mirror of
https://github.com/labmlai/annotated_deep_learning_paper_implementations.git
synced 2025-11-02 04:37:46 +08:00
📚 feedback transformer notes
This commit is contained in:
Varuna Jayasiri
@ -253,14 +253,21 @@
|
||||
|
||||
<url>
|
||||
<loc>https://nn.labml.ai/transformers/feedback/experiment.html</loc>
|
||||
<lastmod>2021-01-14T16:30:00+00:00</lastmod>
|
||||
<lastmod>2021-01-29T16:30:00+00:00</lastmod>
|
||||
<priority>1.00</priority>
|
||||
</url>
|
||||
|
||||
|
||||
<url>
|
||||
<loc>https://nn.labml.ai/transformers/feedback/index.html</loc>
|
||||
<lastmod>2021-01-25T16:30:00+00:00</lastmod>
|
||||
<lastmod>2021-01-29T16:30:00+00:00</lastmod>
|
||||
<priority>1.00</priority>
|
||||
</url>
|
||||
|
||||
|
||||
<url>
|
||||
<loc>https://nn.labml.ai/transformers/feedback/experiment.html</loc>
|
||||
<lastmod>2021-01-29T16:30:00+00:00</lastmod>
|
||||
<priority>1.00</priority>
|
||||
</url>
|
||||
|
||||
|
||||
422
docs/transformers/feedback/experiment.html
Normal file
422
docs/transformers/feedback/experiment.html
Normal file
@ -0,0 +1,422 @@
|
||||
<!DOCTYPE html>
|
||||
<html>
|
||||
<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 a feedback transformer."/>
|
||||
|
||||
<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="Train Feedback Transformer"/>
|
||||
<meta name="twitter:description" content="This is training code with notes for a feedback transformer."/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/transformers/feedback/experiment.html"/>
|
||||
<meta property="og:title" content="Train Feedback Transformer"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="LabML Neural Networks"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="Train Feedback Transformer"/>
|
||||
<meta property="og:description" content="This is training code with notes for a feedback transformer."/>
|
||||
|
||||
<title>Train Feedback Transformer</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../../pylit.css">
|
||||
<link rel="canonical" href="https://nn.labml.ai/transformers/feedback/experiment.html"/>
|
||||
<!-- 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">transformers</a>
|
||||
<a class="parent" href="index.html">feedback</a>
|
||||
</p>
|
||||
<p>
|
||||
|
||||
<a href="https://github.com/lab-ml/labml_nn/tree/master/labml_nn/transformers/feedback/experiment.py">
|
||||
<img alt="Github"
|
||||
src="https://img.shields.io/github/stars/lab-ml/nn?style=social"
|
||||
style="max-width:100%;"/></a>
|
||||
<a href="https://join.slack.com/t/labforml/shared_invite/zt-egj9zvq9-Dl3hhZqobexgT7aVKnD14g/"
|
||||
rel="nofollow">
|
||||
<img alt="Join Slact"
|
||||
src="https://img.shields.io/badge/slack-chat-green.svg?logo=slack"
|
||||
style="max-width:100%;"/></a>
|
||||
<a href="https://twitter.com/labmlai"
|
||||
rel="nofollow">
|
||||
<img alt="Twitter"
|
||||
src="https://img.shields.io/twitter/follow/labmlai?style=social"
|
||||
style="max-width:100%;"/></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>Train Feedback Transformer</h1>
|
||||
<p>This trains a <a href="index.html">feedback transformer</a> model for auto-regression.
