mirror of
https://github.com/labmlai/annotated_deep_learning_paper_implementations.git
synced 2025-10-29 01:26:44 +08:00
Chineese translation
This commit is contained in:
Varuna Jayasiri
@ -12,7 +12,7 @@
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/RWKV/configs.html"/>
|
||||
<meta property="og:url" content="https://nn.labml.ai/rwkv/configs.html"/>
|
||||
<meta property="og:title" content="configs.py"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="configs.py"/>
|
||||
@ -23,7 +23,7 @@
|
||||
<title>configs.py</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/RWKV/configs.html"/>
|
||||
<link rel="canonical" href="https://nn.labml.ai/rwkv/configs.html"/>
|
||||
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.13.18/dist/katex.min.css" integrity="sha384-zTROYFVGOfTw7JV7KUu8udsvW2fx4lWOsCEDqhBreBwlHI4ioVRtmIvEThzJHGET" crossorigin="anonymous">
|
||||
|
||||
<!-- Global site tag (gtag.js) - Google Analytics -->
|
||||
@ -47,7 +47,7 @@
|
||||
<div class='docs'>
|
||||
<p>
|
||||
<a class="parent" href="/">home</a>
|
||||
<a class="parent" href="index.html">RWKV</a>
|
||||
<a class="parent" href="index.html">rwkv</a>
|
||||
</p>
|
||||
<p>
|
||||
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations" target="_blank">
|
||||
@ -60,7 +60,7 @@
|
||||
style="max-width:100%;"/></a>
|
||||
</p>
|
||||
<p>
|
||||
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/RWKV/configs.py" target="_blank">
|
||||
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/rwkv/configs.py" target="_blank">
|
||||
View code on Github</a>
|
||||
</p>
|
||||
</div>
|
||||
|
||||
@ -12,7 +12,7 @@
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/RWKV/experiment.html"/>
|
||||
<meta property="og:url" content="https://nn.labml.ai/rwkv/experiment.html"/>
|
||||
<meta property="og:title" content="experiment.py"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="experiment.py"/>
|
||||
@ -23,7 +23,7 @@
|
||||
<title>experiment.py</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/RWKV/experiment.html"/>
|
||||
<link rel="canonical" href="https://nn.labml.ai/rwkv/experiment.html"/>
|
||||
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.13.18/dist/katex.min.css" integrity="sha384-zTROYFVGOfTw7JV7KUu8udsvW2fx4lWOsCEDqhBreBwlHI4ioVRtmIvEThzJHGET" crossorigin="anonymous">
|
||||
|
||||
<!-- Global site tag (gtag.js) - Google Analytics -->
|
||||
@ -47,7 +47,7 @@
|
||||
<div class='docs'>
|
||||
<p>
|
||||
<a class="parent" href="/">home</a>
|
||||
<a class="parent" href="index.html">RWKV</a>
|
||||
<a class="parent" href="index.html">rwkv</a>
|
||||
</p>
|
||||
<p>
|
||||
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations" target="_blank">
|
||||
@ -60,7 +60,7 @@
|
||||
style="max-width:100%;"/></a>
|
||||
</p>
|
||||
<p>
|
||||
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/RWKV/experiment.py" target="_blank">
|
||||
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/rwkv/experiment.py" target="_blank">
|
||||
View code on Github</a>
|
||||
</p>
|
||||
</div>
|
||||
@ -78,10 +78,10 @@
|
||||
<span class="lineno">3</span>
|
||||
<span class="lineno">4</span><span class="kn">import</span> <span class="nn">torch</span>
|
||||
<span class="lineno">5</span><span class="kn">import</span> <span class="nn">torch.nn</span> <span class="k">as</span> <span class="nn">nn</span>
|
||||
<span class="lineno">6</span><span class="kn">from</span> <span class="nn">labml_nn.RWKV.configs</span> <span class="kn">import</span> <span class="n">RWKVConfigs</span>
|
||||
<span class="lineno">6</span><span class="kn">from</span> <span class="nn">labml_nn.rwkv.configs</span> <span class="kn">import</span> <span class="n">RWKVConfigs</span>
|
||||
<span class="lineno">7</span>
|
||||
<span class="lineno">8</span><span class="kn">from</span> <span class="nn">labml_nn.RWKV</span> <span class="kn">import</span> <span class="n">RWKV</span>
|
||||
<span class="lineno">9</span><span class="kn">from</span> <span class="nn">labml_nn.RWKV</span> <span class="kn">import</span> <span class="n">TimeMixing</span>
|
||||
<span class="lineno">8</span><span class="kn">from</span> <span class="nn">labml_nn.rwkv</span> <span class="kn">import</span> <span class="n">RWKV</span>
|
||||
<span class="lineno">9</span><span class="kn">from</span> <span class="nn">labml_nn.rwkv</span> <span class="kn">import</span> <span class="n">TimeMixing</span>
|
||||
<span class="lineno">10</span><span class="kn">from</span> <span class="nn">labml</span> <span class="kn">import</span> <span class="n">experiment</span>
|
||||
<span class="lineno">11</span><span class="kn">from</span> <span class="nn">labml.configs</span> <span class="kn">import</span> <span class="n">option</span>
|
||||
<span class="lineno">12</span><span class="kn">from</span> <span class="nn">labml_nn.experiments.nlp_autoregression</span> <span class="kn">import</span> <span class="n">NLPAutoRegressionConfigs</span></pre></div>
|
||||
|
||||
@ -12,7 +12,7 @@
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/RWKV/index.html"/>
|
||||
<meta property="og:url" content="https://nn.labml.ai/rwkv/index.html"/>
|
||||
<meta property="og:title" content="Receptance Weighted Key Value (RWKV)"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="Receptance Weighted Key Value (RWKV)"/>
|
||||
@ -23,7 +23,7 @@
|
||||
<title>Receptance Weighted Key Value (RWKV)</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/RWKV/index.html"/>
|
||||
<link rel="canonical" href="https://nn.labml.ai/rwkv/index.html"/>
|
||||
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.13.18/dist/katex.min.css" integrity="sha384-zTROYFVGOfTw7JV7KUu8udsvW2fx4lWOsCEDqhBreBwlHI4ioVRtmIvEThzJHGET" crossorigin="anonymous">
|
||||
|
||||
<!-- Global site tag (gtag.js) - Google Analytics -->
|
||||
@ -47,7 +47,7 @@
|
||||
<div class='docs'>
|
||||
<p>
|
||||
<a class="parent" href="/">home</a>
|
||||
<a class="parent" href="index.html">RWKV</a>
|
||||
<a class="parent" href="index.html">rwkv</a>
|
||||
</p>
|
||||
<p>
|
||||
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations" target="_blank">
|
||||
@ -60,7 +60,7 @@
|
||||
style="max-width:100%;"/></a>
|
||||
</p>
|
||||
<p>
|
||||
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/RWKV/__init__.py" target="_blank">
|
||||
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/rwkv/__init__.py" target="_blank">
|
||||
View code on Github</a>
|
||||
</p>
|
||||
</div>
|
||||
|
||||
File diff suppressed because one or more lines are too long
480
docs/zh/lora/experiment.html
Normal file
480
docs/zh/lora/experiment.html
Normal file
@ -0,0 +1,480 @@
|
||||
<!DOCTYPE html>
|
||||
<html lang="zh">
|
||||
<head>
|
||||
<meta http-equiv="content-type" content="text/html;charset=utf-8"/>
|
||||
<meta name="viewport" content="width=device-width, initial-scale=1.0"/>
|
||||
<meta name="description" content="This is training code with notes for fine-tuning pre-trained GPT-2 model with LoRA."/>
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="Finetune GPT-2 with LoRA"/>
|
||||
<meta name="twitter:description" content="This is training code with notes for fine-tuning pre-trained GPT-2 model with LoRA."/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/lora/experiment.html"/>
|
||||
<meta property="og:title" content="Finetune GPT-2 with LoRA"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="Finetune GPT-2 with LoRA"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="Finetune GPT-2 with LoRA"/>
|
||||
<meta property="og:description" content="This is training code with notes for fine-tuning pre-trained GPT-2 model with LoRA."/>
|
||||
|
||||
<title>Finetune GPT-2 with LoRA</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/lora/experiment.html"/>
|
||||
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.13.18/dist/katex.min.css" integrity="sha384-zTROYFVGOfTw7JV7KUu8udsvW2fx4lWOsCEDqhBreBwlHI4ioVRtmIvEThzJHGET" crossorigin="anonymous">
|
||||
|
||||
<!-- Global site tag (gtag.js) - Google Analytics -->
|
||||
<script async src="https://www.googletagmanager.com/gtag/js?id=G-4V3HC8HBLH"></script>
|
||||
<script>
|
||||
window.dataLayer = window.dataLayer || [];
|
||||
|
||||
function gtag() {
|
||||
dataLayer.push(arguments);
|
||||
}
|
||||
|
||||
gtag('js', new Date());
|
||||
|
||||
gtag('config', 'G-4V3HC8HBLH');
|
||||
</script>
|
||||
</head>
|
||||
<body>
|
||||
<div id='container'>
|
||||
<div id="background"></div>
|
||||
<div class='section'>
|
||||
<div class='docs'>
|
||||
<p>
|
||||
<a class="parent" href="/">home</a>
|
||||
<a class="parent" href="index.html">lora</a>
|
||||
</p>
|
||||
<p>
|
||||
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations" target="_blank">
|
||||
<img alt="Github"
|
||||
src="https://img.shields.io/github/stars/labmlai/annotated_deep_learning_paper_implementations?style=social"
|
||||
style="max-width:100%;"/></a>
|
||||
<a href="https://twitter.com/labmlai" rel="nofollow" target="_blank">
|
||||
<img alt="Twitter"
|
||||
src="https://img.shields.io/twitter/follow/labmlai?style=social"
|
||||
style="max-width:100%;"/></a>
|
||||
</p>
|
||||
<p>
|
||||
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/lora/experiment.py" target="_blank">
|
||||
View code on Github</a>
|
||||
</p>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-0'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-0'>#</a>
|
||||
</div>
|
||||
<h1>Finetune GPT-2 with <a href="index.html">LoRA</a></h1>
|
||||
<p>Here's a Colab notebook for training a feedback transformer on Tiny Shakespeare dataset.</p>
|
||||
<p><a href="https://colab.research.google.com/github/labmlai/annotated_deep_learning_paper_implementations/blob/master/labml_nn/lora/experiment.ipynb"><img alt="Open In Colab" src="https://colab.research.google.com/assets/colab-badge.svg"></a></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">14</span><span></span><span class="kn">import</span> <span class="nn">torch</span>
|
||||
<span class="lineno">15</span><span class="kn">from</span> <span class="nn">labml</span> <span class="kn">import</span> <span class="n">lab</span><span class="p">,</span> <span class="n">monit</span><span class="p">,</span> <span class="n">tracker</span>
|
||||
<span class="lineno">16</span><span class="kn">from</span> <span class="nn">labml.configs</span> <span class="kn">import</span> <span class="n">BaseConfigs</span><span class="p">,</span> <span class="n">option</span>
|
||||
<span class="lineno">17</span><span class="kn">from</span> <span class="nn">labml.utils.download</span> <span class="kn">import</span> <span class="n">download_file</span>
|
||||
<span class="lineno">18</span><span class="kn">from</span> <span class="nn">labml_helpers.device</span> <span class="kn">import</span> <span class="n">DeviceConfigs</span>
|
||||
<span class="lineno">19</span><span class="kn">from</span> <span class="nn">torch.optim</span> <span class="kn">import</span> <span class="n">Adam</span>
|
||||
<span class="lineno">20</span><span class="kn">from</span> <span class="nn">torch.utils.data</span> <span class="kn">import</span> <span class="n">DataLoader</span><span class="p">,</span> <span class="n">TensorDataset</span>
|
||||
<span class="lineno">21</span><span class="kn">from</span> <span class="nn">transformers</span> <span class="kn">import</span> <span class="n">AutoTokenizer</span><span class="p">,</span> <span class="n">AutoModelForCausalLM</span>
|
||||
<span class="lineno">22</span><span class="kn">from</span> <span class="nn">labml_nn.lora.gpt2</span> <span class="kn">import</span> <span class="n">GPTModel</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-1'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-1'>#</a>
|
||||
</div>
|
||||
<h2>Trainer configurations and the training loop</h2>
|
||||
<p>The default configs can and will be over-ridden when we start the experiment</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">25</span><span class="k">class</span> <span class="nc">Trainer</span><span class="p">(</span><span class="n">BaseConfigs</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-2'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-2'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">31</span> <span class="n">device</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">device</span> <span class="o">=</span> <span class="n">DeviceConfigs</span><span class="p">()</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-3'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-3'>#</a>
|
||||
</div>
|
||||
<p>GPT-2 configs </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">34</span> <span class="n">layer_norm_epsilon</span><span class="p">:</span> <span class="nb">float</span> <span class="o">=</span> <span class="mf">1e-05</span>
|
||||
<span class="lineno">35</span> <span class="n">n_embed</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">768</span>
|
||||
<span class="lineno">36</span> <span class="n">n_layer</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">12</span>
|
||||
<span class="lineno">37</span> <span class="n">n_positions</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">1024</span>
|
||||
<span class="lineno">38</span> <span class="n">vocab_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">50257</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-4'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-4'>#</a>
|
||||
</div>
|
||||
<p>Training configs </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">41</span> <span class="n">epochs</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">10</span>
|
||||
<span class="lineno">42</span> <span class="n">batch_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">32</span>
|
||||
<span class="lineno">43</span> <span class="n">learning_rate</span><span class="p">:</span> <span class="nb">float</span> <span class="o">=</span> <span class="mf">1e-4</span>
|
||||
<span class="lineno">44</span> <span class="n">context_len</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">512</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-5'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-5'>#</a>
|
||||
</div>
|
||||
<p>LoRA rank </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">47</span> <span class="n">lora_r</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">32</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-6'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-6'>#</a>
|
||||
</div>
|
||||
<p>Dataset </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">50</span> <span class="n">text</span><span class="p">:</span> <span class="n">TensorDataset</span> <span class="o">=</span> <span class="s2">"tiny_shakespeare"</span>
|
||||
<span class="lineno">51</span> <span class="n">tokenizer</span> <span class="o">=</span> <span class="n">AutoTokenizer</span><span class="o">.</span><span class="n">from_pretrained</span><span class="p">(</span><span class="s2">"gpt2"</span><span class="p">)</span>
|
||||
<span class="lineno">52</span> <span class="n">model</span><span class="p">:</span> <span class="n">GPTModel</span>
|
||||
<span class="lineno">53</span> <span class="n">optimizer</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">optim</span><span class="o">.</span><span class="n">Adam</span>
|
||||
<span class="lineno">54</span> <span class="n">criterion</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">CrossEntropyLoss</span><span class="p">()</span>
|
||||
<span class="lineno">55</span> <span class="n">data_loader</span><span class="p">:</span> <span class="n">DataLoader</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-7'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-7'>#</a>
|
||||
</div>
|
||||
<h3>Load pre-trained <a href="https://huggingface.co/openai-community/gpt2">GPT-2 from huggingface</a></h3>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">57</span> <span class="k">def</span> <span class="nf">_load_pretrained_weights</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-8'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-8'>#</a>
|
||||
</div>
|
||||
