mirror of
https://github.com/labmlai/annotated_deep_learning_paper_implementations.git
synced 2025-08-14 09:31:42 +08:00
alibi docs
This commit is contained in:
Varuna Jayasiri
@ -388,10 +388,10 @@ This will keep the accuracy metric stats separate for training and validation.</
|
||||
<div class='section-link'>
|
||||
<a href='#section-27'>#</a>
|
||||
</div>
|
||||
<p>Move data to the device</p>
|
||||
<p>Set training/eval mode</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">112</span> <span class="n">data</span><span class="p">,</span> <span class="n">target</span> <span class="o">=</span> <span class="n">batch</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">),</span> <span class="n">batch</span><span class="p">[</span><span class="mi">1</span><span class="p">]</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">112</span> <span class="bp">self</span><span class="o">.</span><span class="n">model</span><span class="o">.</span><span class="n">train</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">mode</span><span class="o">.</span><span class="n">is_train</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-28'>
|
||||
@ -399,11 +399,10 @@ This will keep the accuracy metric stats separate for training and validation.</
|
||||
<div class='section-link'>
|
||||
<a href='#section-28'>#</a>
|
||||
</div>
|
||||
<p>Update global step (number of tokens processed) when in training mode</p>
|
||||
<p>Move data to the device</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">115</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mode</span><span class="o">.</span><span class="n">is_train</span><span class="p">:</span>
|
||||
<span class="lineno">116</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add_global_step</span><span class="p">(</span><span class="n">data</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">*</span> <span class="n">data</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">1</span><span class="p">])</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">115</span> <span class="n">data</span><span class="p">,</span> <span class="n">target</span> <span class="o">=</span> <span class="n">batch</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">),</span> <span class="n">batch</span><span class="p">[</span><span class="mi">1</span><span class="p">]</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'>
|
||||
@ -411,10 +410,11 @@ This will keep the accuracy metric stats separate for training and validation.</
|
||||
<div class='section-link'>
|
||||
<a href='#section-29'>#</a>
|
||||
</div>
|
||||
<p>Whether to capture model outputs</p>
|
||||
<p>Update global step (number of tokens processed) when in training mode</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">119</span> <span class="k">with</span> <span class="bp">self</span><span class="o">.</span><span class="n">mode</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="n">is_log_activations</span><span class="o">=</span><span class="n">batch_idx</span><span class="o">.</span><span class="n">is_last</span><span class="p">):</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">118</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mode</span><span class="o">.</span><span class="n">is_train</span><span class="p">:</span>
|
||||
<span class="lineno">119</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add_global_step</span><span class="p">(</span><span class="n">data</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">*</span> <span class="n">data</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">1</span><span class="p">])</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-30'>
|
||||
@ -422,12 +422,10 @@ This will keep the accuracy metric stats separate for training and validation.</
|
||||
<div class='section-link'>
|
||||
<a href='#section-30'>#</a>
|
||||
</div>
|
||||
<p>Get model outputs.
|
||||
It’s returning a tuple for states when using RNNs.
|
||||
This is not implemented yet. 😜</p>
|
||||
<p>Whether to capture model outputs</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">123</span> <span class="n">output</span><span class="p">,</span> <span class="o">*</span><span class="n">_</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">data</span><span class="p">)</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">122</span> <span class="k">with</span> <span class="bp">self</span><span class="o">.</span><span class="n">mode</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="n">is_log_activations</span><span class="o">=</span><span class="n">batch_idx</span><span class="o">.</span><span class="n">is_last</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-31'>
|
||||
@ -435,11 +433,12 @@ This is not implemented yet. 😜</p>
|
||||
<div class='section-link'>
|
||||
<a href='#section-31'>#</a>
|
||||
</div>
|
||||
<p>Calculate and log loss</p>
|
||||
<p>Get model outputs.
|
||||
It’s returning a tuple for states when using RNNs.