|
||||
You can pick the original feedback transformer or the new version
|
||||
where the keys and values are precalculated.</p>
|
||||
<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/lab-ml/nn/blob/master/labml_nn/transformers/feedback/experiment.ipynb"><img alt="Open In Colab" src="https://colab.research.google.com/assets/colab-badge.svg" /></a>
|
||||
<a href="https://web.lab-ml.com/run?uuid=d8eb9416530a11eb8fb50242ac1c0002"><img alt="View Run" src="https://img.shields.io/badge/labml-experiment-brightgreen" /></a></p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">19</span><span></span><span class="kn">import</span> <span class="nn">torch</span>
|
||||
<span class="lineno">20</span><span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">nn</span>
|
||||
<span class="lineno">21</span>
|
||||
<span class="lineno">22</span><span class="kn">from</span> <span class="nn">labml</span> <span class="kn">import</span> <span class="n">experiment</span>
|
||||
<span class="lineno">23</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">24</span><span class="kn">from</span> <span class="nn">labml.utils.pytorch</span> <span class="kn">import</span> <span class="n">get_modules</span>
|
||||
<span class="lineno">25</span><span class="kn">from</span> <span class="nn">labml_helpers.module</span> <span class="kn">import</span> <span class="n">Module</span>
|
||||
<span class="lineno">26</span>
|
||||
<span class="lineno">27</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>
|
||||
<span class="lineno">28</span><span class="kn">from</span> <span class="nn">labml_nn.transformers</span> <span class="kn">import</span> <span class="n">Encoder</span><span class="p">,</span> <span class="n">Generator</span><span class="p">,</span> <span class="n">TransformerConfigs</span>
|
||||
<span class="lineno">29</span><span class="kn">from</span> <span class="nn">labml_nn.transformers.utils</span> <span class="kn">import</span> <span class="n">subsequent_mask</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>Auto regressive model</h2>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">32</span><span class="k">class</span> <span class="nc">AutoregressiveModel</span><span class="p">(</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">37</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_vocab</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">d_model</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">transformer</span><span class="p">:</span> <span class="n">Module</span><span class="p">):</span>
|
||||
<span class="lineno">38</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>Token embedding module</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">40</span> <span class="bp">self</span><span class="o">.</span><span class="n">src_embed</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Embedding</span><span class="p">(</span><span class="n">n_vocab</span><span class="p">,</span> <span class="n">d_model</span><span class="p">)</span>
|
||||
<span class="lineno">41</span> <span class="bp">self</span><span class="o">.</span><span class="n">transformer</span> <span class="o">=</span> <span class="n">transformer</span>
|
||||
<span class="lineno">42</span> <span class="bp">self</span><span class="o">.</span><span class="n">generator</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Linear</span><span class="p">(</span><span class="n">d_model</span><span class="p">,</span> <span class="n">n_vocab</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">44</span> <span class="k">def</span> <span class="fm">__call__</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-5'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-5'>#</a>
|
||||
</div>
|
||||
<p>Embed the tokens</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">46</span> <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">src_embed</span><span class="p">(</span><span class="n">x</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>Run it through the the transformer</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">48</span> <span class="n">res</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">transformer</span><span class="p">(</span><span class="n">x</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>Generate logits of the next token</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">50</span> <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">generator</span><span class="p">(</span><span class="n">res</span><span class="p">),</span> <span class="kc">None</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-8'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-8'>#</a>
|
||||
</div>
|
||||
<h2>Configurations</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">53</span><span class="k">class</span> <span class="nc">Configs</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">60</span> <span class="n">model</span><span class="p">:</span> <span class="n">AutoregressiveModel</span>
|
||||
<span class="lineno">61</span>
|
||||
<span class="lineno">62</span> <span class="n">d_model</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">512</span>
|
||||
<span class="lineno">63</span> <span class="n">heads</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">8</span>
|
||||
<span class="lineno">64</span> <span class="n">dropout</span><span class="p">:</span> <span class="nb">float</span> <span class="o">=</span> <span class="mf">0.0</span>
|
||||
<span class="lineno">65</span> <span class="n">d_ff</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">2048</span>
|
||||
<span class="lineno">66</span> <span class="n">n_layers</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">6</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>Create <a href="index.html">original feedback transformer</a>.</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">69</span><span class="nd">@option</span><span class="p">(</span><span class="n">Configs</span><span class="o">.</span><span class="n">model</span><span class="p">)</span>