<p>Load the huggingface model and get the parameters </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">63</span> <span class="n">hf_model</span> <span class="o">=</span> <span class="n">AutoModelForCausalLM</span><span class="o">.</span><span class="n">from_pretrained</span><span class="p">(</span><span class="s2">"gpt2"</span><span class="p">)</span>
|
||||
<span class="lineno">64</span> <span class="n">state_dict</span> <span class="o">=</span> <span class="n">hf_model</span><span class="o">.</span><span class="n">state_dict</span><span class="p">()</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-9'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-9'>#</a>
|
||||
</div>
|
||||
<p>Transformer embedding and prediction layer parameter mapping (<code class="highlight"><span></span><span class="n">hf</span><span class="p">:</span> <span class="n">ours</span></code>
|
||||
) </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">67</span> <span class="n">mapping</span> <span class="o">=</span> <span class="p">{</span>
|
||||
<span class="lineno">68</span> <span class="s1">'transformer.wte.weight'</span><span class="p">:</span> <span class="s1">'token_embedding.weight'</span><span class="p">,</span>
|
||||
<span class="lineno">69</span> <span class="s1">'transformer.wpe.weight'</span><span class="p">:</span> <span class="s1">'position_embedding.weight'</span><span class="p">,</span>
|
||||
<span class="lineno">70</span> <span class="s1">'transformer.ln_f.weight'</span><span class="p">:</span> <span class="s1">'final_norm.weight'</span><span class="p">,</span>
|
||||
<span class="lineno">71</span> <span class="s1">'transformer.ln_f.bias'</span><span class="p">:</span> <span class="s1">'final_norm.bias'</span><span class="p">,</span>
|
||||
<span class="lineno">72</span> <span class="s1">'lm_head.weight'</span><span class="p">:</span> <span class="s1">'lm_head.weight'</span>
|
||||
<span class="lineno">73</span> <span class="p">}</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-10'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-10'>#</a>
|
||||
</div>
|
||||
<p>Mapping (<code class="highlight"><span></span><span class="n">hf</span><span class="p">:</span> <span class="n">ours</span></code>
|
||||
) of decoder layers </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">76</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">12</span><span class="p">):</span>
|
||||
<span class="lineno">77</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">'transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ln_1.weight'</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">'blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.pre_norm.weight'</span>
|
||||
<span class="lineno">78</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">'transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ln_1.bias'</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">'blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.pre_norm.bias'</span>
|
||||
<span class="lineno">79</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">'transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_attn.weight'</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">'blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_att.weight'</span>
|
||||
<span class="lineno">80</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">'transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_attn.bias'</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">'blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_att.bias'</span>
|
||||
<span class="lineno">81</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">'transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_proj.weight'</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">'blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_proj.weight'</span>
|
||||
<span class="lineno">82</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">'transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_proj.bias'</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">'blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_proj.bias'</span>
|
||||
<span class="lineno">83</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">'transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ln_2.weight'</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">'blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.post_norm.weight'</span>
|
||||
<span class="lineno">84</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">'transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ln_2.bias'</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">'blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.post_norm.bias'</span>
|
||||
<span class="lineno">85</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">'transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.mlp.c_fc.weight'</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">'blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ffn.c_fc.weight'</span>
|
||||
<span class="lineno">86</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">'transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.mlp.c_fc.bias'</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">'blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ffn.c_fc.bias'</span>
|
||||
<span class="lineno">87</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">'transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.mlp.c_proj.weight'</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">'blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ffn.c_proj.weight'</span>
|
||||
<span class="lineno">88</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">'transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.mlp.c_proj.bias'</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">'blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ffn.c_proj.bias'</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-11'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-11'>#</a>
|
||||
</div>
|
||||
<p>Move the parameters based on mapping </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">91</span> <span class="n">new_state_dict</span> <span class="o">=</span> <span class="p">{}</span>
|
||||
<span class="lineno">92</span> <span class="k">for</span> <span class="n">old_key</span><span class="p">,</span> <span class="n">new_key</span> <span class="ow">in</span> <span class="n">mapping</span><span class="o">.</span><span class="n">items</span><span class="p">():</span>
|
||||
<span class="lineno">93</span> <span class="k">if</span> <span class="n">old_key</span> <span class="ow">in</span> <span class="n">state_dict</span><span class="p">:</span>
|
||||
<span class="lineno">94</span> <span class="n">new_state_dict</span><span class="p">[</span><span class="n">new_key</span><span class="p">]</span> <span class="o">=</span> <span class="n">state_dict</span><span class="p">[</span><span class="n">old_key</span><span class="p">]</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-12'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-12'>#</a>
|
||||
</div>
|
||||
<p>GPT-2 hugging face uses 1D Convolution layers. We need to transpose those weights since we use linear layers </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">97</span> <span class="n">convo_layers</span> <span class="o">=</span> <span class="p">([</span><span class="sa">f</span><span class="s1">'blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ffn.c_fc.weight'</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">12</span><span class="p">)]</span> <span class="o">+</span>
|
||||
<span class="lineno">98</span> <span class="p">[</span><span class="sa">f</span><span class="s1">'blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ffn.c_proj.weight'</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">12</span><span class="p">)]</span> <span class="o">+</span>
|
||||
<span class="lineno">99</span> <span class="p">[</span><span class="sa">f</span><span class="s1">'blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_att.weight'</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">12</span><span class="p">)]</span> <span class="o">+</span>
|
||||
<span class="lineno">100</span> <span class="p">[</span><span class="sa">f</span><span class="s1">'blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_proj.weight'</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">12</span><span class="p">)])</span>
|
||||
<span class="lineno">101</span>
|
||||
<span class="lineno">102</span> <span class="k">for</span> <span class="n">layer</span> <span class="ow">in</span> <span class="n">convo_layers</span><span class="p">:</span>
|
||||
<span class="lineno">103</span> <span class="n">new_state_dict</span><span class="p">[</span><span class="n">layer</span><span class="p">]</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">transpose</span><span class="p">(</span><span class="n">new_state_dict</span><span class="p">[</span><span class="n">layer</span><span class="p">],</span> <span class="mi">0</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-13'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-13'>#</a>
|
||||
</div>
|
||||
<p>Load out model </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">106</span> <span class="bp">self</span><span class="o">.</span><span class="n">model</span><span class="o">.</span><span class="n">load_state_dict</span><span class="p">(</span><span class="n">new_state_dict</span><span class="p">,</span> <span class="n">strict</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span> <span class="c1"># state dict does not have lora weights</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-14'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-14'>#</a>
|
||||
</div>
|
||||
<h3>Initialize the model, optimizer and dataloader</h3>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">108</span> <span class="k">def</span> <span class="nf">initialize</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-15'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-15'>#</a>
|
||||
</div>
|
||||
<p>Initialize the model </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">113</span> <span class="bp">self</span><span class="o">.</span><span class="n">model</span> <span class="o">=</span> <span class="n">GPTModel</span><span class="p">(</span>
|
||||
<span class="lineno">114</span> <span class="n">layer_norm_epsilon</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">layer_norm_epsilon</span><span class="p">,</span>
|
||||
<span class="lineno">115</span> <span class="n">n_embd</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">n_embed</span><span class="p">,</span>
|
||||
<span class="lineno">116</span> <span class="n">n_layer</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">n_layer</span><span class="p">,</span>
|
||||
<span class="lineno">117</span> <span class="n">n_positions</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">n_positions</span><span class="p">,</span>
|
||||
<span class="lineno">118</span> <span class="n">vocab_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">vocab_size</span><span class="p">,</span>
|
||||
<span class="lineno">119</span> <span class="n">r</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">lora_r</span><span class="p">,</span>
|
||||
<span class="lineno">120</span> <span class="p">)</span>
|
||||
<span class="lineno">121</span> <span class="bp">self</span><span class="o">.</span><span class="n">model</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-16'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-16'>#</a>
|
||||
</div>
|
||||
<p>Load pre-trained model weights </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">123</span> <span class="bp">self</span><span class="o">.</span><span class="n">_load_pretrained_weights</span><span class="p">()</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-17'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-17'>#</a>
|
||||
</div>
|
||||
<p>Initialize the optimizer </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">126</span> <span class="bp">self</span><span class="o">.</span><span class="n">optimizer</span> <span class="o">=</span> <span class="n">Adam</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">model</span><span class="o">.</span><span class="n">parameters</span><span class="p">(),</span> <span class="n">lr</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">learning_rate</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-18'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-18'>#</a>
|
||||
</div>
|
||||
<p>Initialize the data loader </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">129</span> <span class="bp">self</span><span class="o">.</span><span class="n">data_loader</span> <span class="o">=</span> <span class="n">DataLoader</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">text</span><span class="p">,</span> <span class="n">batch_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">batch_size</span><span class="p">,</span> <span class="n">shuffle</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-19'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-19'>#</a>
|
||||
</div>
|
||||
<h3>Training loop</h3>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">131</span> <span class="k">def</span> <span class="nf">run</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-20'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-20'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">136</span> <span class="k">for</span> <span class="n">_</span> <span class="ow">in</span> <span class="n">monit</span><span class="o">.</span><span class="n">loop</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">epochs</span><span class="p">):</span>
|
||||
<span class="lineno">137</span> <span class="k">for</span> <span class="n">i</span><span class="p">,</span> <span class="n">batch</span> <span class="ow">in</span> <span class="n">monit</span><span class="o">.</span><span class="n">enum</span><span class="p">(</span><span class="s1">'Train'</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">data_loader</span><span class="p">):</span>
|
||||
<span class="lineno">138</span> <span class="n">inputs</span> <span class="o">=</span> <span class="n">batch</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
|
||||
<span class="lineno">139</span> <span class="n">inputs</span> <span class="o">=</span> <span class="n">inputs</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
|
||||
<span class="lineno">140</span> <span class="n">labels</span> <span class="o">=</span> <span class="n">inputs</span><span class="o">.</span><span class="n">clone</span><span class="p">()</span>
|
||||
<span class="lineno">141</span>
|
||||
<span class="lineno">142</span> <span class="n">outputs</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">model</span><span class="p">(</span><span class="n">inputs</span><span class="p">)</span>
|
||||
<span class="lineno">143</span>
|
||||
<span class="lineno">144</span> <span class="n">shift_logits</span> <span class="o">=</span> <span class="n">outputs</span><span class="p">[</span><span class="o">...</span><span class="p">,</span> <span class="p">:</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="p">:]</span>
|
||||
<span class="lineno">145</span> <span class="n">shift_labels</span> <span class="o">=</span> <span class="n">labels</span><span class="p">[</span><span class="o">...</span><span class="p">,</span> <span class="mi">1</span><span class="p">:]</span>
|
||||
<span class="lineno">146</span>
|
||||
<span class="lineno">147</span> <span class="n">loss</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">criterion</span><span class="p">(</span><span class="n">shift_logits</span><span class="o">.</span><span class="n">reshape</span><span class="p">(</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="n">shift_logits</span><span class="o">.</span><span class="n">size</span><span class="p">(</span><span class="o">-</span><span class="mi">1</span><span class="p">)),</span> <span class="n">shift_labels</span><span class="o">.</span><span class="n">reshape</span><span class="p">(</span><span class="o">-</span><span class="mi">1</span><span class="p">))</span>