|
||||
This is not implemented yet. 😜</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">126</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">output</span><span class="p">,</span> <span class="n">target</span><span class="p">)</span>
|
||||
<span class="lineno">127</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add</span><span class="p">(</span><span class="s2">"loss."</span><span class="p">,</span> <span class="n">loss</span><span class="p">)</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">126</span> <span class="n">output</span><span class="p">,</span> <span class="o">*</span><span class="n">_</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">data</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-32'>
|
||||
@ -447,11 +446,11 @@ This is not implemented yet. 😜</p>
|
||||
<div class='section-link'>
|
||||
<a href='#section-32'>#</a>
|
||||
</div>
|
||||
<p>Calculate and log accuracy</p>
|
||||
<p>Calculate and log loss</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">130</span> <span class="bp">self</span><span class="o">.</span><span class="n">accuracy</span><span class="p">(</span><span class="n">output</span><span class="p">,</span> <span class="n">target</span><span class="p">)</span>
|
||||
<span class="lineno">131</span> <span class="bp">self</span><span class="o">.</span><span class="n">accuracy</span><span class="o">.</span><span class="n">track</span><span class="p">()</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">129</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">output</span><span class="p">,</span> <span class="n">target</span><span class="p">)</span>
|
||||
<span class="lineno">130</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add</span><span class="p">(</span><span class="s2">"loss."</span><span class="p">,</span> <span class="n">loss</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-33'>
|
||||
@ -459,10 +458,13 @@ This is not implemented yet. 😜</p>
|
||||
<div class='section-link'>
|
||||
<a href='#section-33'>#</a>
|
||||
</div>
|
||||
<p>Train the model</p>
|
||||
<p>Calculate and log accuracy</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">134</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mode</span><span class="o">.</span><span class="n">is_train</span><span class="p">:</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">133</span> <span class="bp">self</span><span class="o">.</span><span class="n">accuracy</span><span class="p">(</span><span class="n">output</span><span class="p">,</span> <span class="n">target</span><span class="p">)</span>
|
||||
<span class="lineno">134</span> <span class="bp">self</span><span class="o">.</span><span class="n">accuracy</span><span class="o">.</span><span class="n">track</span><span class="p">()</span>
|
||||
<span class="lineno">135</span>
|
||||
<span class="lineno">136</span> <span class="bp">self</span><span class="o">.</span><span class="n">other_metrics</span><span class="p">(</span><span class="n">output</span><span class="p">,</span> <span class="n">target</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-34'>
|
||||
@ -470,10 +472,10 @@ This is not implemented yet. 😜</p>
|
||||
<div class='section-link'>
|
||||
<a href='#section-34'>#</a>
|
||||
</div>
|
||||
<p>Calculate gradients</p>
|
||||
<p>Train the model</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">136</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">139</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mode</span><span class="o">.</span><span class="n">is_train</span><span class="p">:</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-35'>
|
||||
@ -481,10 +483,10 @@ This is not implemented yet. 😜</p>
|
||||
<div class='section-link'>
|
||||
<a href='#section-35'>#</a>
|
||||
</div>
|
||||
<p>Clip gradients</p>
|
||||
<p>Calculate gradients</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">138</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">clip_grad_norm_</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">max_norm</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">grad_norm_clip</span><span class="p">)</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">141</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-36'>
|
||||
@ -492,10 +494,10 @@ This is not implemented yet. 😜</p>
|
||||
<div class='section-link'>
|
||||
<a href='#section-36'>#</a>
|
||||
</div>
|
||||
<p>Take optimizer step</p>
|
||||
<p>Clip gradients</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">140</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">143</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">clip_grad_norm_</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">max_norm</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">grad_norm_clip</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-37'>
|
||||
@ -503,11 +505,10 @@ This is not implemented yet. 😜</p>
|
||||
<div class='section-link'>
|
||||
<a href='#section-37'>#</a>
|
||||
</div>
|
||||
<p>Log the model parameters and gradients on last batch of every epoch</p>
|
||||
<p>Take optimizer step</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">142</span> <span class="k">if</span> <span class="n">batch_idx</span><span class="o">.</span><span class="n">is_last</span><span class="p">:</span>
|
||||