|
||||
<span class="lineno">70</span><span class="k">def</span> <span class="nf">feedback_transformer</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">Configs</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">74</span> <span class="kn">from</span> <span class="nn">labml_nn.transformers.feedback</span> <span class="kn">import</span> <span class="n">FeedbackTransformer</span><span class="p">,</span> <span class="n">FeedbackTransformerLayer</span><span class="p">,</span> \
|
||||
<span class="lineno">75</span> <span class="n">FeedbackAttention</span><span class="p">,</span> <span class="n">FeedForward</span>
|
||||
<span class="lineno">76</span>
|
||||
<span class="lineno">77</span> <span class="k">return</span> <span class="n">AutoregressiveModel</span><span class="p">(</span>
|
||||
<span class="lineno">78</span> <span class="n">c</span><span class="o">.</span><span class="n">n_tokens</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">d_model</span><span class="p">,</span>
|
||||
<span class="lineno">79</span> <span class="n">FeedbackTransformer</span><span class="p">(</span>
|
||||
<span class="lineno">80</span> <span class="n">FeedbackTransformerLayer</span><span class="p">(</span><span class="n">d_model</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">d_model</span><span class="p">,</span>
|
||||
<span class="lineno">81</span> <span class="n">attn</span><span class="o">=</span><span class="n">FeedbackAttention</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">heads</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">d_model</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">dropout</span><span class="p">),</span>
|
||||
<span class="lineno">82</span> <span class="n">feed_forward</span><span class="o">=</span><span class="n">FeedForward</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">d_model</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">d_ff</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">dropout</span><span class="p">),</span>
|
||||
<span class="lineno">83</span> <span class="n">dropout_prob</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">dropout</span><span class="p">),</span>
|
||||
<span class="lineno">84</span> <span class="n">c</span><span class="o">.</span><span class="n">n_layers</span><span class="p">))</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">device</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-12'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-12'>#</a>
|
||||
</div>
|
||||
<p>Create <a href="index.html#kv_shared">updated feedback transformer</a>, with precalculated keys and values.</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">87</span><span class="nd">@option</span><span class="p">(</span><span class="n">Configs</span><span class="o">.</span><span class="n">model</span><span class="p">)</span>
|
||||
<span class="lineno">88</span><span class="k">def</span> <span class="nf">feedback_transformer_kv</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">Configs</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">92</span> <span class="kn">from</span> <span class="nn">labml_nn.transformers.feedback</span> <span class="kn">import</span> <span class="n">FeedbackTransformerKV</span><span class="p">,</span> <span class="n">FeedbackTransformerLayer</span><span class="p">,</span> \
|
||||
<span class="lineno">93</span> <span class="n">FeedbackAttention</span><span class="p">,</span> <span class="n">FeedForward</span>
|
||||
<span class="lineno">94</span>
|
||||
<span class="lineno">95</span> <span class="k">return</span> <span class="n">AutoregressiveModel</span><span class="p">(</span>
|
||||
<span class="lineno">96</span> <span class="n">c</span><span class="o">.</span><span class="n">n_tokens</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">d_model</span><span class="p">,</span>
|
||||
<span class="lineno">97</span> <span class="n">FeedbackTransformerKV</span><span class="p">(</span>
|
||||
<span class="lineno">98</span> <span class="n">FeedbackTransformerLayer</span><span class="p">(</span><span class="n">d_model</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">d_model</span><span class="p">,</span>
|
||||
<span class="lineno">99</span> <span class="n">attn</span><span class="o">=</span><span class="n">FeedbackAttention</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">heads</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">d_model</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">dropout</span><span class="p">,</span>
|
||||
<span class="lineno">100</span> <span class="n">is_kv_precomputed</span><span class="o">=</span><span class="kc">True</span><span class="p">),</span>
|
||||
<span class="lineno">101</span> <span class="n">feed_forward</span><span class="o">=</span><span class="n">FeedForward</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">d_model</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">d_ff</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">dropout</span><span class="p">),</span>
|
||||
<span class="lineno">102</span> <span class="n">dropout_prob</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">dropout</span><span class="p">),</span>
|
||||