|
||||
<span class="lineno">148</span>
|
||||
<span class="lineno">149</span> <span class="bp">self</span><span class="o">.</span><span class="n">optimizer</span><span class="o">.</span><span class="n">zero_grad</span><span class="p">()</span>
|
||||
<span class="lineno">150</span> <span class="n">loss</span><span class="o">.</span><span class="n">backward</span><span class="p">()</span>
|
||||
<span class="lineno">151</span> <span class="bp">self</span><span class="o">.</span><span class="n">optimizer</span><span class="o">.</span><span class="n">step</span><span class="p">()</span>
|
||||
<span class="lineno">152</span>
|
||||
<span class="lineno">153</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add</span><span class="p">({</span><span class="s1">'loss'</span><span class="p">:</span> <span class="n">loss</span><span class="p">})</span>
|
||||
<span class="lineno">154</span>
|
||||
<span class="lineno">155</span> <span class="n">tracker</span><span class="o">.</span><span class="n">save</span><span class="p">()</span>
|
||||
<span class="lineno">156</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add_global_step</span><span class="p">()</span>
|
||||
<span class="lineno">157</span> <span class="n">tracker</span><span class="o">.</span><span class="n">new_line</span><span class="p">()</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-21'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-21'>#</a>
|
||||
</div>
|
||||
<h3>Tiny Shakespeare dataset</h3>
|
||||
<p>It will download from the url if not present</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">160</span><span class="nd">@option</span><span class="p">(</span><span class="n">Trainer</span><span class="o">.</span><span class="n">text</span><span class="p">)</span>
|
||||
<span class="lineno">161</span><span class="k">def</span> <span class="nf">tiny_shakespeare</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">Trainer</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-22'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-22'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">167</span> <span class="n">path</span> <span class="o">=</span> <span class="n">lab</span><span class="o">.</span><span class="n">get_data_path</span><span class="p">()</span> <span class="o">/</span> <span class="s1">'tiny_shakespeare.txt'</span>
|
||||
<span class="lineno">168</span> <span class="k">if</span> <span class="ow">not</span> <span class="n">path</span><span class="o">.</span><span class="n">exists</span><span class="p">():</span>
|
||||
<span class="lineno">169</span> <span class="n">download_file</span><span class="p">(</span><span class="s2">"https://raw.githubusercontent.com/karpathy/char-rnn/master/data/tinyshakespeare/input.txt"</span><span class="p">,</span> <span class="n">path</span><span class="p">)</span>
|
||||
<span class="lineno">170</span> <span class="k">with</span> <span class="nb">open</span><span class="p">(</span><span class="n">path</span><span class="p">,</span> <span class="s1">'r'</span><span class="p">,</span> <span class="n">encoding</span><span class="o">=</span><span class="s1">'utf-8'</span><span class="p">)</span> <span class="k">as</span> <span class="n">f</span><span class="p">:</span>
|
||||
<span class="lineno">171</span> <span class="n">text</span> <span class="o">=</span> <span class="n">f</span><span class="o">.</span><span class="n">read</span><span class="p">()</span>
|
||||
<span class="lineno">172</span>
|
||||
<span class="lineno">173</span> <span class="n">tokens</span> <span class="o">=</span> <span class="n">c</span><span class="o">.</span><span class="n">tokenizer</span><span class="o">.</span><span class="n">encode</span><span class="p">(</span><span class="n">text</span><span class="p">)</span>
|
||||
<span class="lineno">174</span> <span class="n">num_batches</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="n">tokens</span><span class="p">)</span> <span class="o">//</span> <span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">batch_size</span> <span class="o">*</span> <span class="n">c</span><span class="o">.</span><span class="n">context_len</span><span class="p">)</span>
|
||||
<span class="lineno">175</span> <span class="n">tokens</span> <span class="o">=</span> <span class="n">tokens</span><span class="p">[:</span><span class="n">num_batches</span> <span class="o">*</span> <span class="n">c</span><span class="o">.</span><span class="n">batch_size</span> <span class="o">*</span> <span class="n">c</span><span class="o">.</span><span class="n">context_len</span><span class="p">]</span>
|
||||
<span class="lineno">176</span> <span class="n">input_ids</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">(</span><span class="n">tokens</span><span class="p">)</span><span class="o">.</span><span class="n">view</span><span class="p">(</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">context_len</span><span class="p">)</span>
|
||||
<span class="lineno">177</span> <span class="k">return</span> <span class="n">TensorDataset</span><span class="p">(</span><span class="n">input_ids</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='footer'>
|
||||
<a href="https://labml.ai">labml.ai</a>
|
||||
</div>
|
||||
</div>
|
||||
<script src=../interactive.js?v=1"></script>
|
||||
<script>
|
||||
function handleImages() {
|
||||
var images = document.querySelectorAll('p>img')
|
||||
|
||||
for (var i = 0; i < images.length; ++i) {
|
||||
handleImage(images[i])
|
||||
}
|
||||
}
|
||||
|
||||
function handleImage(img) {
|
||||
img.parentElement.style.textAlign = 'center'
|
||||
|
||||
var modal = document.createElement('div')
|
||||
modal.id = 'modal'
|
||||
|
||||
var modalContent = document.createElement('div')
|
||||
modal.appendChild(modalContent)
|
||||
|
||||
var modalImage = document.createElement('img')
|
||||
modalContent.appendChild(modalImage)
|
||||
|
||||
var span = document.createElement('span')
|
||||
span.classList.add('close')
|
||||
span.textContent = 'x'
|
||||
modal.appendChild(span)
|
||||
|
||||
img.onclick = function () {
|
||||
console.log('clicked')
|
||||
document.body.appendChild(modal)
|
||||
modalImage.src = img.src
|
||||
}
|
||||
|
||||
span.onclick = function () {
|
||||
document.body.removeChild(modal)
|
||||
}
|
||||
}
|
||||
|
||||
handleImages()
|
||||
</script>
|
||||
</body>
|
||||
</html>
|
||||
493
docs/zh/lora/gpt2.html
Normal file
493
docs/zh/lora/gpt2.html
Normal file
@ -0,0 +1,493 @@
|
||||
<!DOCTYPE html>
|
||||
<html lang="zh">
|
||||
<head>
|
||||
<meta http-equiv="content-type" content="text/html;charset=utf-8"/>
|
||||
<meta name="viewport" content="width=device-width, initial-scale=1.0"/>
|
||||
<meta name="description" content=""/>
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="gpt2.py"/>
|
||||
<meta name="twitter:description" content=""/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/lora/gpt2.html"/>
|
||||
<meta property="og:title" content="gpt2.py"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="gpt2.py"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="gpt2.py"/>
|
||||
<meta property="og:description" content=""/>
|
||||
|
||||
<title>gpt2.py</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/lora/gpt2.html"/>
|
||||
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.13.18/dist/katex.min.css" integrity="sha384-zTROYFVGOfTw7JV7KUu8udsvW2fx4lWOsCEDqhBreBwlHI4ioVRtmIvEThzJHGET" crossorigin="anonymous">
|
||||
|
||||
<!-- Global site tag (gtag.js) - Google Analytics -->
|
||||
<script async src="https://www.googletagmanager.com/gtag/js?id=G-4V3HC8HBLH"></script>
|
||||
<script>
|
||||
window.dataLayer = window.dataLayer || [];
|
||||
|
||||
function gtag() {
|
||||
dataLayer.push(arguments);
|
||||
}
|
||||
|
||||
gtag('js', new Date());
|
||||
|
||||
gtag('config', 'G-4V3HC8HBLH');
|
||||
</script>
|
||||
</head>
|
||||
<body>
|
||||
<div id='container'>
|
||||
<div id="background"></div>
|
||||
<div class='section'>
|
||||
<div class='docs'>
|
||||
<p>
|
||||
<a class="parent" href="/">home</a>
|
||||
<a class="parent" href="index.html">lora</a>
|
||||
</p>
|
||||
<p>
|
||||
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations" target="_blank">
|
||||
<img alt="Github"
|
||||
src="https://img.shields.io/github/stars/labmlai/annotated_deep_learning_paper_implementations?style=social"
|
||||
style="max-width:100%;"/></a>
|
||||
<a href="https://twitter.com/labmlai" rel="nofollow" target="_blank">
|
||||
<img alt="Twitter"
|
||||
src="https://img.shields.io/twitter/follow/labmlai?style=social"
|
||||
style="max-width:100%;"/></a>
|
||||
</p>
|
||||
<p>
|
||||
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/lora/gpt2.py" target="_blank">
|
||||
View code on Github</a>
|
||||
</p>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-0'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-0'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">1</span><span></span><span class="kn">import</span> <span class="nn">torch</span>
|
||||
<span class="lineno">2</span><span class="kn">import</span> <span class="nn">torch.nn</span> <span class="k">as</span> <span class="nn">nn</span>
|
||||
<span class="lineno">3</span><span class="kn">from</span> <span class="nn">labml_nn.lora</span> <span class="kn">import</span> <span class="n">Linear</span><span class="p">,</span> <span class="n">Embedding</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-1'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-1'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">6</span><span class="k">class</span> <span class="nc">FFN</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-2'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-2'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">7</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">dim</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">n_embed</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">r</span><span class="p">:</span> <span class="nb">int</span><span class="p">):</span>
|
||||
<span class="lineno">8</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-3'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-3'>#</a>
|
||||
</div>
|
||||
<p>lin1 </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">10</span> <span class="bp">self</span><span class="o">.</span><span class="n">c_fc</span> <span class="o">=</span> <span class="n">Linear</span><span class="p">(</span><span class="n">n_embed</span><span class="p">,</span> <span class="n">dim</span><span class="p">,</span> <span class="n">r</span><span class="o">=</span><span class="n">r</span><span class="p">,</span> <span class="n">bias</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-4'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-4'>#</a>
|
||||
</div>
|
||||
<p>lin2 </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">12</span> <span class="bp">self</span><span class="o">.</span><span class="n">c_proj</span> <span class="o">=</span> <span class="n">Linear</span><span class="p">(</span><span class="n">dim</span><span class="p">,</span> <span class="n">n_embed</span><span class="p">,</span> <span class="n">r</span><span class="o">=</span><span class="n">r</span><span class="p">,</span> <span class="n">bias</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
|
||||
<span class="lineno">13</span> <span class="bp">self</span><span class="o">.</span><span class="n">act</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">functional</span><span class="o">.</span><span class="n">gelu</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-5'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-5'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">15</span> <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">):</span>
|
||||
<span class="lineno">16</span> <span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">c_fc</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
|
||||
<span class="lineno">17</span> <span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">act</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
|
||||
<span class="lineno">18</span> <span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">c_proj</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
|
||||
<span class="lineno">19</span> <span class="k">return</span> <span class="n">hidden_states</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-6'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-6'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">22</span><span class="k">class</span> <span class="nc">MultiHeadAttention</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-7'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-7'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">23</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">n_embed</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">r</span><span class="p">:</span> <span class="nb">int</span><span class="p">):</span>
|
||||
<span class="lineno">24</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
|
||||
<span class="lineno">25</span> <span class="bp">self</span><span class="o">.</span><span class="n">embed_dim</span> <span class="o">=</span> <span class="n">n_embed</span>
|
||||
<span class="lineno">26</span> <span class="bp">self</span><span class="o">.</span><span class="n">num_heads</span> <span class="o">=</span> <span class="n">n_embed</span>
|
||||
<span class="lineno">27</span> <span class="bp">self</span><span class="o">.</span><span class="n">head_dim</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">embed_dim</span> <span class="o">//</span> <span class="bp">self</span><span class="o">.</span><span class="n">num_heads</span>
|
||||
<span class="lineno">28</span> <span class="bp">self</span><span class="o">.</span><span class="n">split_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">embed_dim</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-8'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-8'>#</a>
|
||||
</div>
|
||||
<p>qkv </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">31</span> <span class="bp">self</span><span class="o">.</span><span class="n">c_att</span> <span class="o">=</span> <span class="n">Linear</span><span class="p">(</span><span class="n">n_embed</span><span class="p">,</span> <span class="n">n_embed</span> <span class="o">*</span> <span class="mi">3</span><span class="p">,</span> <span class="n">r</span><span class="o">=</span><span class="n">r</span><span class="p">,</span> <span class="n">bias</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-9'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-9'>#</a>
|
||||
</div>
|
||||
<p>out </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">33</span> <span class="bp">self</span><span class="o">.</span><span class="n">c_proj</span> <span class="o">=</span> <span class="n">Linear</span><span class="p">(</span><span class="n">n_embed</span><span class="p">,</span> <span class="n">n_embed</span><span class="p">,</span> <span class="n">r</span><span class="o">=</span><span class="n">r</span><span class="p">,</span> <span class="n">bias</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-10'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-10'>#</a>
|
||||
</div>
|
||||
<p> Splits hidden_size dim into attn_head_size and num_heads</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">35</span> <span class="k">def</span> <span class="nf">_split_heads</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">tensor</span><span class="p">,</span> <span class="n">num_heads</span><span class="p">,</span> <span class="n">attn_head_size</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-11'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-11'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">39</span> <span class="n">new_shape</span> <span class="o">=</span> <span class="n">tensor</span><span class="o">.</span><span class="n">size</span><span class="p">()[:</span><span class="o">-</span><span class="mi">1</span><span class="p">]</span> <span class="o">+</span> <span class="p">(</span><span class="n">num_heads</span><span class="p">,</span> <span class="n">attn_head_size</span><span class="p">)</span>