<span class="lineno">143</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add</span><span class="p">(</span><span class="s1">'model'</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">model</span><span class="p">)</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">145</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-38'>
|
||||
@ -515,10 +516,11 @@ This is not implemented yet. 😜</p>
|
||||
<div class='section-link'>
|
||||
<a href='#section-38'>#</a>
|
||||
</div>
|
||||
<p>Clear the gradients</p>
|
||||
<p>Log the model parameters and gradients on last batch of every epoch</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">145</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">147</span> <span class="k">if</span> <span class="n">batch_idx</span><span class="o">.</span><span class="n">is_last</span><span class="p">:</span>
|
||||
<span class="lineno">148</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add</span><span class="p">(</span><span class="s1">'model'</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">model</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-39'>
|
||||
@ -526,32 +528,32 @@ This is not implemented yet. 😜</p>
|
||||
<div class='section-link'>
|
||||
<a href='#section-39'>#</a>
|
||||
</div>
|
||||
<p>Save the tracked metrics</p>
|
||||
<p>Clear the gradients</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">148</span> <span class="n">tracker</span><span class="o">.</span><span class="n">save</span><span class="p">()</span></pre></div>
|
||||
<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>
|
||||
</div>
|
||||
<div class='section' id='section-40'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-40'>#</a>
|
||||
</div>
|
||||
<p>Save the tracked metrics</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">153</span> <span class="n">tracker</span><span class="o">.</span><span class="n">save</span><span class="p">()</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-41'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-41'>#</a>
|
||||
</div>
|
||||
<h3>Sampling function to generate samples periodically while training</h3>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">150</span> <span class="k">def</span> <span class="nf">sample</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-41'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-41'>#</a>
|
||||
</div>
|
||||
<p>Starting prompt</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">156</span> <span class="n">prompt</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">prompt</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">155</span> <span class="k">def</span> <span class="nf">sample</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-42'>
|
||||
@ -559,10 +561,10 @@ This is not implemented yet. 😜</p>
|
||||
<div class='section-link'>
|
||||
<a href='#section-42'>#</a>
|
||||
</div>
|
||||
<p>Collect output for printing</p>
|
||||
<p>Starting prompt</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">158</span> <span class="n">log</span> <span class="o">=</span> <span class="p">[(</span><span class="n">prompt</span><span class="p">,</span> <span class="n">Text</span><span class="o">.</span><span class="n">subtle</span><span class="p">)]</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">161</span> <span class="n">prompt</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">prompt</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-43'>
|
||||
@ -570,10 +572,10 @@ This is not implemented yet. 😜</p>
|
||||
<div class='section-link'>
|
||||
<a href='#section-43'>#</a>
|
||||
</div>
|
||||
<p>Sample 25 tokens</p>
|
||||
<p>Collect output for printing</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">160</span> <span class="k">for</span> <span class="n">i</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">'Sample'</span><span class="p">,</span> <span class="mi">25</span><span class="p">):</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">163</span> <span class="n">log</span> <span class="o">=</span> <span class="p">[(</span><span class="n">prompt</span><span class="p">,</span> <span class="n">Text</span><span class="o">.</span><span class="n">subtle</span><span class="p">)]</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-44'>
|
||||
@ -581,11 +583,10 @@ This is not implemented yet. 😜</p>
|
||||
<div class='section-link'>
|
||||
<a href='#section-44'>#</a>
|
||||
</div>
|
||||
<p>Tokenize the prompt</p>
|
||||
<p>Sample 25 tokens</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">162</span> <span class="n">data</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">text</span><span class="o">.</span><span class="n">text_to_i</span><span class="p">(</span><span class="n">prompt</span><span class="p">)</span><span class="o">.</span><span class="n">unsqueeze</span><span class="p">(</span><span class="o">-</span><span class="mi">1</span><span class="p">)</span>
|
||||
<span class="lineno">163</span> <span class="n">data</span> <span class="o">=</span> <span class="n">data</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">165</span> <span class="k">for</span> <span class="n">i</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">'Sample'</span><span class="p">,</span> <span class="mi">25</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-45'>
|
||||
@ -593,10 +594,11 @@ This is not implemented yet. 😜</p>
|
||||
<div class='section-link'>
|
||||
<a href='#section-45'>#</a>
|
||||
</div>
|
||||
<p>Get the model output</p>
|
||||
