mirror of
https://github.com/labmlai/annotated_deep_learning_paper_implementations.git
synced 2025-08-26 08:41:23 +08:00
LoRA GPT2 n_heads fix and notes
This commit is contained in:
Varuna Jayasiri
@ -70,7 +70,7 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-0'>#</a>
|
||||
</div>
|
||||
<h1>Finetune GPT-2 with <a href="index.html">LoRA</a></h1>
|
||||
<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>
|
||||
|
||||
@ -165,24 +165,19 @@
|
||||
|
||||
</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>
|
||||
<span class="lineno">53</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">54</span> <span class="n">model</span><span class="p">:</span> <span class="n">GPTModel</span>
|
||||
<span class="lineno">55</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">56</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>
|
||||
<span class="lineno">57</span> <span class="n">data_loader</span><span class="p">:</span> <span class="n">DataLoader</span></pre></div>
|
||||
<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 doc-strings'>
|
||||
<div class='docs'>
|
||||
<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>
|
||||
<p>Huggingface tokenizer </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">59</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 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'>
|
||||
@ -190,12 +185,11 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-8'>#</a>
|
||||
</div>
|
||||
<p>Load the huggingface model and get the parameters </p>
|
||||
<p><a href="gpt2.html">GPT2 model</a> </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">65</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">66</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 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'>
|
||||
@ -203,18 +197,11 @@
|
||||
<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>
|
||||
<p>Optimizer </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">69</span> <span class="n">mapping</span> <span class="o">=</span> <span class="p">{</span>
|
||||
<span class="lineno">70</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">71</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">72</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">73</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">74</span> <span class="s1">'lm_head.weight'</span><span class="p">:</span> <span class="s1">'lm_head.weight'</span>
|
||||
<span class="lineno">75</span> <span class="p">}</span></pre></div>
|
||||
<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'>
|
||||
@ -222,24 +209,11 @@
|
||||
<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>
|
||||
<p>Cross entropy loss </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">78</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">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">.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">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">.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">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_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">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_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">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">.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">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">.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">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_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">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">.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">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_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">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_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">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">.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">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">.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 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'>
|
||||
@ -247,32 +221,23 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-11'>#</a>
|
||||
</div>
|
||||
<p>Move the parameters based on mapping </p>
|
||||
<p>Dataloader </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">93</span> <span class="n">new_state_dict</span> <span class="o">=</span> <span class="p">{}</span>
|
||||
<span class="lineno">94</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">95</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">96</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 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'>
|
||||
<div class='docs doc-strings'>
|
||||
<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>
|
||||
<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">99</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">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">.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">101</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">102</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">103</span>
|
||||
<span class="lineno">104</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">105</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 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'>
|
||||
@ -280,24 +245,31 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-13'>#</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>
|
||||
<p>Load the huggingface model and get the parameters </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">108</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 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 doc-strings'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-14'>#</a>
|
||||
</div>
|
||||
<h3>Initialize the model, optimizer and dataloader</h3>
|
||||
<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">110</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 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'>
|
||||
@ -305,20 +277,24 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-15'>#</a>
|
||||
</div>
|
||||
<p>Initialize the model </p>
|
||||
<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">115</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">116</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">117</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">118</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">119</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">120</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">121</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">122</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">123</span> <span class="p">)</span>
|
||||
<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">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 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">.pre_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">.pre_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">.post_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">.post_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'>
|
||||
@ -326,11 +302,14 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-16'>#</a>
|
||||
</div>
|
||||
<p>Load pre-trained model weights </p>
|
||||
<p>Move the parameters based on mapping </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">_load_pretrained_weights</span><span class="p">()</span></pre></div>
|
||||
<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'>
|
||||
@ -338,11 +317,17 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-17'>#</a>
|
||||
</div>
|
||||
<p>Initialize the optimizer </p>
|
||||
<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">129</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 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'>
|
||||
@ -350,11 +335,12 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-18'>#</a>
|
||||
</div>
|
||||
<p>Initialize the data loader </p>
|
||||
<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">132</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 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">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'>