|
||||
<span class="lineno">40</span> <span class="n">tensor</span> <span class="o">=</span> <span class="n">tensor</span><span class="o">.</span><span class="n">view</span><span class="p">(</span><span class="n">new_shape</span><span class="p">)</span>
|
||||
<span class="lineno">41</span> <span class="k">return</span> <span class="n">tensor</span><span class="o">.</span><span class="n">permute</span><span class="p">(</span><span class="mi">0</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="mi">3</span><span class="p">)</span> <span class="c1"># (batch, head, seq_length, head_features)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-12'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-12'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">43</span> <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">):</span>
|
||||
<span class="lineno">44</span> <span class="n">batch_size</span><span class="p">,</span> <span class="n">seq_length</span><span class="p">,</span> <span class="n">_</span> <span class="o">=</span> <span class="n">hidden_states</span><span class="o">.</span><span class="n">size</span><span class="p">()</span>
|
||||
<span class="lineno">45</span>
|
||||
<span class="lineno">46</span> <span class="n">query</span><span class="p">,</span> <span class="n">key</span><span class="p">,</span> <span class="n">value</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">c_att</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">split_size</span><span class="p">,</span> <span class="n">dim</span><span class="o">=</span><span class="mi">2</span><span class="p">)</span>
|
||||
<span class="lineno">47</span>
|
||||
<span class="lineno">48</span> <span class="n">query</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_split_heads</span><span class="p">(</span><span class="n">query</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">num_heads</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">head_dim</span><span class="p">)</span>
|
||||
<span class="lineno">49</span> <span class="n">key</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_split_heads</span><span class="p">(</span><span class="n">key</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">num_heads</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">head_dim</span><span class="p">)</span>
|
||||
<span class="lineno">50</span> <span class="n">value</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_split_heads</span><span class="p">(</span><span class="n">value</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">num_heads</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">head_dim</span><span class="p">)</span>
|
||||
<span class="lineno">51</span>
|
||||
<span class="lineno">52</span> <span class="n">attn_output</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">functional</span><span class="o">.</span><span class="n">scaled_dot_product_attention</span><span class="p">(</span>
|
||||
<span class="lineno">53</span> <span class="n">query</span><span class="p">,</span>
|
||||
<span class="lineno">54</span> <span class="n">key</span><span class="p">,</span>
|
||||
<span class="lineno">55</span> <span class="n">value</span><span class="p">,</span>
|
||||
<span class="lineno">56</span> <span class="n">attn_mask</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
||||
<span class="lineno">57</span> <span class="n">dropout_p</span><span class="o">=</span><span class="mf">0.0</span><span class="p">,</span>
|
||||
<span class="lineno">58</span> <span class="n">is_causal</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># for the triangular mask</span>
|
||||
<span class="lineno">59</span> <span class="p">)</span>
|
||||
<span class="lineno">60</span>
|
||||
<span class="lineno">61</span> <span class="n">attn_output</span> <span class="o">=</span> <span class="n">attn_output</span><span class="o">.</span><span class="n">transpose</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">)</span><span class="o">.</span><span class="n">contiguous</span><span class="p">()</span>
|
||||
<span class="lineno">62</span> <span class="n">attn_output</span> <span class="o">=</span> <span class="n">attn_output</span><span class="o">.</span><span class="n">view</span><span class="p">(</span><span class="n">batch_size</span><span class="p">,</span> <span class="n">seq_length</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">embed_dim</span><span class="p">)</span>
|
||||
<span class="lineno">63</span>
|
||||
<span class="lineno">64</span> <span class="n">attn_output</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">c_proj</span><span class="p">(</span><span class="n">attn_output</span><span class="p">)</span>
|
||||
<span class="lineno">65</span>
|
||||
<span class="lineno">66</span> <span class="k">return</span> <span class="n">attn_output</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-13'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-13'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">69</span><span class="k">class</span> <span class="nc">Block</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-14'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-14'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">70</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">n_embed</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">layer_norm_epsilon</span><span class="p">:</span> <span class="nb">float</span><span class="p">,</span> <span class="n">r</span><span class="p">:</span> <span class="nb">int</span><span class="p">):</span>
|
||||
<span class="lineno">71</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
|
||||
<span class="lineno">72</span> <span class="bp">self</span><span class="o">.</span><span class="n">pre_norm</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">LayerNorm</span><span class="p">(</span><span class="n">n_embed</span><span class="p">,</span> <span class="n">eps</span><span class="o">=</span><span class="n">layer_norm_epsilon</span><span class="p">)</span>
|
||||
<span class="lineno">73</span> <span class="bp">self</span><span class="o">.</span><span class="n">attn</span> <span class="o">=</span> <span class="n">MultiHeadAttention</span><span class="p">(</span><span class="n">n_embed</span><span class="p">,</span> <span class="n">r</span><span class="p">)</span>
|
||||
<span class="lineno">74</span> <span class="bp">self</span><span class="o">.</span><span class="n">post_norm</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">LayerNorm</span><span class="p">(</span><span class="n">n_embed</span><span class="p">,</span> <span class="n">eps</span><span class="o">=</span><span class="n">layer_norm_epsilon</span><span class="p">)</span>
|
||||
<span class="lineno">75</span> <span class="bp">self</span><span class="o">.</span><span class="n">ffn</span> <span class="o">=</span> <span class="n">FFN</span><span class="p">(</span><span class="n">n_embed</span> <span class="o">*</span> <span class="mi">4</span><span class="p">,</span> <span class="n">n_embed</span><span class="p">,</span> <span class="n">r</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-15'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-15'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">77</span> <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">):</span>
|
||||
<span class="lineno">78</span> <span class="n">residual</span> <span class="o">=</span> <span class="n">hidden_states</span>
|
||||
<span class="lineno">79</span> <span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">pre_norm</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
|
||||
<span class="lineno">80</span>
|
||||
<span class="lineno">81</span> <span class="n">attn_output</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">attn</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
|
||||
<span class="lineno">82</span>
|
||||
<span class="lineno">83</span> <span class="n">hidden_states</span> <span class="o">=</span> <span class="n">attn_output</span> <span class="o">+</span> <span class="n">residual</span>
|
||||
<span class="lineno">84</span> <span class="n">residual</span> <span class="o">=</span> <span class="n">hidden_states</span>
|
||||
<span class="lineno">85</span> <span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">post_norm</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
|
||||
<span class="lineno">86</span> <span class="n">feed_forward_output</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">ffn</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
|
||||
<span class="lineno">87</span> <span class="n">hidden_states</span> <span class="o">=</span> <span class="n">feed_forward_output</span> <span class="o">+</span> <span class="n">residual</span>
|
||||
<span class="lineno">88</span>
|
||||
<span class="lineno">89</span> <span class="k">return</span> <span class="n">hidden_states</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-16'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-16'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">92</span><span class="k">class</span> <span class="nc">GPTModel</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-17'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-17'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">93</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">layer_norm_epsilon</span><span class="p">:</span> <span class="nb">float</span><span class="p">,</span> <span class="n">n_embd</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">n_layer</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">n_positions</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span>
|
||||
<span class="lineno">94</span> <span class="n">vocab_size</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">r</span><span class="p">:</span> <span class="nb">int</span><span class="p">):</span>
|
||||
<span class="lineno">95</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
|
||||
<span class="lineno">96</span>
|
||||
<span class="lineno">97</span> <span class="bp">self</span><span class="o">.</span><span class="n">token_embedding</span> <span class="o">=</span> <span class="n">Embedding</span><span class="p">(</span><span class="n">vocab_size</span><span class="p">,</span> <span class="n">n_embd</span><span class="p">,</span> <span class="n">r</span><span class="o">=</span><span class="n">r</span><span class="p">)</span>
|
||||
<span class="lineno">98</span> <span class="bp">self</span><span class="o">.</span><span class="n">position_embedding</span> <span class="o">=</span> <span class="n">Embedding</span><span class="p">(</span><span class="n">n_positions</span><span class="p">,</span> <span class="n">n_embd</span><span class="p">,</span> <span class="n">r</span><span class="o">=</span><span class="n">r</span><span class="p">)</span>
|
||||
<span class="lineno">99</span>
|
||||
<span class="lineno">100</span> <span class="bp">self</span><span class="o">.</span><span class="n">blocks</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">ModuleList</span><span class="p">([</span><span class="n">Block</span><span class="p">(</span><span class="n">n_embd</span><span class="p">,</span> <span class="n">layer_norm_epsilon</span><span class="p">,</span> <span class="n">r</span><span class="o">=</span><span class="n">r</span><span class="p">)</span>
|
||||
<span class="lineno">101</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">n_layer</span><span class="p">)])</span>
|
||||
<span class="lineno">102</span>
|
||||
<span class="lineno">103</span> <span class="bp">self</span><span class="o">.</span><span class="n">final_norm</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">LayerNorm</span><span class="p">(</span><span class="n">n_embd</span><span class="p">,</span> <span class="n">eps</span><span class="o">=</span><span class="n">layer_norm_epsilon</span><span class="p">)</span>
|
||||
<span class="lineno">104</span>
|
||||
<span class="lineno">105</span> <span class="bp">self</span><span class="o">.</span><span class="n">lm_head</span> <span class="o">=</span> <span class="n">Linear</span><span class="p">(</span><span class="n">n_embd</span><span class="p">,</span> <span class="n">vocab_size</span><span class="p">,</span> <span class="n">r</span><span class="o">=</span><span class="n">r</span><span class="p">,</span> <span class="n">bias</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-18'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-18'>#</a>
|
||||
</div>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">input_ids</span></code>
|
||||
has shape <code class="highlight"><span></span><span class="p">[</span><span class="n">batch_size</span><span class="p">,</span> <span class="n">seq_len</span><span class="p">]</span></code>
|
||||
</li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">107</span> <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">input_ids</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-19'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-19'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">111</span> <span class="n">batch_size</span><span class="p">,</span> <span class="n">seq_len</span> <span class="o">=</span> <span class="n">input_ids</span><span class="o">.</span><span class="n">shape</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-20'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-20'>#</a>
|
||||
</div>
|
||||
<p>Get token embeddings </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">114</span> <span class="n">token_embeddings</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">token_embedding</span><span class="p">(</span><span class="n">input_ids</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-21'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-21'>#</a>
|
||||
</div>
|
||||
<p>Get position ids </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">116</span> <span class="n">position_ids</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">arange</span><span class="p">(</span><span class="n">seq_len</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="n">input_ids</span><span class="o">.</span><span class="n">device</span><span class="p">)[</span><span class="kc">None</span><span class="p">,</span> <span class="p">:]</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-22'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-22'>#</a>
|
||||
</div>
|
||||
<p>Get position embeddings </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">118</span> <span class="n">position_embeddings</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">position_embedding</span><span class="p">(</span><span class="n">position_ids</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-23'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-23'>#</a>
|
||||
</div>
|
||||
<p>Add position embeddings </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">121</span> <span class="n">x</span> <span class="o">=</span> <span class="n">token_embeddings</span> <span class="o">+</span> <span class="n">position_embeddings</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-24'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-24'>#</a>
|
||||
</div>
|
||||
<p>Run through transformer blocks </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">124</span> <span class="k">for</span> <span class="n">block</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">blocks</span><span class="p">:</span>
|
||||
<span class="lineno">125</span> <span class="n">x</span> <span class="o">=</span> <span class="n">block</span><span class="p">(</span><span class="n">x</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-25'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-25'>#</a>