<span class="lineno">103</span> <span class="n">c</span><span class="o">.</span><span class="n">n_layers</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">d_model</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">heads</span><span class="p">))</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">device</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">106</span><span class="k">def</span> <span class="nf">main</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>Create experiment</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">108</span> <span class="n">experiment</span><span class="o">.</span><span class="n">create</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s2">"feedback_transformer"</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>Create configs</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">110</span> <span class="n">conf</span> <span class="o">=</span> <span class="n">Configs</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>Load configurations</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">112</span> <span class="n">experiment</span><span class="o">.</span><span class="n">configs</span><span class="p">(</span><span class="n">conf</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>A dictionary of configurations to override</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">114</span> <span class="p">{</span><span class="s1">'tokenizer'</span><span class="p">:</span> <span class="s1">'character'</span><span class="p">,</span>
|
||||
<span class="lineno">115</span> <span class="s1">'text'</span><span class="p">:</span> <span class="s1">'tiny_shakespeare'</span><span class="p">,</span>
|
||||
<span class="lineno">116</span> <span class="s1">'optimizer.learning_rate'</span><span class="p">:</span> <span class="mf">1.0</span><span class="p">,</span>
|
||||
<span class="lineno">117</span> <span class="s1">'optimizer.optimizer'</span><span class="p">:</span> <span class="s1">'Noam'</span><span class="p">,</span>
|
||||
<span class="lineno">118</span> <span class="s1">'prompt'</span><span class="p">:</span> <span class="s1">'It is'</span><span class="p">,</span>
|
||||
<span class="lineno">119</span> <span class="s1">'prompt_separator'</span><span class="p">:</span> <span class="s1">''</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>
|
||||
<p>Use <code>feedback_transformer</code> for original feedback transformer</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">122</span> <span class="s1">'model'</span><span class="p">:</span> <span class="s1">'feedback_transformer_kv'</span><span class="p">,</span>
|
||||
<span class="lineno">123</span>
|
||||
<span class="lineno">124</span> <span class="s1">'train_loader'</span><span class="p">:</span> <span class="s1">'shuffled_train_loader'</span><span class="p">,</span>
|
||||
<span class="lineno">125</span> <span class="s1">'valid_loader'</span><span class="p">:</span> <span class="s1">'shuffled_valid_loader'</span><span class="p">,</span>
|
||||
<span class="lineno">126</span>
|
||||
<span class="lineno">127</span> <span class="s1">'seq_len'</span><span class="p">:</span> <span class="mi">128</span><span class="p">,</span>
|
||||
<span class="lineno">128</span> <span class="s1">'epochs'</span><span class="p">:</span> <span class="mi">128</span><span class="p">,</span>
|
||||
<span class="lineno">129</span> <span class="s1">'batch_size'</span><span class="p">:</span> <span class="mi">64</span><span class="p">,</span>
|
||||
<span class="lineno">130</span> <span class="s1">'inner_iterations'</span><span class="p">:</span> <span class="mi">25</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>
|
||||
<p>Set models for saving and loading</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">133</span> <span class="n">experiment</span><span class="o">.</span><span class="n">add_pytorch_models</span><span class="p">(</span><span class="n">get_modules</span><span class="p">(</span><span class="n">conf</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>Start the experiment</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">136</span> <span class="k">with</span> <span class="n">experiment</span><span class="o">.</span><span class="n">start</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>Run the training loop</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">138</span> <span class="n">conf</span><span class="o">.</span><span class="n">run</span><span class="p">()</span>
|
||||
<span class="lineno">139</span>
|
||||
<span class="lineno">140</span>
|
||||
<span class="lineno">141</span><span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s1">'__main__'</span><span class="p">:</span>
|
||||
<span class="lineno">142</span> <span class="n">main</span><span class="p">()</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
</div>
|
||||
</div>
|
||||
<script src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.4/MathJax.js?config=TeX-AMS_HTML">
|
||||
</script>
|
||||
<!-- MathJax configuration -->
|
||||
<script type="text/x-mathjax-config">
|
||||
MathJax.Hub.Config({
|
||||
tex2jax: {
|
||||
inlineMath: [ ['$','$'] ],
|
||||
displayMath: [ ['$$','$$'] ],
|
||||
processEscapes: true,
|
||||
processEnvironments: true
|
||||
},
|
||||
// Center justify equations in code and markdown cells. Elsewhere
|
||||
// we use CSS to left justify single line equations in code cells.
|
||||
displayAlign: 'center',
|
||||
"HTML-CSS": { fonts: ["TeX"] }
|
||||
});
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
</script>
|
||||
</body>
|
||||
</html>
|
||||
File diff suppressed because it is too large
Load Diff
@ -22,11 +22,19 @@ In order to speed up the training the paper discusses starting with a short sequ
|
||||
gradually increasing it.
|
||||
They also discuss using a pretrained parallel transformer as the starting point.
|
||||
|
||||
The feedback transformer doesn't keep the outputs of all layers.
|
||||
The original feedback transformer doesn't keep the outputs of all layers.
|
||||
Instead it keeps weighted sum of the output of all layers.
|
||||
This reduces the memory used for caching during prediction.
|
||||
The first half of this file implements this.
|
||||
|
||||
Here's a notebook for training a feedback transformer on Tiny Shakespeare dataset.
