mirror of
https://github.com/labmlai/annotated_deep_learning_paper_implementations.git
synced 2025-10-31 18:58:43 +08:00
643 lines
46 KiB
HTML
643 lines
46 KiB
HTML
<!DOCTYPE html>
|
|
<html lang="en">
|
|
<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 <a href="gpt2.html">GPT-2</a> 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">torch.optim</span> <span class="kn">import</span> <span class="n">Adam</span>
|
|
<span class="lineno">16</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">17</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">18</span>
|
|
<span class="lineno">19</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">20</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">21</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">22</span><span class="kn">from</span> <span class="nn">labml_nn.helpers.device</span> <span class="kn">import</span> <span class="n">DeviceConfigs</span>
|
|
<span class="lineno">23</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">26</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">32</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">35</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">36</span> <span class="n">d_model</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">768</span>
|
|
<span class="lineno">37</span> <span class="n">n_layers</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">12</span>
|
|
<span class="lineno">38</span> <span class="n">n_heads</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">12</span>
|
|
<span class="lineno">39</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">40</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">43</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">44</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">45</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">46</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">49</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">52</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></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-7'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-7'>#</a>
|
|
</div>
|
|
<p>Huggingface tokenizer </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">54</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></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-8'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-8'>#</a>
|
|
</div>
|
|
<p><a href="gpt2.html">GPT2 model</a> </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">56</span> <span class="n">model</span><span class="p">:</span> <span class="n">GPTModel</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>Optimizer </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">58</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></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-10'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-10'>#</a>
|
|
</div>
|
|
<p>Cross entropy loss </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">60</span> <span class="n">loss_func</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></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-11'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-11'>#</a>
|
|
</div>
|
|
<p>Dataloader </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">62</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-12'>
|
|
<div class='docs doc-strings'>
|
|
<div class='section-link'>
|
|
<a href='#section-12'>#</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">64</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-13'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-13'>#</a>
|
|
</div>
|
|
<p>Load the huggingface model and get the parameters </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">70</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">71</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-14'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-14'>#</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">74</span> <span class="n">mapping</span> <span class="o">=</span> <span class="p">{</span>
|
|
<span class="lineno">75</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">76</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">77</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">78</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">79</span> <span class="s1">'lm_head.weight'</span><span class="p">:</span> <span class="s1">'lm_head.weight'</span>
|
|
<span class="lineno">80</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>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">83</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">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_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">.attn_norm.weight'</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">.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">.attn_norm.bias'</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">.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.qkv_projection.weight'</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">.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.qkv_projection.bias'</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">.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.output_projection.weight'</span>
|
|
<span class="lineno">89</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.output_projection.bias'</span>
|
|
<span class="lineno">90</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">.ffn_norm.weight'</span>
|
|
<span class="lineno">91</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">.ffn_norm.bias'</span>
|
|
<span class="lineno">92</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.linear_in.weight'</span>
|
|
<span class="lineno">93</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.linear_in.bias'</span>
|
|
<span class="lineno">94</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.linear_out.weight'</span>
|
|
<span class="lineno">95</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.linear_out.bias'</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>Move the parameters based on mapping </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">98</span> <span class="n">new_state_dict</span> <span class="o">=</span> <span class="p">{}</span>
|
|
<span class="lineno">99</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">100</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">101</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-17'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-17'>#</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">104</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.linear_in.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">105</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.linear_out.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">106</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.qkv_projection.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">107</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.output_projection.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">108</span>
|
|
<span class="lineno">109</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">110</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-18'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-18'>#</a>
|
|
</div>
|
|
<p>Load out model. We use <code class="highlight"><span></span><span class="n">strict</span> <span class="o">=</span> <span class="kc">False</span></code>
|
|
because the state does not have LoRA weights </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">113</span> <span class="n">missing_keys</span><span class="p">,</span> <span class="n">unexpected_keys</span> <span class="o">=</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></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-19'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-19'>#</a>