|
||||
</div>
|
||||
<p>Final normalization </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">128</span> <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">final_norm</span><span class="p">(</span><span class="n">x</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-26'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-26'>#</a>
|
||||
</div>
|
||||
<p>Get logits from projection layer </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">130</span> <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">lm_head</span><span class="p">(</span><span class="n">x</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='footer'>
|
||||
<a href="https://labml.ai">labml.ai</a>
|
||||
</div>
|
||||
</div>
|
||||
<script src=../interactive.js?v=1"></script>
|
||||
<script>
|
||||
function handleImages() {
|
||||
var images = document.querySelectorAll('p>img')
|
||||
|
||||
for (var i = 0; i < images.length; ++i) {
|
||||
handleImage(images[i])
|
||||
}
|
||||
}
|
||||
|
||||
function handleImage(img) {
|
||||
img.parentElement.style.textAlign = 'center'
|
||||
|
||||
var modal = document.createElement('div')
|
||||
modal.id = 'modal'
|
||||
|
||||
var modalContent = document.createElement('div')
|
||||
modal.appendChild(modalContent)
|
||||
|
||||
var modalImage = document.createElement('img')
|
||||
modalContent.appendChild(modalImage)
|
||||
|
||||
var span = document.createElement('span')
|
||||
span.classList.add('close')
|
||||
span.textContent = 'x'
|
||||
modal.appendChild(span)
|
||||
|
||||
img.onclick = function () {
|
||||
console.log('clicked')
|
||||
document.body.appendChild(modal)
|
||||
modalImage.src = img.src
|
||||
}
|
||||
|
||||
span.onclick = function () {
|
||||
document.body.removeChild(modal)
|
||||
}
|
||||
}
|
||||
|
||||
handleImages()
|
||||
</script>
|
||||
</body>
|
||||
</html>
|
||||
534
docs/zh/lora/index.html
Normal file
534
docs/zh/lora/index.html
Normal file
@ -0,0 +1,534 @@
|
||||
<!DOCTYPE html>
|
||||
<html lang="zh">
|
||||
<head>
|
||||
<meta http-equiv="content-type" content="text/html;charset=utf-8"/>
|
||||
<meta name="viewport" content="width=device-width, initial-scale=1.0"/>
|
||||
<meta name="description" content="Annotated implementation of RoRA from paper LoRA: Low-Rank Adaptation of Large Language Models"/>
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="Low-Rank Adaptation (LoRA)"/>
|
||||
<meta name="twitter:description" content="Annotated implementation of RoRA from paper LoRA: Low-Rank Adaptation of Large Language Models"/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/lora/index.html"/>
|
||||
<meta property="og:title" content="Low-Rank Adaptation (LoRA)"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="Low-Rank Adaptation (LoRA)"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="Low-Rank Adaptation (LoRA)"/>
|
||||
<meta property="og:description" content="Annotated implementation of RoRA from paper LoRA: Low-Rank Adaptation of Large Language Models"/>
|
||||
|
||||
<title>Low-Rank Adaptation (LoRA)</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/lora/index.html"/>
|
||||
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.13.18/dist/katex.min.css" integrity="sha384-zTROYFVGOfTw7JV7KUu8udsvW2fx4lWOsCEDqhBreBwlHI4ioVRtmIvEThzJHGET" crossorigin="anonymous">
|
||||
|
||||
<!-- Global site tag (gtag.js) - Google Analytics -->
|
||||
<script async src="https://www.googletagmanager.com/gtag/js?id=G-4V3HC8HBLH"></script>
|
||||
<script>
|
||||
window.dataLayer = window.dataLayer || [];
|
||||
|
||||
function gtag() {
|
||||
dataLayer.push(arguments);
|
||||
}
|
||||
|
||||
gtag('js', new Date());
|
||||
|
||||
gtag('config', 'G-4V3HC8HBLH');
|
||||
</script>
|
||||
</head>
|
||||
<body>
|
||||
<div id='container'>
|
||||
<div id="background"></div>
|
||||
<div class='section'>
|
||||
<div class='docs'>
|
||||
<p>
|
||||
<a class="parent" href="/">home</a>
|
||||
<a class="parent" href="index.html">lora</a>
|
||||
</p>
|
||||
<p>
|
||||
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations" target="_blank">
|
||||
<img alt="Github"
|
||||
src="https://img.shields.io/github/stars/labmlai/annotated_deep_learning_paper_implementations?style=social"
|
||||
style="max-width:100%;"/></a>
|
||||
<a href="https://twitter.com/labmlai" rel="nofollow" target="_blank">
|
||||
<img alt="Twitter"
|
||||
src="https://img.shields.io/twitter/follow/labmlai?style=social"
|
||||
style="max-width:100%;"/></a>
|
||||
</p>
|
||||
<p>
|
||||
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/lora/__init__.py" target="_blank">
|
||||
View code on Github</a>
|
||||
</p>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-0'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-0'>#</a>
|
||||
</div>
|
||||
<h1>Low-Rank Adaptation (LoRA)</h1>
|
||||
<p>This is an implementation of <a href="https://arxiv.org/abs/2106.09685">Low-Rank Adaptation (LoRA)</a> in <a href="https://pytorch.org">PyTorch</a>.</p>
|
||||
<p>Low-Rank Adaptation (LoRA) freezes pre-trained model weights and injects trainable rank decomposition matrices into each layer of the transformer. This makes it possible to efficiently fine-tune large langauge models by reducing trainable parameters by a large factor.</p>
|
||||
<p>Here's <a href="experiment.html">the training code</a> for training a GPT2 model with LoRA on Tiny Shakespeare dataset.</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">24</span><span></span><span class="kn">import</span> <span class="nn">torch</span>
|
||||
<span class="lineno">25</span><span class="kn">import</span> <span class="nn">torch.nn</span> <span class="k">as</span> <span class="nn">nn</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-1'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-1'>#</a>
|
||||
</div>
|
||||
<h2>LoRA Linear Layer</h2>
|
||||
<p>LoRA linear layer adds a low-rank decomposition to the pre-trained weight matrix (<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.9991079999999999em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqc" style=""><span class="mord" style=""><span class="mord coloredeq eqq" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">∈</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathbb" style="">R</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.8491079999999999em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">d</span><span class="mbin mtight" style="">×</span><span class="mord mathnormal mtight" style="margin-right:0.03148em">k</span></span></span></span></span></span></span></span></span></span></span></span></span></span>) of the linear layer.</p>
|
||||
<p><span ><span class="katex-display"><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.83333em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqf" style=""><span class="mord" style=""><span class="mord coloredeq eqq" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin" style="">+</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mord" style=""><span class="mord coloredeq eqo" style="">Δ</span><span class="mord mathnormal coloredeq eqo" style="margin-right:0.13889em">W</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord coloredeq eqq" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin" style="">+</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mord" style=""><span class="mord mathnormal coloredeq equ" style="margin-right:0.05017em">B</span></span><span class="mord" style=""><span class="mord mathnormal coloredeq eqt" style="">A</span></span></span></span></span></span></span></span></p>
|
||||
<p>, where <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8882079999999999em;vertical-align:-0.0391em;"></span><span class="mord coloredeq eqe" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq equ" style="margin-right:0.05017em">B</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">∈</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathbb" style="">R</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.8491079999999999em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">d</span><span class="mbin mtight" style="">×</span><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span></span></span></span></span></span></span></span></span></span></span>, <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8882079999999999em;vertical-align:-0.0391em;"></span><span class="mord coloredeq eqd" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqt" style="">A</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">∈</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathbb" style="">R</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.8491079999999999em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span><span class="mbin mtight" style="">×</span><span class="mord mathnormal mtight" style="margin-right:0.03148em">k</span></span></span></span></span></span></span></span></span></span></span></span></span></span>, and the rank <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.5782em;vertical-align:-0.0391em;"></span><span class="mord coloredeq eqv" style=""><span class="mord mathnormal" style="margin-right:0.02778em">r</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel">≪</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span></span><span class="base"><span class="strut" style="height:1em;vertical-align:-0.25em;"></span><span class="mord mathnormal">min</span><span class="mopen">(</span><span class="mord mathnormal">d</span><span class="mpunct">,</span><span class="mspace" style="margin-right:0.16666666666666666em;"></span><span class="mord mathnormal" style="margin-right:0.03148em;">k</span><span class="mclose">)</span></span></span></span></span>.</p>
|
||||
<p>All parameters are frozen except <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq eqt" style=""><span class="mord mathnormal" style="">A</span></span></span></span></span></span> and <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq equ" style=""><span class="mord mathnormal" style="margin-right:0.05017em">B</span></span></span></span></span></span>.</p>
|
||||
<p><span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq eqo" style=""><span class="mord" style="">Δ</span><span class="mord mathnormal" style="margin-right:0.13889em">W</span></span></span></span></span></span> is initialized to be zero at the beginning of the training.</p>
|
||||
<p>They multiple <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq eqm" style=""><span class="mord" style=""><span class="mord coloredeq eqo" style="">Δ</span><span class="mord mathnormal coloredeq eqo" style="margin-right:0.13889em">W</span></span><span class="mord mathnormal" style="">x</span></span></span></span></span></span> by <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.040392em;vertical-align:-0.345em;"></span><span class="mord coloredeq eqi" style=""><span class="mord" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.695392em;"><span style="top:-2.6550000000000002em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.394em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqp" style="margin-right:0.0037em">α</span></span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.345em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span></span></span></span></span></span> where <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqp" style=""><span class="mord mathnormal" style="margin-right:0.0037em">α</span></span></span></span></span></span> is a hyper-parameter. Once <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqp" style=""><span class="mord mathnormal" style="margin-right:0.0037em">α</span></span></span></span></span></span> is tuned it can be kept the same when varying <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqv" style=""><span class="mord mathnormal" style="margin-right:0.02778em">r</span></span></span></span></span></span>.</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">28</span><span class="k">class</span> <span class="nc">Linear</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-2'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-2'>#</a>
|
||||
</div>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">in_features</span></code>
|
||||
is the number of input features of the linear layer </li>
|
||||
<li><code class="highlight"><span></span><span class="n">out_features</span></code>
|
||||
is the number of output features of the linear layer </li>
|
||||
<li><code class="highlight"><span></span><span class="n">bias</span></code>
|
||||
is a flag indicating if there is a bias parameter </li>
|
||||
<li><code class="highlight"><span></span><span class="n">r</span></code>
|
||||
is the rank of the decomposition <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqv" style=""><span class="mord mathnormal" style="margin-right:0.02778em">r</span></span></span></span></span></span> </li>
|
||||
<li><code class="highlight"><span></span><span class="n">alpha</span></code>
|
||||
is the scaling factor <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqp" style=""><span class="mord mathnormal" style="margin-right:0.0037em">α</span></span></span></span></span></span></li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">49</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">in_features</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">out_features</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">bias</span><span class="p">:</span> <span class="nb">bool</span><span class="p">,</span>
|
||||
<span class="lineno">50</span> <span class="n">r</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">alpha</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-3'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-3'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">58</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-4'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-4'>#</a>
|
||||
</div>
|
||||
<p>Set <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqn" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqp" style="margin-right:0.0037em">α</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathnormal coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span></span></span> is not provided. i.e. make the scaling factor <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.040392em;vertical-align:-0.345em;"></span><span class="mord coloredeq eqh" style=""><span class="mord" style=""><span class="mord coloredeq eqi" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.695392em;"><span style="top:-2.6550000000000002em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.394em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqp" style="margin-right:0.0037em">α</span></span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.345em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style="">1</span></span></span></span></span></span>. </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">61</span> <span class="k">if</span> <span class="n">alpha</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
||||
<span class="lineno">62</span> <span class="n">alpha</span> <span class="o">=</span> <span class="n">r</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-5'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-5'>#</a>
|
||||
</div>
|
||||
<p>The pre-trained weight <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.83333em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqq" style=""><span class="mord" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span></span> </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">65</span> <span class="bp">self</span><span class="o">.</span><span class="n">weight</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">empty</span><span class="p">((</span><span class="n">out_features</span><span class="p">,</span> <span class="n">in_features</span><span class="p">)))</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-6'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-6'>#</a>
|
||||
</div>
|
||||
