</span><span class="o">.</span><span class="n">has_fp8_qdq</span><span class="p">()</span>
|
|
<span class="k">assert</span> <span class="n">use_fp8_qdq</span>
|
|
|
|
<span class="k">for</span> <span class="n">layer_idx</span><span class="p">,</span> <span class="n">layer</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">model</span><span class="o">.</span><span class="n">layers</span><span class="p">):</span>
|
|
<span class="n">layer</span> <span class="o">=</span> <span class="n">_quantize_layer</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="n">layer_idx</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">,</span> <span class="n">quant_scales</span><span class="p">)</span>
|
|
|
|
<span class="c1"># TODO: add lm_head</span>
|
|
|
|
<span class="k">return</span> <span class="n">model</span>
|
|
|
|
|
|
<span class="k">def</span> <span class="nf">_fp8_quantize</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">:</span> <span class="n">QuantMode</span><span class="p">,</span> <span class="n">quant_scales</span><span class="p">:</span> <span class="nb">dict</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span>
|
|
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span>
|
|
<span class="n">model</span><span class="p">,</span>
|
|
<span class="p">(</span><span class="n">FalconForCausalLM</span><span class="p">,</span> <span class="n">GPTJForCausalLM</span><span class="p">,</span> <span class="n">GPTLMHeadModel</span><span class="p">,</span> <span class="n">LLaMAForCausalLM</span><span class="p">)):</span>
|
|
<span class="k">return</span> <span class="n">_default_fp8_quantize</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">quant_mode</span><span class="p">,</span> <span class="n">quant_scales</span><span class="p">)</span>
|
|
<span class="k">raise</span> <span class="ne">NotImplementedError</span><span class="p">(</span>
|
|
<span class="sa">f</span><span class="s2">"Model </span><span class="si">{</span><span class="n">model</span><span class="si">}</span><span class="s2"> is not implemented by fp8_quantize yet"</span><span class="p">)</span>
|
|
</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> |