|
||||
The updated feedback transformer shares weights $W^l_k$ and $W^l_v$ used
|
||||
to calculate keys and values for among the layers.
|
||||
We then calculate the keys and values for each step only once and keep
|
||||
them cached.
|
||||
The [second half](#shared_kv) of this file implements this.
|
||||
We implemented a custom PyTorch function to improve performance.
|
||||
|
||||
Here's [the training code](experiment.html) and a notebook for training a feedback transformer on Tiny Shakespeare dataset.
|
||||
|
||||
[](https://colab.research.google.com/github/lab-ml/nn/blob/master/labml_nn/transformers/feedback/experiment.ipynb)
|
||||
[](https://web.lab-ml.com/run?uuid=d8eb9416530a11eb8fb50242ac1c0002)
|
||||
@ -39,8 +47,8 @@ import torch
|
||||
from torch import nn
|
||||
|
||||
from labml_helpers.module import Module
|
||||
from labml_nn.transformers.mha import PrepareForMultiHeadAttention
|
||||
from labml_nn.transformers.feed_forward import FeedForward
|
||||
from labml_nn.transformers.mha import PrepareForMultiHeadAttention
|
||||
from labml_nn.utils import clone_module_list
|
||||
|
||||
|
||||
@ -61,6 +69,7 @@ class FeedbackAttention(Module):
|
||||
* 'heads' is the number of attention heads
|
||||
* `d_model` is the number of features in the transformer
|
||||
* `dropout_prob` is the attention dropout probability
|
||||
* `is_kv_precomputed` is whether key, value tensors are already calculated
|
||||
"""
|
||||
|
||||
super().__init__()
|
||||
@ -70,11 +79,13 @@ class FeedbackAttention(Module):
|
||||
#
|
||||
self.heads = heads
|
||||
|
||||
# These transform the `query`, `key` and `value` vectors for multi-headed attention.
|
||||
# These transform the `query` multi-headed attention.
|
||||
self.query = PrepareForMultiHeadAttention(d_model, heads, self.d_k, bias=False)
|
||||
# These transform the `key` and `value` fir multi-headed attention.
|
||||
if not is_kv_precomputed:
|
||||
self.key = PrepareForMultiHeadAttention(d_model, heads, self.d_k, bias=False)
|
||||
self.value = PrepareForMultiHeadAttention(d_model, heads, self.d_k, bias=True)
|
||||
# Keys and values are already calculated
|
||||
else:
|
||||
self.key = None
|
||||
self.value = None
|
||||
@ -94,6 +105,8 @@ class FeedbackAttention(Module):
|
||||
|
||||
# Relative positional embeddings for key relative to the query.
|
||||
self.key_pos_embeddings = nn.Parameter(torch.zeros((self.P, heads, self.d_k)), requires_grad=True)
|
||||
# Relative positional embedding bias for key relative to the query.
|
||||
self.key_pos_bias = nn.Parameter(torch.zeros((self.P, heads)), requires_grad=True)
|
||||
# Positional embeddings for the query is independent of the position of the query
|
||||
self.query_pos_bias = nn.Parameter(torch.zeros((heads, self.d_k)), requires_grad=True)
|
||||
|
||||
@ -104,27 +117,42 @@ class FeedbackAttention(Module):
|
||||
"""
|
||||
### Get attention scores
|
||||
|
||||
We use relative positional encodings for attention, similar
|
||||
to [relative multi-head attention form Transformer-XL paper](../relative_mha.html).
|
||||
|
||||
Attention from current step's query to key in step $j$ (relative to current step) is,
|
||||
|
||||
\begin{align}
|
||||
A_{j} &= Q^\top K_j \\
|
||||
&= lin_q(X^q + P_q)^\top lin_k(X^k_j + P_j) \\
|
||||
&= (Q + U^Q)^\top(K_j + U^K_j)
|
||||
&= (Q + U^Q)^\top(K_j + U^K_j) \\
|
||||
&= \underset{\color{lightgreen}{A}}{Q^\top K_j} +
|
||||
\underset{\color{lightgreen}{B}}{Q^\top U^K_j} +
|
||||
\underset{\color{lightgreen}{C}}{{U^Q}^\top K_j} +
|
||||
\underset{\color{lightgreen}{D}}{{U^Q}^\top U^K_j}
|
||||
\end{align}
|
||||
|
||||
where $Q, K_j$, are linear transformations of
|
||||
original embeddings $X^q, X^k_j$
|
||||
and $U^Q, U^K_j$ are linear transformations of
|
||||
absolute positional encodings $P_q, P_j$.