<p>Freeze it </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">67</span> <span class="bp">self</span><span class="o">.</span><span class="n">weight</span><span class="o">.</span><span class="n">requires_grad</span> <span class="o">=</span> <span class="kc">False</span>
|
||||
<span class="lineno">68</span>
|
||||
<span class="lineno">69</span> <span class="k">if</span> <span class="n">bias</span><span class="p">:</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-7'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-7'>#</a>
|
||||
</div>
|
||||
<p>Bias parameter <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.84444em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqr" style=""><span class="mord" style=""><span class="mord mathnormal" style="">b</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:0em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span></span> (also frozen) </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">71</span> <span class="bp">self</span><span class="o">.</span><span class="n">bias</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">empty</span><span class="p">(</span><span class="n">out_features</span><span class="p">))</span>
|
||||
<span class="lineno">72</span> <span class="bp">self</span><span class="o">.</span><span class="n">bias</span><span class="o">.</span><span class="n">requires_grad</span> <span class="o">=</span> <span class="kc">False</span>
|
||||
<span class="lineno">73</span> <span class="k">else</span><span class="p">:</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-8'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-8'>#</a>
|
||||
</div>
|
||||
<p>No bias parameter </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">75</span> <span class="bp">self</span><span class="o">.</span><span class="n">bias</span> <span class="o">=</span> <span class="kc">None</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-9'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-9'>#</a>
|
||||
</div>
|
||||
<p>scaling factor <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.040392em;vertical-align:-0.345em;"></span><span class="mord coloredeq eqi" style=""><span class="mord" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.695392em;"><span style="top:-2.6550000000000002em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.394em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqp" style="margin-right:0.0037em">α</span></span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.345em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span></span></span></span></span></span> </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">78</span> <span class="bp">self</span><span class="o">.</span><span class="n">scaling</span> <span class="o">=</span> <span class="n">alpha</span> <span class="o">/</span> <span class="n">r</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-10'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-10'>#</a>
|
||||
</div>
|
||||
<p>Matrix <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8882079999999999em;vertical-align:-0.0391em;"></span><span class="mord coloredeq eqd" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqt" style="">A</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">∈</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathbb" style="">R</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.8491079999999999em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span><span class="mbin mtight" style="">×</span><span class="mord mathnormal mtight" style="margin-right:0.03148em">k</span></span></span></span></span></span></span></span></span></span></span></span></span></span> </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">80</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_a</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">empty</span><span class="p">((</span><span class="n">in_features</span><span class="p">,</span> <span class="n">r</span><span class="p">)))</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-11'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-11'>#</a>
|
||||
</div>
|
||||
<p>Matrix <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8882079999999999em;vertical-align:-0.0391em;"></span><span class="mord coloredeq eqe" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq equ" style="margin-right:0.05017em">B</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">∈</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathbb" style="">R</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.8491079999999999em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">d</span><span class="mbin mtight" style="">×</span><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span></span></span></span></span></span></span></span></span></span></span>, we keep <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq eqt" style=""><span class="mord mathnormal" style="">A</span></span></span></span></span></span> and <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq equ" style=""><span class="mord mathnormal" style="margin-right:0.05017em">B</span></span></span></span></span></span> transposed </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">82</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_b</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">empty</span><span class="p">((</span><span class="n">r</span><span class="p">,</span> <span class="n">out_features</span><span class="p">)))</span>
|
||||
<span class="lineno">83</span>
|
||||
<span class="lineno">84</span> <span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">():</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-12'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-12'>#</a>
|
||||
</div>
|
||||
<p>Initialize <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq eqt" style=""><span class="mord mathnormal" style="">A</span></span></span></span></span></span> similar to a weight matrix in a normal linear layer </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">86</span> <span class="n">nn</span><span class="o">.</span><span class="n">init</span><span class="o">.</span><span class="n">kaiming_uniform_</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">lora_a</span><span class="p">,</span> <span class="n">a</span><span class="o">=</span><span class="mi">5</span> <span class="o">**</span> <span class="mf">0.5</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-13'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-13'>#</a>
|
||||
</div>
|
||||
<p>Initialize <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq equ" style=""><span class="mord mathnormal" style="margin-right:0.05017em">B</span></span></span></span></span></span> to <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.64444em;vertical-align:0em;"></span><span class="mord coloredeq eqs" style=""><span class="mord" style="">0</span></span></span></span></span></span> so that <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq eqk" style=""><span class="mord" style=""><span class="mord coloredeq eqo" style="">Δ</span><span class="mord mathnormal coloredeq eqo" style="margin-right:0.13889em">W</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathnormal coloredeq equ" style="margin-right:0.05017em">B</span></span><span class="mord" style=""><span class="mord mathnormal coloredeq eqt" style="">A</span></span></span></span></span></span></span> is <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.64444em;vertical-align:0em;"></span><span class="mord coloredeq eqs" style=""><span class="mord" style="">0</span></span></span></span></span></span> at initialization </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">88</span> <span class="n">nn</span><span class="o">.</span><span class="n">init</span><span class="o">.</span><span class="n">zeros_</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">lora_b</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-14'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-14'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">90</span> <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-15'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-15'>#</a>
|
||||
</div>
|
||||
<p>Compute <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.83333em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqq" style=""><span class="mord" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span><span class="mord mathnormal">x</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin">+</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span></span><span class="base"><span class="strut" style="height:0.84444em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqr" style=""><span class="mord" style=""><span class="mord mathnormal" style="">b</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:0em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span></span> </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">92</span> <span class="n">result</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">functional</span><span class="o">.</span><span class="n">linear</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">weight</span><span class="p">,</span> <span class="n">bias</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">bias</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-16'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-16'>#</a>
|
||||
</div>
|
||||
<p>Add <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.040392em;vertical-align:-0.345em;"></span><span class="mord coloredeq eqi" style=""><span class="mord" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.695392em;"><span style="top:-2.6550000000000002em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.394em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqp" style="margin-right:0.0037em">α</span></span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.345em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span></span><span class="mord coloredeq eqm" style=""><span class="mord" style=""><span class="mord coloredeq eqo" style="">Δ</span><span class="mord mathnormal coloredeq eqo" style="margin-right:0.13889em">W</span></span><span class="mord mathnormal" style="">x</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span></span><span class="base"><span class="strut" style="height:1.040392em;vertical-align:-0.345em;"></span><span class="mord coloredeq eqi" style=""><span class="mord" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.695392em;"><span style="top:-2.6550000000000002em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.394em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqp" style="margin-right:0.0037em">α</span></span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.345em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span></span><span class="mord coloredeq equ" style=""><span class="mord mathnormal" style="margin-right:0.05017em">B</span></span><span class="mord coloredeq eqt" style=""><span class="mord mathnormal" style="">A</span></span><span class="mord mathnormal">x</span></span></span></span></span> </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">95</span> <span class="n">result</span> <span class="o">+=</span> <span class="p">(</span><span class="n">x</span> <span class="o">@</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_a</span> <span class="o">@</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_b</span><span class="p">)</span> <span class="o">*</span> <span class="bp">self</span><span class="o">.</span><span class="n">scaling</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-17'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-17'>#</a>
|
||||
</div>
|
||||
<p> </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">98</span> <span class="k">return</span> <span class="n">result</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-18'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-18'>#</a>
|
||||
</div>
|
||||
<h2>LoRA Embedding Layer</h2>
|
||||
<p>Similar to LoRA linear layer this adds a low-rank decomposition to the pre-trained embedding weights matrix (<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.9991079999999999em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqc" style=""><span class="mord" style=""><span class="mord coloredeq eqq" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">∈</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathbb" style="">R</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.8491079999999999em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">d</span><span class="mbin mtight" style="">×</span><span class="mord mathnormal mtight" style="margin-right:0.03148em">k</span></span></span></span></span></span></span></span></span></span></span></span></span></span>).</p>
|
||||
<p><span ><span class="katex-display"><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.83333em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqf" style=""><span class="mord" style=""><span class="mord coloredeq eqq" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin" style="">+</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mord" style=""><span class="mord coloredeq eqo" style="">Δ</span><span class="mord mathnormal coloredeq eqo" style="margin-right:0.13889em">W</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord coloredeq eqq" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin" style="">+</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mord" style=""><span class="mord mathnormal coloredeq equ" style="margin-right:0.05017em">B</span></span><span class="mord" style=""><span class="mord mathnormal coloredeq eqt" style="">A</span></span></span></span></span></span></span></span></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">101</span><span class="k">class</span> <span class="nc">Embedding</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-19'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-19'>#</a>
|
||||
</div>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">num_embeddings</span></code>
|
||||
is the number of embeddings </li>
|
||||
<li><code class="highlight"><span></span><span class="n">embedding_dim</span></code>
|
||||
is the number embedding dimensions </li>
|
||||
<li><code class="highlight"><span></span><span class="n">r</span></code>
|
||||
is the rank of the decomposition <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqv" style=""><span class="mord mathnormal" style="margin-right:0.02778em">r</span></span></span></span></span></span> </li>
|
||||
<li><code class="highlight"><span></span><span class="n">alpha</span></code>
|
||||
is the scaling factor <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqp" style=""><span class="mord mathnormal" style="margin-right:0.0037em">α</span></span></span></span></span></span></li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">111</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">num_embeddings</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">embedding_dim</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span>
|
||||
<span class="lineno">112</span> <span class="n">r</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">alpha</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-20'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-20'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">120</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-21'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-21'>#</a>
|
||||
</div>
|
||||
<p>Set <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqn" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqp" style="margin-right:0.0037em">α</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathnormal coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span></span></span> is not provided. i.e. make the scaling factor <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.040392em;vertical-align:-0.345em;"></span><span class="mord coloredeq eqh" style=""><span class="mord" style=""><span class="mord coloredeq eqi" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.695392em;"><span style="top:-2.6550000000000002em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.394em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqp" style="margin-right:0.0037em">α</span></span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.345em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style="">1</span></span></span></span></span></span>. </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">123</span> <span class="k">if</span> <span class="n">alpha</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