|
|
</div>
|
|
<p>make sure that only lora weights are not loaded </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">116</span> <span class="k">assert</span> <span class="nb">all</span><span class="p">(</span><span class="s1">'lora'</span> <span class="ow">in</span> <span class="n">key</span> <span class="k">for</span> <span class="n">key</span> <span class="ow">in</span> <span class="n">missing_keys</span><span class="p">)</span>
|
|
<span class="lineno">117</span> <span class="k">assert</span> <span class="ow">not</span> <span class="n">unexpected_keys</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-20'>
|
|
<div class='docs doc-strings'>
|
|
<div class='section-link'>
|
|
<a href='#section-20'>#</a>
|
|
</div>
|
|
<h3>Initialize the model, optimizer and dataloader</h3>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">119</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-21'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-21'>#</a>
|
|
</div>
|
|
<p>Initialize the <a href="gpt2.html">GPT2 model</a> </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">124</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">125</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">126</span> <span class="n">d_model</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">d_model</span><span class="p">,</span>
|
|
<span class="lineno">127</span> <span class="n">n_layers</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">n_layers</span><span class="p">,</span>
|
|
<span class="lineno">128</span> <span class="n">n_heads</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">n_heads</span><span class="p">,</span>
|
|
<span class="lineno">129</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">130</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">131</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">132</span> <span class="p">)</span>
|
|
<span class="lineno">133</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-22'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-22'>#</a>
|
|
</div>
|
|
<p>Load pre-trained model weights </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">_load_pretrained_weights</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>Initialize the optimizer </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">138</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-24'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-24'>#</a>
|
|
</div>
|
|
<p>Initialize the data loader </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">141</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-25'>
|
|
<div class='docs doc-strings'>
|
|
<div class='section-link'>
|
|
<a href='#section-25'>#</a>
|
|
</div>
|
|
<h3>Training loop</h3>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">143</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-26'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-26'>#</a>
|
|
</div>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">148</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></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-27'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-27'>#</a>
|
|
</div>
|
|
<p><code class="highlight"><span></span><span class="n">inputs</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>
|
|
</p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">150</span> <span class="k">for</span> <span class="p">(</span><span class="n">inputs</span><span class="p">,)</span> <span class="ow">in</span> <span class="n">monit</span><span class="o">.</span><span class="n">iterate</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></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-28'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-28'>#</a>
|
|
</div>
|
|
<p>Move <code class="highlight"><span></span><span class="n">inputs</span></code>
|
|
to device </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">152</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></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-29'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-29'>#</a>
|
|
</div>
|
|
<p>Call the model, with the all but the last token </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">154</span> <span class="n">logits</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="p">:</span><span class="o">-</span><span class="mi">1</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>Get cross entropy loss </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">156</span> <span class="n">loss</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">loss_func</span><span class="p">(</span><span class="n">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">logits</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">]),</span> <span class="n">inputs</span><span class="p">[:,</span> <span class="mi">1</span><span class="p">:]</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></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-31'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-31'>#</a>
|
|
</div>
|
|
<p>Make gradients 0 </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">159</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></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-32'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-32'>#</a>
|
|
</div>
|
|
<p>Compute gradients </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">161</span> <span class="n">loss</span><span class="o">.</span><span class="n">backward</span><span class="p">()</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-33'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-33'>#</a>
|
|
</div>
|
|
<p>Optimize </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">163</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></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-34'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-34'>#</a>
|
|
</div>
|
|
<p>Log the loss </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">166</span> <span class="n">tracker</span><span class="o">.</span><span class="n">save</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">167</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add_global_step</span><span class="p">()</span></pre></div>
|
|
</div>
|
|
</div>
|
|
<div class='section' id='section-35'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-35'>#</a>
|
|
</div>
|
|
<p> </p>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">169</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-36'>
|
|
<div class='docs doc-strings'>
|
|
<div class='section-link'>
|
|
<a href='#section-36'>#</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">172</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">173</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-37'>
|
|
<div class='docs'>
|
|
<div class='section-link'>
|
|
<a href='#section-37'>#</a>
|
|
</div>
|
|
|
|
</div>
|
|
<div class='code'>
|
|
<div class="highlight"><pre><span class="lineno">179</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">180</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">181</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">182</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">183</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">184</span>
|
|
<span class="lineno">185</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">186</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">187</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">188</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">189</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> |