|
||||
positional encodings $P_q, P_j$.
|
||||
|
||||
We replace term $\color{lightgreen}{D}$ with $S_j$.
|
||||
"""
|
||||
|
||||
# $U^K_j$
|
||||
key_pos_emb = self.key_pos_embeddings[-key.shape[0]:]
|
||||
# $U^Q$
|
||||
query_pos_bias = self.query_pos_bias[None, :, :]
|
||||
# $S_j$
|
||||
key_pos_bias = self.key_pos_bias[-key.shape[0]:]
|
||||
|
||||
# $(Q + U^Q)^\top(K_j + U^K_j)$
|
||||
# $\underset{\color{lightgreen}{A}}{Q^\top K_j} + \underset{\color{lightgreen}{C}}{{U^Q}^\top K_j}$
|
||||
ac = torch.einsum('bhd,jbhd->jbh', query + query_pos_bias, key)
|
||||
bd = torch.einsum('bhd,jhd->jbh', query, key_pos_emb)
|
||||
# $\underset{\color{lightgreen}{B}}{Q^\top U^K_j} + \underset{\color{lightgreen}{D}}{S_j}$
|
||||
bd = torch.einsum('bhd,jhd->jbh', query, key_pos_emb) + key_pos_bias[:, None, :]
|
||||
|
||||
# $A_j$
|
||||
return ac + bd
|
||||
|
||||
def __call__(self, *,
|
||||
@ -283,23 +311,69 @@ class FeedbackTransformer(Module):
|
||||
return self.norm(res)
|
||||
|
||||
|
||||
# <a id="shared_kv">
|
||||
# # Shared keys and values for among layers
|
||||
# </a>
|
||||
|
||||
class StackFunction(torch.autograd.Function):
|
||||
"""
|
||||
### Stack Function implementation
|
||||
|
||||
We implement a custom function instead of appending to a python list
|
||||
and then doing `torch.stack`.
|
||||
This greatly improves the performance over calling `torch.stack` at
|
||||
each step along the sequence.
|
||||
Everytime `torch.stack` is called it creates a new tensor, while
|
||||
this method and the accompanying class `Stack` share memory for each step.
|
||||
"""
|
||||
|
||||
@staticmethod
|
||||
def forward(ctx, memory, memory_grad, last, n):
|
||||
"""
|
||||
* `ctx` is the context of the function (which lets as cache stuff)
|
||||
* `memory` is the shared memory tensor where we stack and store the values of each step (keys & values)
|
||||
* `memory_grad` is the shared memory tensor to store and accumulate gradients of each step
|
||||
* `last` is the last value stacked
|
||||
* `n` is the number of steps (i.e. size of the stack)
|
||||
|
||||
This returns the stacked tensor for steps upto `n`.
|
||||
"""
|
||||
|
||||
# Cache accumulated gradients
|
||||
ctx._mem_grad = memory_grad
|
||||
# Cache the size of the stack
|
||||
ctx._n = n
|
||||
# Return the stack
|
||||
return memory[:n + 1]
|
||||
|
||||
@staticmethod
|
||||
def backward(ctx, grad_output):
|
||||
"""
|
||||
* `grad_output` is the gradient with respect to the output of about `forward` function
|
||||
|
||||
This accumulates the gradients in the shared memory tensor and return the
|
||||
gradients with respect to the `last` result in the stack.
|
||||
"""
|
||||
# Get the current size of the stack
|
||||
n = ctx._n
|
||||
# Get the accumulated gradients
|
||||
memory_grad = ctx._mem_grad
|
||||
# Add the gradients
|
||||
memory_grad[:n + 1] += grad_output
|
||||
# Return the gradients w.r.t to last value in the stack
|
||||
return None, None, memory_grad[n], None
|
||||
|
||||
|
||||
class Stack:
|
||||
"""
|
||||
### Stack Module
|
||||
|
||||
This uses the stack function defined above, and does the necessary initializations.