||||
<span class="lineno">124</span> <span class="n">alpha</span> <span class="o">=</span> <span class="n">r</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-22'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-22'>#</a>
|
||||
</div>
|
||||
<p>The pre-trained embedding weights <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.83333em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqq" style=""><span class="mord" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span></span> (frozen) </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">127</span> <span class="bp">self</span><span class="o">.</span><span class="n">weight</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">empty</span><span class="p">((</span><span class="n">num_embeddings</span><span class="p">,</span> <span class="n">embedding_dim</span><span class="p">)))</span>
|
||||
<span class="lineno">128</span> <span class="bp">self</span><span class="o">.</span><span class="n">weight</span><span class="o">.</span><span class="n">requires_grad</span> <span class="o">=</span> <span class="kc">False</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-23'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-23'>#</a>
|
||||
</div>
|
||||
<p>scaling factor <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.040392em;vertical-align:-0.345em;"></span><span class="mord coloredeq eqi" style=""><span class="mord" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.695392em;"><span style="top:-2.6550000000000002em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.394em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqp" style="margin-right:0.0037em">α</span></span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.345em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span></span></span></span></span></span> </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">131</span> <span class="bp">self</span><span class="o">.</span><span class="n">scaling</span> <span class="o">=</span> <span class="n">alpha</span> <span class="o">/</span> <span class="n">r</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-24'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-24'>#</a>
|
||||
</div>
|
||||
<p>Matrix <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8882079999999999em;vertical-align:-0.0391em;"></span><span class="mord coloredeq eqd" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqt" style="">A</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">∈</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathbb" style="">R</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.8491079999999999em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span><span class="mbin mtight" style="">×</span><span class="mord mathnormal mtight" style="margin-right:0.03148em">k</span></span></span></span></span></span></span></span></span></span></span></span></span></span> </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">133</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_a</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">empty</span><span class="p">((</span><span class="n">num_embeddings</span><span class="p">,</span> <span class="n">r</span><span class="p">)))</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-25'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-25'>#</a>
|
||||
</div>
|
||||
<p>Matrix <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8882079999999999em;vertical-align:-0.0391em;"></span><span class="mord coloredeq eqe" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq equ" style="margin-right:0.05017em">B</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">∈</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathbb" style="">R</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.8491079999999999em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">d</span><span class="mbin mtight" style="">×</span><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span></span></span></span></span></span></span></span></span></span></span> </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">135</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_b</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">empty</span><span class="p">((</span><span class="n">r</span><span class="p">,</span> <span class="n">embedding_dim</span><span class="p">)))</span>
|
||||
<span class="lineno">136</span>
|
||||
<span class="lineno">137</span> <span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">():</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-26'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-26'>#</a>
|
||||
</div>
|
||||
<p>Initialize <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq eqt" style=""><span class="mord mathnormal" style="">A</span></span></span></span></span></span> with a normal distribution </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">139</span> <span class="n">nn</span><span class="o">.</span><span class="n">init</span><span class="o">.</span><span class="n">normal_</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">lora_a</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-27'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-27'>#</a>
|
||||
</div>
|
||||
<p>Initialize <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq equ" style=""><span class="mord mathnormal" style="margin-right:0.05017em">B</span></span></span></span></span></span> to <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.64444em;vertical-align:0em;"></span><span class="mord coloredeq eqs" style=""><span class="mord" style="">0</span></span></span></span></span></span> so that <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq eqk" style=""><span class="mord" style=""><span class="mord coloredeq eqo" style="">Δ</span><span class="mord mathnormal coloredeq eqo" style="margin-right:0.13889em">W</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathnormal coloredeq equ" style="margin-right:0.05017em">B</span></span><span class="mord" style=""><span class="mord mathnormal coloredeq eqt" style="">A</span></span></span></span></span></span></span> is <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.64444em;vertical-align:0em;"></span><span class="mord coloredeq eqs" style=""><span class="mord" style="">0</span></span></span></span></span></span> at initialization </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">141</span> <span class="n">nn</span><span class="o">.</span><span class="n">init</span><span class="o">.</span><span class="n">zeros_</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">lora_b</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-28'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-28'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">143</span> <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-29'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-29'>#</a>
|
||||
</div>
|
||||
<p>Compute the embeddings <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1em;vertical-align:-0.25em;"></span><span class="mord coloredeq eqq" style=""><span class="mord" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span><span class="mord text"><span class="mord">onehot</span></span><span class="mopen">(</span><span class="mord mathnormal">x</span><span class="mclose">)</span></span></span></span></span> </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">145</span> <span class="n">result</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">functional</span><span class="o">.</span><span class="n">embedding</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">weight</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-30'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-30'>#</a>
|
||||
</div>
|
||||
<p>Add <span ><strong style="">Error</strong></span> </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">148</span> <span class="n">result</span> <span class="o">+=</span> <span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">functional</span><span class="o">.</span><span class="n">embedding</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_a</span><span class="p">)</span> <span class="o">@</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_b</span><span class="p">)</span> <span class="o">*</span> <span class="bp">self</span><span class="o">.</span><span class="n">scaling</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-31'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-31'>#</a>
|
||||
</div>
|
||||
<p> </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">151</span> <span class="k">return</span> <span class="n">result</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='footer'>
|
||||
<a href="https://labml.ai">labml.ai</a>
|
||||
</div>
|
||||
</div>
|
||||
<script src=../interactive.js?v=1"></script>
|
||||
<script>
|
||||
function handleImages() {
|
||||
var images = document.querySelectorAll('p>img')
|
||||
|
||||
for (var i = 0; i < images.length; ++i) {
|
||||
handleImage(images[i])
|
||||
}
|
||||
}
|
||||
|
||||
function handleImage(img) {
|
||||
img.parentElement.style.textAlign = 'center'
|
||||
|
||||
var modal = document.createElement('div')
|
||||
modal.id = 'modal'
|
||||
|
||||
var modalContent = document.createElement('div')
|
||||
modal.appendChild(modalContent)
|
||||
|
||||
var modalImage = document.createElement('img')
|
||||
modalContent.appendChild(modalImage)
|
||||
|
||||
var span = document.createElement('span')
|
||||
span.classList.add('close')
|
||||
span.textContent = 'x'
|
||||
modal.appendChild(span)
|
||||
|
||||
img.onclick = function () {
|
||||
console.log('clicked')
|
||||
document.body.appendChild(modal)
|
||||
modalImage.src = img.src
|
||||
}
|
||||
|
||||
span.onclick = function () {
|
||||
document.body.removeChild(modal)
|
||||
}
|
||||
}
|
||||
|
||||
handleImages()
|
||||
</script>
|
||||
</body>
|
||||
</html>
|
||||
@ -8,7 +8,7 @@
|
||||
|
||||
<url>
|
||||
<loc>https://nn.labml.ai/gan/wasserstein/index.html</loc>
|
||||
<lastmod>2023-10-24T16:30:00+00:00</lastmod>
|
||||
<lastmod>2024-07-15T16:30:00+00:00</lastmod>
|
||||
<priority>1.00</priority>
|
||||
</url>
|
||||
|
||||
@ -504,22 +504,22 @@
|
||||
|
||||
|
||||
<url>
|
||||
<loc>https://nn.labml.ai/RWKV/configs.html</loc>
|
||||
<lastmod>2024-03-17T16:30:00+00:00</lastmod>
|
||||
<loc>https://nn.labml.ai/rwkv/configs.html</loc>
|
||||
<lastmod>2024-08-02T16:30:00+00:00</lastmod>
|
||||
<priority>1.00</priority>
|
||||
</url>
|
||||
|
||||
|
||||
<url>
|
||||
<loc>https://nn.labml.ai/RWKV/index.html</loc>
|
||||
<lastmod>2024-03-17T16:30:00+00:00</lastmod>
|
||||
<loc>https://nn.labml.ai/rwkv/index.html</loc>
|
||||
<lastmod>2024-08-02T16:30:00+00:00</lastmod>
|
||||
<priority>1.00</priority>
|
||||
</url>
|
||||
|
||||
|
||||
<url>
|
||||
<loc>https://nn.labml.ai/RWKV/experiment.html</loc>
|
||||
<lastmod>2024-03-17T16:30:00+00:00</lastmod>
|
||||
<loc>https://nn.labml.ai/rwkv/experiment.html</loc>
|
||||
<lastmod>2024-08-02T16:30:00+00:00</lastmod>
|
||||
<priority>1.00</priority>
|
||||
</url>
|
||||
|
||||
@ -1294,6 +1294,27 @@
|
||||
</url>
|
||||
|
||||
|
||||
<url>
|
||||
<loc>https://nn.labml.ai/lora/gpt2.html</loc>
|
||||
<lastmod>2024-08-16T16:30:00+00:00</lastmod>
|
||||
<priority>1.00</priority>
|
||||
</url>
|
||||
|
||||
|
||||
<url>
|
||||
<loc>https://nn.labml.ai/lora/index.html</loc>
|
||||
<lastmod>2024-08-03T16:30:00+00:00</lastmod>
|
||||
<priority>1.00</priority>
|
||||
</url>
|
||||
|
||||
|
||||
<url>
|
||||
<loc>https://nn.labml.ai/lora/experiment.html</loc>
|
||||
<lastmod>2024-08-16T16:30:00+00:00</lastmod>
|
||||
<priority>1.00</priority>
|
||||
</url>
|
||||
|
||||
|
||||
<url>
|
||||
<loc>https://nn.labml.ai/graphs/gat/index.html</loc>
|
||||
<lastmod>2023-10-24T16:30:00+00:00</lastmod>
|
||||
|
||||
@ -3,24 +3,24 @@
|
||||
<head>
|
||||
<meta http-equiv="content-type" content="text/html;charset=utf-8"/>
|
||||
<meta name="viewport" content="width=device-width, initial-scale=1.0"/>
|
||||
<meta name="description" content="这是 PyTorch 中AFT（免注意变压器）的带注释的实现/教程。"/>
|
||||
<meta name="description" content="这是一个 PyTorch实现的 AFT （无注意力 Transformer ）带注释实现/教程。"/>
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="免注意的变压器"/>
|
||||
<meta name="twitter:description" content="这是 PyTorch 中AFT（免注意变压器）的带注释的实现/教程。"/>
|
||||
<meta name="twitter:title" content="一种无注意力的 Transformer"/>
|
||||
<meta name="twitter:description" content="这是一个 PyTorch实现的 AFT （无注意力 Transformer ）带注释实现/教程。"/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/transformers/aft/index.html"/>
|
||||
<meta property="og:title" content="免注意的变压器"/>
|
||||
<meta property="og:title" content="一种无注意力的 Transformer"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="免注意的变压器"/>
|
||||
<meta property="og:site_name" content="一种无注意力的 Transformer"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="免注意的变压器"/>
|
||||
<meta property="og:description" content="这是 PyTorch 中AFT（免注意变压器）的带注释的实现/教程。"/>
|
||||
<meta property="og:title" content="一种无注意力的 Transformer"/>
|
||||
<meta property="og:description" content="这是一个 PyTorch实现的 AFT （无注意力 Transformer ）带注释实现/教程。"/>
|
||||
|
||||
<title>免注意的变压器</title>
|
||||
<title>一种无注意力的 Transformer</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/transformers/aft/index.html"/>
|
||||
|
||||
@ -7,20 +7,20 @@
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="免注意的变压器"/>
|
||||
<meta name="twitter:title" content="一种无注意力的 Transformer"/>
|
||||
<meta name="twitter:description" content=""/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/transformers/aft/readme.html"/>
|
||||
<meta property="og:title" content="免注意的变压器"/>
|
||||
<meta property="og:title" content="一种无注意力的 Transformer"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="免注意的变压器"/>
|
||||
<meta property="og:site_name" content="一种无注意力的 Transformer"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="免注意的变压器"/>
|
||||
<meta property="og:title" content="一种无注意力的 Transformer"/>
|
||||
<meta property="og:description" content=""/>
|
||||
|
||||
<title>免注意的变压器</title>
|
||||
<title>一种无注意力的 Transformer</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/transformers/aft/readme.html"/>
|
||||
@ -71,10 +71,10 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-0'>#</a>
|
||||
</div>
|
||||
<h1><a href="https://nn.labml.ai/transformers/aft/index.html">一款无注意的变形金刚</a></h1>
|
||||
<p>这是 <a href="https://pytorch.org">PyTorch 对</a>《<a href="https://arxiv.org/abs/2105.14103">无注意力的变形金刚》一文的</a>实现。</p>
|
||||
<p>本文用一种新的高效运算取代了<a href="https://nn.labml.ai/transformers/mha.html">自我注意力层</a>，该运算的存储复杂度为O（Td），其中 T 是序列长度，<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.69444em;vertical-align:0em;"></span><span class="mord mathnormal">d</span></span></span></span></span>是嵌入的维度。</p>
|
||||
<p>本文介绍了 AFT 以及 AFT-Local 和 AFT-conv。这里我们实现了 aft-Local，它关注自回归模型中的 cloby 代币。</p>
|
||||
<h1><a href="https://nn.labml.ai/transformers/aft/index.html">一种无注意力的 Transformer </a></h1>
|
||||
<p>这是论文 <a href="https://arxiv.org/abs/2105.14103">《一种无注意力的 Transformer 》</a>的<a href="https://pytorch.org">PyTorch </a>实现。</p>
|
||||
<p>这篇论文用一种新的高效操作替代了<a href="https://nn.labml.ai/transformers/mha.html">自注意力层</a>，该运算的存储复杂度为O（Td），其中 T 是序列长度，<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.69444em;vertical-align:0em;"></span><span class="mord mathnormal">d</span></span></span></span></span>是嵌入的维度。</p>
|
||||
<p>该论文介绍了 AFT 以及 AFT-local 和 AFT-conv 。这里我们实现了 AFT-local ，它会在自回归模型中关注邻近的 token 。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
|
||||
@ -308,7 +308,7 @@
|
||||