<p>Tokenize the prompt</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">165</span> <span class="n">output</span><span class="p">,</span> <span class="o">*</span><span class="n">_</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">data</span><span class="p">)</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">167</span> <span class="n">data</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">text</span><span class="o">.</span><span class="n">text_to_i</span><span class="p">(</span><span class="n">prompt</span><span class="p">)</span><span class="o">.</span><span class="n">unsqueeze</span><span class="p">(</span><span class="o">-</span><span class="mi">1</span><span class="p">)</span>
|
||||
<span class="lineno">168</span> <span class="n">data</span> <span class="o">=</span> <span class="n">data</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-46'>
|
||||
@ -604,10 +606,10 @@ This is not implemented yet. 😜</p>
|
||||
<div class='section-link'>
|
||||
<a href='#section-46'>#</a>
|
||||
</div>
|
||||
<p>Get the model prediction (greedy)</p>
|
||||
<p>Get the model output</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">167</span> <span class="n">output</span> <span class="o">=</span> <span class="n">output</span><span class="o">.</span><span class="n">argmax</span><span class="p">(</span><span class="n">dim</span><span class="o">=-</span><span class="mi">1</span><span class="p">)</span><span class="o">.</span><span class="n">squeeze</span><span class="p">()</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">170</span> <span class="n">output</span><span class="p">,</span> <span class="o">*</span><span class="n">_</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">data</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-47'>
|
||||
@ -615,10 +617,10 @@ This is not implemented yet. 😜</p>
|
||||
<div class='section-link'>
|
||||
<a href='#section-47'>#</a>
|
||||
</div>
|
||||
<p>Add the prediction to prompt</p>
|
||||
<p>Get the model prediction (greedy)</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">169</span> <span class="n">prompt</span> <span class="o">+=</span> <span class="bp">self</span><span class="o">.</span><span class="n">prompt_separator</span> <span class="o">+</span> <span class="bp">self</span><span class="o">.</span><span class="n">text</span><span class="o">.</span><span class="n">itos</span><span class="p">[</span><span class="n">output</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">172</span> <span class="n">output</span> <span class="o">=</span> <span class="n">output</span><span class="o">.</span><span class="n">argmax</span><span class="p">(</span><span class="n">dim</span><span class="o">=-</span><span class="mi">1</span><span class="p">)</span><span class="o">.</span><span class="n">squeeze</span><span class="p">()</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-48'>
|
||||
@ -626,10 +628,10 @@ This is not implemented yet. 😜</p>
|
||||
<div class='section-link'>
|
||||
<a href='#section-48'>#</a>
|
||||
</div>
|
||||
<p>Add the prediction for logging</p>
|
||||
<p>Add the prediction to prompt</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">171</span> <span class="n">log</span> <span class="o">+=</span> <span class="p">[(</span><span class="bp">self</span><span class="o">.</span><span class="n">prompt_separator</span> <span class="o">+</span> <span class="bp">self</span><span class="o">.</span><span class="n">text</span><span class="o">.</span><span class="n">itos</span><span class="p">[</span><span class="n">output</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">]],</span> <span class="n">Text</span><span class="o">.</span><span class="n">value</span><span class="p">)]</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">174</span> <span class="n">prompt</span> <span class="o">+=</span> <span class="bp">self</span><span class="o">.</span><span class="n">prompt_separator</span> <span class="o">+</span> <span class="bp">self</span><span class="o">.</span><span class="n">text</span><span class="o">.</span><span class="n">itos</span><span class="p">[</span><span class="n">output</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-49'>
|
||||
@ -637,67 +639,78 @@ This is not implemented yet. 😜</p>
|
||||
<div class='section-link'>
|
||||
<a href='#section-49'>#</a>
|
||||
</div>
|
||||
<p>Print the sampled output</p>
|
||||
<p>Add the prediction for logging</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">174</span> <span class="n">logger</span><span class="o">.</span><span class="n">log</span><span class="p">(</span><span class="n">log</span><span class="p">)</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">176</span> <span class="n">log</span> <span class="o">+=</span> <span class="p">[(</span><span class="bp">self</span><span class="o">.</span><span class="n">prompt_separator</span> <span class="o">+</span> <span class="bp">self</span><span class="o">.</span><span class="n">text</span><span class="o">.</span><span class="n">itos</span><span class="p">[</span><span class="n">output</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">]],</span> <span class="n">Text</span><span class="o">.</span><span class="n">value</span><span class="p">)]</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-50'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-50'>#</a>
|
||||
</div>
|
||||
<p>Print the sampled output</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">179</span> <span class="n">logger</span><span class="o">.</span><span class="n">log</span><span class="p">(</span><span class="n">log</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-51'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-51'>#</a>