|
||||
"""
|
||||
def __init__(self, max_len: int):
|
||||
"""
|
||||
* `max_len` is the maximum size of the stack
|
||||
"""
|
||||
self.max_len = max_len
|
||||
self.memory = None
|
||||
self.memory_grad = None
|
||||
@ -307,37 +381,70 @@ class Stack:
|
||||
self.n = -1
|
||||
self.last_get_n = -1
|
||||
|
||||
def append(self, n: int, vector: torch.Tensor):
|
||||
def append(self, n: int, value: torch.Tensor):
|
||||
"""
|
||||
* `n` is the size of the stack
|
||||
* `value` is the tensor that needs to be added to the stack
|
||||
"""
|
||||
|
||||
# You need to get (use) the stack after adding a value.
|
||||
# Otherwise this implementation fails
|
||||
assert n == 0 or self.last_get_n == n - 1, f"{n}, {self.last_get_n}"
|
||||
|
||||
# Do this without gradients
|
||||
with torch.no_grad():
|
||||
if self.memory is None or self.memory.shape[1:] != vector.shape:
|
||||
# Initialize the shared memory tensor to keep the stack
|
||||
if self.memory is None or self.memory.shape[1:] != value.shape:
|
||||
# This should only happen when the stack is empty
|
||||
assert n == 0
|
||||
self.memory = vector.new_zeros(self.max_len, *vector.shape, requires_grad=False)
|
||||
self.memory_grad = vector.new_zeros(self.memory.shape, requires_grad=False)
|
||||
# Create a tensor for the stack
|
||||
self.memory = value.new_zeros(self.max_len, *value.shape, requires_grad=False)
|
||||
# Create a tensor to accumulate the gradients
|
||||
self.memory_grad = value.new_zeros(self.memory.shape, requires_grad=False)
|
||||
# The memory is already initialized but we are resetting the stack.
|
||||
#
|
||||
# This could have been another function like `reset`, but
|
||||
# we found this easier to use.
|
||||
elif n == 0:
|
||||
# Reset accumulated gradients
|
||||
self.memory_grad.fill_(0.)
|
||||
|
||||
# memory[n] = vector.detach()
|
||||
self.memory.data[n] = vector.detach()
|
||||
# Set the value in the correct position of the stack
|
||||
self.memory.data[n] = value.detach()
|
||||
# Keep track of the stack (for debugging)
|
||||
self.n = n
|
||||
|
||||
self.last = vector
|
||||
# Keep track of the last value added to the stack.
|
||||
# We need this to be passed on to `StackFunction` in order
|
||||
# to get the gradients propagated backwards.
|
||||
self.last = value
|
||||
|
||||
def get(self):
|
||||
"""
|
||||
Returns the stack
|
||||
"""
|
||||
|
||||
# Keep track of the size of the stack when it was used.
|
||||
# This is used for a sanity check in `append`.
|
||||
self.last_get_n = self.n
|
||||
# Take it all through `StackFunction` so that `StackFunction.backwards`
|
||||
# is called by PyTorch during backpropagation.
|
||||
return StackFunction.apply(self.memory, self.memory_grad, self.last, self.n)
|
||||
|
||||
|
||||
class FeedbackTransformerKV(Module):
|
||||
"""
|
||||
## Feedback Transformer Module
|
||||
## Updated Feedback Transformer Module
|
||||
|
||||
This is the updated feedback transformer module that caches the keys and values.