|
||||
@ -362,11 +348,11 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-19'>#</a>
|
||||
</div>
|
||||
<h3>Training loop</h3>
|
||||
<h3>Initialize the model, optimizer and dataloader</h3>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">134</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 class="highlight"><pre><span class="lineno">115</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-20'>
|
||||
@ -374,10 +360,20 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-20'>#</a>
|
||||
</div>
|
||||
<p>Initialize the <a href="gpt2.html">GPT2 model</a> </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">139</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 class="highlight"><pre><span class="lineno">120</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">121</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">122</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">123</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">124</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">125</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">126</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">127</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">128</span> <span class="p">)</span>
|
||||
<span class="lineno">129</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-21'>
|
||||
@ -385,13 +381,11 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-21'>#</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>
|
||||
<p>Load pre-trained model weights </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">141</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 class="highlight"><pre><span class="lineno">131</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-22'>
|
||||
@ -399,12 +393,11 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-22'>#</a>
|
||||
</div>
|
||||
<p>Move <code class="highlight"><span></span><span class="n">inputs</span></code>
|
||||
to device </p>
|
||||
<p>Initialize the optimizer </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">143</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 class="highlight"><pre><span class="lineno">134</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-23'>
|
||||
@ -412,23 +405,23 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-23'>#</a>
|
||||
</div>
|
||||
<p>Call the model, with the all but the last token </p>
|
||||
<p>Initialize the data loader </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">145</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 class="highlight"><pre><span class="lineno">137</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-24'>
|
||||
<div class='docs'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-24'>#</a>
|
||||
</div>
|
||||
<p>Get cross entropy loss </p>
|
||||
<h3>Training loop</h3>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><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">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 class="highlight"><pre><span class="lineno">139</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-25'>
|
||||
@ -436,11 +429,10 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-25'>#</a>
|
||||
</div>
|
||||
<p>Make gradients 0 </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">150</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 class="highlight"><pre><span class="lineno">144</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-26'>
|
||||
@ -448,11 +440,13 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-26'>#</a>
|
||||
</div>
|
||||
<p>Compute gradients </p>
|
||||
<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">152</span> <span class="n">loss</span><span class="o">.</span><span class="n">backward</span><span class="p">()</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">146</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-27'>
|
||||
@ -460,11 +454,12 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-27'>#</a>
|
||||
</div>
|
||||
<p>Optimize </p>
|
||||
<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">154</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 class="highlight"><pre><span class="lineno">148</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-28'>
|
||||
@ -472,12 +467,11 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-28'>#</a>
|
||||
</div>
|
||||
<p>Log the loss </p>
|
||||
<p>Call the model, with the all but the last token </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">157</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">158</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add_global_step</span><span class="p">()</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">150</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-29'>
|
||||
@ -485,25 +479,23 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-29'>#</a>
|
||||
</div>
|
||||
<p> </p>
|
||||
<p>Get cross entropy loss </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">160</span> <span class="n">tracker</span><span class="o">.</span><span class="n">new_line</span><span class="p">()</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">152</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-30'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-30'>#</a>
|
||||
</div>
|
||||
<h3>Tiny Shakespeare dataset</h3>
|
||||
<p>It will download from the url if not present</p>
|
||||
<p>Make gradients 0 </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">163</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">164</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 class="highlight"><pre><span class="lineno">155</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-31'>
|
||||
@ -511,20 +503,83 @@
|
||||
<div class='section-link'>
|
||||
<a href='#section-31'>#</a>
|
||||
</div>
|
||||
<p>Compute gradients </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">170</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">171</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">172</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">173</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">174</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">175</span>
|
||||
<span class="lineno">176</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">177</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">178</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">179</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">180</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 class="highlight"><pre><span class="lineno">157</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-32'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-32'>#</a>
|
||||
</div>
|
||||
<p>Optimize </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">step</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>Log the loss </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">162</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">163</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-34'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-34'>#</a>
|
||||
</div>
|
||||
<p> </p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">165</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-35'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-35'>#</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">168</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">169</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-36'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-36'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">175</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">176</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">177</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">178</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">179</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">180</span>
|
||||
<span class="lineno">181</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">182</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">183</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">184</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">185</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'>
|
||||
|
||||
Reference in New Issue
Block a user