<a href='#section-18'>#</a>
|
||||
</div>
|
||||
<h2>GLU 变体</h2>
|
||||
<p>这些是在论文 <a href="https://arxiv.org/abs/2002.05202">《 GLU Variants Improve Transformer 》</a>中包含的各种带门控隐藏层的 ffn 变体。我们已按照论文规定省略了偏置项。</p>
|
||||
<p>这些是在论文 <a href="https://arxiv.org/abs/2002.05202">《 GLU Variants Improve Transformer 》</a>中包含的各种带门控隐藏层的 FFN 变体。我们已按照论文规定省略了偏置项。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
|
||||
@ -183,7 +183,7 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-7'>#</a>
|
||||
</div>
|
||||
<p>输出（预测）</p>
|
||||
<p>输出预测值</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
|
||||
@ -72,7 +72,7 @@
|
||||
</div>
|
||||
<h1>多头注意力 (MHA)</h1>
|
||||
<p><a href="https://colab.research.google.com/github/labmlai/annotated_deep_learning_paper_implementations/blob/master/labml_nn/transformers/basic/autoregressive_experiment.ipynb"><img alt="Open In Colab" src="https://colab.research.google.com/assets/colab-badge.svg"></a></p>
|
||||
</a><p>这是论文<a href="https://arxiv.org/abs/1706.03762">《 Attention is All You Need 》</a>中多头注意力的<a href="https://pytorch.org/">PyTorch</a>教程/实现。该实现的灵感来自<a href="https://nlp.seas.harvard.edu/2018/04/03/attention.html">《带注释的变形金刚》</a>。</p>%n<p>这是使用基础 Transformer 和 MHA 进行 NLP 自回归的<a href="basic/autoregressive_experiment.html">训练代码</a>。</p>%n<p>这是一个训练简单transformer的<a href="basic/autoregressive_experiment.html">代码实现</a>。</p>
|
||||
</a><p>这是论文<a href="https://arxiv.org/abs/1706.03762">《 Attention is All You Need 》</a>中多头注意力的<a href="https://pytorch.org/">PyTorch</a>教程/实现。该实现的灵感来自<a href="https://nlp.seas.harvard.edu/2018/04/03/attention.html">《带注释的 Transformer 》</a>。</p><p>这是使用基础 Transformer 和 MHA 进行 NLP 自回归的<a href="basic/autoregressive_experiment.html">训练代码</a>。</p><p>这是一个训练简单 Transformer 的<a href="basic/autoregressive_experiment.html">代码实现</a>。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
@ -116,7 +116,7 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-3'>#</a>
|
||||
</div>
|
||||
<p>线性层用于线性变换/p>
|
||||
<p>线性层用于线性变换</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
@ -234,7 +234,7 @@ s-225.272,467,-225.272,467s-235,486,-235,486c-2.7,4.7,-9,7,-19,7
|
||||
c-6,0,-10,-1,-12,-3s-194,-422,-194,-422s-65,47,-65,47z
|
||||
M834 80h400000v40h-400000z"></path></svg></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.18278000000000005em;"><span></span></span></span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.677em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqg" style="">Q</span><span class="mord coloredeq eqg" style=""><span class="mord mathnormal" style="margin-right:0.07153em">K</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.849108em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style="">⊤</span></span></span></span></span></span></span></span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.93em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span></span><span class="mord"><span class="delimsizing size4">)</span></span><span class="mord mathnormal" style="margin-right:0.22222em;">V</span></span></span></span></span></span></p>
|
||||
<p>简单来说，它会找到与查询 (Query) 匹配的键 (key)，并获取这些键 (Key) 的值 (Value) 。</p>
|
||||
<p>它使用查询和键的点积作为衡量它们之间匹配程度的指标。在进行<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8888799999999999em;vertical-align:-0.19444em;"></span><span class="mord coloredeq eqh" style=""><span class="mord mathnormal" style="">so</span><span class="mord mathnormal" style="margin-right:0.10764em">f</span><span class="mord mathnormal" style="">t</span><span class="mord mathnormal" style="">ma</span><span class="mord mathnormal" style="">x</span></span></span></span></span></span>之前，点积会被<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.383108em;vertical-align:-0.538em;"></span><span class="mord"><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.845108em;"><span style="top:-2.5864385em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord sqrt mtight"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.8622307142857143em;"><span class="svg-align" style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="mord mtight" style="padding-left:0.833em;"><span class="mord mtight coloredeq eqi" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">d</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.3448em;"><span style="top:-2.3487714285714287em;margin-left:0em;margin-right:0.07142857142857144em;"><span class="pstrut" style="height:2.5em;"></span><span class="sizing reset-size3 size1 mtight" style=""><span class="mord mathnormal mtight" style="margin-right:0.03148em">k</span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15122857142857138em;"><span></span></span></span></span></span></span></span></span></span><span style="top:-2.8222307142857144em;"><span class="pstrut" style="height:3em;"></span><span class="hide-tail mtight" style="min-width:0.853em;height:1.08em;"><svg height="1.08em" preserveaspectratio="xMinYMin slice" viewbox="0 0 400000 1080" width="400em" xmlns="http://www.w3.org/2000/svg"><path d="M95,702
|
||||
<p>它使用查询和键的点积作为衡量它们之间匹配程度的指标。在进行<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8888799999999999em;vertical-align:-0.19444em;"></span><span class="mord coloredeq eqh" style=""><span class="mord mathnormal" style="">so</span><span class="mord mathnormal" style="margin-right:0.10764em">f</span><span class="mord mathnormal" style="">t</span><span class="mord mathnormal" style="">ma</span><span class="mord mathnormal" style="">x</span></span></span></span></span></span>之前，点积会乘以<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.383108em;vertical-align:-0.538em;"></span><span class="mord"><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.845108em;"><span style="top:-2.5864385em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord sqrt mtight"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.8622307142857143em;"><span class="svg-align" style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="mord mtight" style="padding-left:0.833em;"><span class="mord mtight coloredeq eqi" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">d</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.3448em;"><span style="top:-2.3487714285714287em;margin-left:0em;margin-right:0.07142857142857144em;"><span class="pstrut" style="height:2.5em;"></span><span class="sizing reset-size3 size1 mtight" style=""><span class="mord mathnormal mtight" style="margin-right:0.03148em">k</span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15122857142857138em;"><span></span></span></span></span></span></span></span></span></span><span style="top:-2.8222307142857144em;"><span class="pstrut" style="height:3em;"></span><span class="hide-tail mtight" style="min-width:0.853em;height:1.08em;"><svg height="1.08em" preserveaspectratio="xMinYMin slice" viewbox="0 0 400000 1080" width="400em" xmlns="http://www.w3.org/2000/svg"><path d="M95,702
|
||||
c-2.7,0,-7.17,-2.7,-13.5,-8c-5.8,-5.3,-9.5,-10,-9.5,-14
|
||||
c0,-2,0.3,-3.3,1,-4c1.3,-2.7,23.83,-20.7,67.5,-54
|
||||
c44.2,-33.3,65.8,-50.3,66.5,-51c1.3,-1.3,3,-2,5,-2c4.7,0,8.7,3.3,12,10
|
||||
@ -467,7 +467,7 @@ M834 80h400000v40h-400000z"></path></svg></span></span></span><span class="vlist
|
||||
<a href='#section-28'>#</a>
|
||||
</div>
|
||||
<p><code class="highlight"><span></span><span class="n">query</span></code>
|
||||
<code class="highlight"><span></span><span class="n">key</span></code>
|
||||
、<code class="highlight"><span></span><span class="n">key</span></code>
|
||||
和<code class="highlight"><span></span><span class="n">value</span></code>
|
||||
是存储<em>查询</em>、<em>键</em>和<em>值</em>向量集合的张量。它们的形状为<code class="highlight"><span></span><span class="p">[</span><span class="n">seq_len</span><span class="p">,</span> <span class="n">batch_size</span><span class="p">,</span> <span class="n">d_model</span><span class="p">]</span></code>
|
||||
。</p>
|
||||
@ -530,7 +530,7 @@ M834 80h400000v40h-400000z"></path></svg></span></span></span><span class="vlist
|
||||
<div class='section-link'>
|
||||
<a href='#section-31'>#</a>
|
||||
</div>
|
||||
<p>计算注意力分数<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.043548em;vertical-align:-0.19444em;"></span><span class="mord coloredeq eqg" style=""><span class="mord mathnormal" style="">Q</span><span class="mord" style=""><span class="mord mathnormal" style="margin-right:0.07153em">K</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.849108em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style="">⊤</span></span></span></span></span></span></span></span></span></span></span></span></span>这将得到一个形状为<code class="highlight"><span></span><span class="p">[</span><span class="n">seq_len</span><span class="p">,</span> <span class="n">seq_len</span><span class="p">,</span> <span class="n">batch_size</span><span class="p">,</span> <span class="n">heads</span><span class="p">]</span></code>
|
||||
<p>计算注意力分数<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.043548em;vertical-align:-0.19444em;"></span><span class="mord coloredeq eqg" style=""><span class="mord mathnormal" style="">Q</span><span class="mord" style=""><span class="mord mathnormal" style="margin-right:0.07153em">K</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.849108em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style="">⊤</span></span></span></span></span></span></span></span></span></span></span></span></span>，这将得到一个形状为<code class="highlight"><span></span><span class="p">[</span><span class="n">seq_len</span><span class="p">,</span> <span class="n">seq_len</span><span class="p">,</span> <span class="n">batch_size</span><span class="p">,</span> <span class="n">heads</span><span class="p">]</span></code>
|
||||
的张量。</p>
|
||||
|
||||
</div>
|
||||
@ -579,7 +579,7 @@ M834 80h400000v40h-400000z"></path></svg></span></span></span><span class="vlist
|
||||
<div class='section-link'>
|
||||
<a href='#section-34'>#</a>
|
||||
</div>
|
||||
<p>对 Key 序列维度上的注意力进行<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8888799999999999em;vertical-align:-0.19444em;"></span><span class="mord coloredeq eqh" style=""><span class="mord mathnormal" style="">so</span><span class="mord mathnormal" style="margin-right:0.10764em">f</span><span class="mord mathnormal" style="">t</span><span class="mord mathnormal" style="">ma</span><span class="mord mathnormal" style="">x</span></span></span></span></span></span><span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:3.0000299999999998em;vertical-align:-1.25003em;"></span><span class="mord coloredeq eqc" style=""><span class="mord" style=""><span class="mop op-limits" style=""><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.6944399999999998em;"><span style="top:-2.20556em;margin-left:0em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">se</span><span class="mord mathnormal mtight" style="margin-right:0.03588em">q</span></span></span></span><span style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span><span class="mop" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqh" style="">so</span><span class="mord mathnormal coloredeq eqh" style="margin-right:0.10764em">f</span><span class="mord mathnormal coloredeq eqh" style="">t</span><span class="mord mathnormal coloredeq eqh" style="">ma</span><span class="mord mathnormal coloredeq eqh" style="">x</span></span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:1.030548em;"><span></span></span></span></span></span></span><span class="mord" style=""><span class="delimsizing size4" style=""><span style="">(</span></span></span><span class="mord" style=""><span class="mord coloredeq eqe" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:1.095028em;"><span style="top:-2.5864385em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord sqrt mtight" style=""><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.8622307142857143em;"><span class="svg-align" style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="mord mtight" style="padding-left:0.833em"><span class="mord mtight" style=""><span class="mord mtight coloredeq eqi" style=""><span class="mord mathnormal mtight" style="">d</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.3448em;"><span style="top:-2.3487714285714287em;margin-left:0em;margin-right:0.07142857142857144em;"><span class="pstrut" style="height:2.5em;"></span><span class="sizing reset-size3 size1 mtight" style=""><span class="mord mathnormal mtight" style="margin-right:0.03148em">k</span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15122857142857138em;"><span></span></span></span></span></span></span></span></span></span><span style="top:-2.8222307142857144em;"><span class="pstrut" style="height:3em;"></span><span class="hide-tail mtight" style="min-width:0.853em;height:1.08em"><svg height="1.08em" preserveaspectratio="xMinYMin slice" viewbox="0 0 400000 1080" width="400em" xmlns="http://www.w3.org/2000/svg"><path d="M95,702
|
||||
<p>对 Key 序列维度上的注意力进行<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8888799999999999em;vertical-align:-0.19444em;"></span><span class="mord coloredeq eqh" style=""><span class="mord mathnormal" style="">so</span><span class="mord mathnormal" style="margin-right:0.10764em">f</span><span class="mord mathnormal" style="">t</span><span class="mord mathnormal" style="">ma</span><span class="mord mathnormal" style="">x</span></span></span></span></span></span>操作，<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:3.0000299999999998em;vertical-align:-1.25003em;"></span><span class="mord coloredeq eqc" style=""><span class="mord" style=""><span class="mop op-limits" style=""><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.6944399999999998em;"><span style="top:-2.20556em;margin-left:0em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">se</span><span class="mord mathnormal mtight" style="margin-right:0.03588em">q</span></span></span></span><span style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span><span class="mop" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqh" style="">so</span><span class="mord mathnormal coloredeq eqh" style="margin-right:0.10764em">f</span><span class="mord mathnormal coloredeq eqh" style="">t</span><span class="mord mathnormal coloredeq eqh" style="">ma</span><span class="mord mathnormal coloredeq eqh" style="">x</span></span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:1.030548em;"><span></span></span></span></span></span></span><span class="mord" style=""><span class="delimsizing size4" style=""><span style="">(</span></span></span><span class="mord" style=""><span class="mord coloredeq eqe" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:1.095028em;"><span style="top:-2.5864385em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord sqrt mtight" style=""><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.8622307142857143em;"><span class="svg-align" style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="mord mtight" style="padding-left:0.833em"><span class="mord mtight" style=""><span class="mord mtight coloredeq eqi" style=""><span class="mord mathnormal mtight" style="">d</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.3448em;"><span style="top:-2.3487714285714287em;margin-left:0em;margin-right:0.07142857142857144em;"><span class="pstrut" style="height:2.5em;"></span><span class="sizing reset-size3 size1 mtight" style=""><span class="mord mathnormal mtight" style="margin-right:0.03148em">k</span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15122857142857138em;"><span></span></span></span></span></span></span></span></span></span><span style="top:-2.8222307142857144em;"><span class="pstrut" style="height:3em;"></span><span class="hide-tail mtight" style="min-width:0.853em;height:1.08em"><svg height="1.08em" preserveaspectratio="xMinYMin slice" viewbox="0 0 400000 1080" width="400em" xmlns="http://www.w3.org/2000/svg"><path d="M95,702
|
||||
c-2.7,0,-7.17,-2.7,-13.5,-8c-5.8,-5.3,-9.5,-10,-9.5,-14
|
||||
c0,-2,0.3,-3.3,1,-4c1.3,-2.7,23.83,-20.7,67.5,-54
|
||||
c44.2,-33.3,65.8,-50.3,66.5,-51c1.3,-1.3,3,-2,5,-2c4.7,0,8.7,3.3,12,10
|
||||
|
||||
File diff suppressed because one or more lines are too long
Reference in New Issue
Block a user