|
||||
"""
|
||||
|
||||
def __init__(self, layer: FeedbackTransformerLayer, n_layers: int, d_model: int, heads: int):
|
||||
"""
|
||||
* `layer` is the feedback transformer layer, which we clone for each layer
|
||||
* `n_layers` is the number of layers in the transformer
|
||||
* `d_model` is the number of features in the transformer
|
||||
* 'heads' is the number of attention heads
|
||||
"""
|
||||
|
||||
super().__init__()
|
||||
@ -351,11 +458,16 @@ class FeedbackTransformerKV(Module):
|
||||
# Softmax for weights before taking the weighted sum
|
||||
self.softmax = nn.Softmax(0)
|
||||
|
||||
# Number of features in a head
|
||||
d_k = d_model // heads
|
||||
# Module to transform embeddings (memory) to get keys
|
||||
self.key = PrepareForMultiHeadAttention(d_model, heads, d_k, bias=False)
|
||||
# Module to transform embeddings (memory) to get keys
|
||||
self.value = PrepareForMultiHeadAttention(d_model, heads, d_k, bias=False)
|
||||
|
||||
# Memory for stacked keys
|
||||
self.mem_key = Stack(512)
|
||||
# Memory for stacked values
|
||||
self.mem_value = Stack(512)
|
||||
|
||||
def __call__(self, x_seq: torch.Tensor):
|
||||
@ -372,9 +484,10 @@ class FeedbackTransformerKV(Module):
|
||||
# List to store layer outputs
|
||||
layer_outputs = [x]
|
||||
|
||||
# If there is memory, stack them into a vector
|
||||
# Stack of keys and values
|
||||
key_tensor = None
|
||||
value_tensor = None
|
||||
# Get the keys and values tensors if we are beyond the initial step
|
||||
if step > 0:
|
||||
key_tensor = self.mem_key.get()
|
||||
value_tensor = self.mem_value.get()
|
||||
@ -390,7 +503,9 @@ class FeedbackTransformerKV(Module):
|
||||
layer_outputs = torch.stack(layer_outputs)
|
||||
# Calculate the memory vector as a weighted sum of layer outputs
|
||||
mem = torch.einsum('lbd,l->bd', layer_outputs, self.softmax(self.weights))
|
||||
# Calculate the keys from memory and add it to the stack
|
||||
self.mem_key.append(step, self.key(mem))
|
||||
# Calculate the values from memory and add it to the stack
|
||||
self.mem_value.append(step, self.value(mem))
|
||||
# Append the output to results
|
||||
res.append(x)
|
||||
|
||||
@ -7,6 +7,13 @@ summary: This is training code with notes for a feedback transformer.
|
||||
# Train Feedback Transformer
|
||||
|
||||
This trains a [feedback transformer](index.html) model for auto-regression.
|
||||
You can pick the original feedback transformer or the new version
|
||||
where the keys and values are precalculated.
|
||||
|
||||
Here's a Colab notebook for training a feedback transformer on Tiny Shakespeare dataset.
|
||||
|
||||
[](https://colab.research.google.com/github/lab-ml/nn/blob/master/labml_nn/transformers/feedback/experiment.ipynb)
|
||||
[](https://web.lab-ml.com/run?uuid=d8eb9416530a11eb8fb50242ac1c0002)
|
||||
"""
|
||||
|
||||
import torch
|
||||
@ -35,8 +42,9 @@ class AutoregressiveModel(Module):
|
||||
self.generator = nn.Linear(d_model, n_vocab)
|
||||
|
||||
def __call__(self, x: torch.Tensor):
|
||||
# Embed the tokens
|
||||
x = self.src_embed(x)
|
||||
# Embed the tokens (`src`) and run it through the the transformer
|
||||
# Run it through the the transformer
|
||||
res = self.transformer(x)
|
||||
# Generate logits of the next token
|
||||
return self.generator(res), None
|
||||
@ -60,6 +68,9 @@ class Configs(NLPAutoRegressionConfigs):
|
||||
|
||||
@option(Configs.model)
|
||||
def feedback_transformer(c: Configs):
|
||||
"""
|
||||
Create [original feedback transformer](index.html).
|
||||
"""
|
||||
from labml_nn.transformers.feedback import FeedbackTransformer, FeedbackTransformerLayer, \
|
||||
FeedbackAttention, FeedForward
|
||||
|
||||
@ -75,6 +86,9 @@ def feedback_transformer(c: Configs):
|
||||
|
||||
@option(Configs.model)
|
||||
def feedback_transformer_kv(c: Configs):
|
||||
"""
|
||||
Create [updated feedback transformer](index.html#kv_shared), with precalculated keys and values.
|
||||
"""
|
||||
from labml_nn.transformers.feedback import FeedbackTransformerKV, FeedbackTransformerLayer, \
|
||||
FeedbackAttention, FeedForward
|
||||
|
||||
@ -104,6 +118,7 @@ def main():
|
||||
'prompt': 'It is',
|
||||
'prompt_separator': '',
|
||||
|
||||
# Use `feedback_transformer` for original feedback transformer
|
||||
'model': 'feedback_transformer_kv',
|
||||
|
||||
'train_loader': 'shuffled_train_loader',
|
||||
@ -119,7 +134,7 @@ def main():
|
||||
|
||||
# Start the experiment
|
||||
with experiment.start():
|
||||
# `TrainValidConfigs.run`
|
||||
# Run the training loop
|
||||
conf.run()
|
||||
|
||||
|
||||
|
||||
Reference in New Issue
Block a user