|
||||
</div>
|
||||
<h3>Default <a href="../optimizers/configs.html">optimizer configurations</a></h3>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">177</span><span class="nd">@option</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</span><span class="o">.</span><span class="n">optimizer</span><span class="p">)</span>
|
||||
<span class="lineno">178</span><span class="k">def</span> <span class="nf">_optimizer</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">NLPAutoRegressionConfigs</span><span class="p">):</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">182</span><span class="nd">@option</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</span><span class="o">.</span><span class="n">optimizer</span><span class="p">)</span>
|
||||
<span class="lineno">183</span><span class="k">def</span> <span class="nf">_optimizer</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">NLPAutoRegressionConfigs</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-51'>
|
||||
<div class='section' id='section-52'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-51'>#</a>
|
||||
<a href='#section-52'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">183</span> <span class="n">optimizer</span> <span class="o">=</span> <span class="n">OptimizerConfigs</span><span class="p">()</span>
|
||||
<span class="lineno">184</span> <span class="n">optimizer</span><span class="o">.</span><span class="n">parameters</span> <span class="o">=</span> <span class="n">c</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="lineno">185</span> <span class="n">optimizer</span><span class="o">.</span><span class="n">optimizer</span> <span class="o">=</span> <span class="s1">'Adam'</span>
|
||||
<span class="lineno">186</span> <span class="n">optimizer</span><span class="o">.</span><span class="n">d_model</span> <span class="o">=</span> <span class="n">c</span><span class="o">.</span><span class="n">d_model</span>
|
||||
<span class="lineno">187</span>
|
||||
<span class="lineno">188</span> <span class="k">return</span> <span class="n">optimizer</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">188</span> <span class="n">optimizer</span> <span class="o">=</span> <span class="n">OptimizerConfigs</span><span class="p">()</span>
|
||||
<span class="lineno">189</span> <span class="n">optimizer</span><span class="o">.</span><span class="n">parameters</span> <span class="o">=</span> <span class="n">c</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="lineno">190</span> <span class="n">optimizer</span><span class="o">.</span><span class="n">optimizer</span> <span class="o">=</span> <span class="s1">'Adam'</span>
|
||||
<span class="lineno">191</span> <span class="n">optimizer</span><span class="o">.</span><span class="n">d_model</span> <span class="o">=</span> <span class="n">c</span><span class="o">.</span><span class="n">d_model</span>
|
||||
<span class="lineno">192</span>
|
||||
<span class="lineno">193</span> <span class="k">return</span> <span class="n">optimizer</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-52'>
|
||||
<div class='section' id='section-53'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-52'>#</a>
|
||||
<a href='#section-53'>#</a>
|
||||
</div>
|
||||
<p>Get number of tokens</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">191</span><span class="nd">@option</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</span><span class="o">.</span><span class="n">n_tokens</span><span class="p">)</span>
|
||||
<span class="lineno">192</span><span class="k">def</span> <span class="nf">_n_tokens</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">NLPAutoRegressionConfigs</span><span class="p">):</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">196</span><span class="nd">@option</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</span><span class="o">.</span><span class="n">n_tokens</span><span class="p">)</span>
|
||||
<span class="lineno">197</span><span class="k">def</span> <span class="nf">_n_tokens</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">NLPAutoRegressionConfigs</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-53'>
|
||||
<div class='section' id='section-54'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-53'>#</a>
|
||||
<a href='#section-54'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">196</span> <span class="k">return</span> <span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="o">.</span><span class="n">n_tokens</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">201</span> <span class="k">return</span> <span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="o">.</span><span class="n">n_tokens</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-54'>
|
||||
<div class='section' id='section-55'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-54'>#</a>
|
||||
<a href='#section-55'>#</a>
|
||||
</div>
|
||||
<h3>Basic english tokenizer</h3>
|
||||
<p>We use character level tokenizer in this experiment.
|
||||
@ -708,168 +721,157 @@ You can switch by setting,</p>
|
||||
<p>as the configurations dictionary when starting the experiment.</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">199</span><span class="nd">@option</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</span><span class="o">.</span><span class="n">tokenizer</span><span class="p">)</span>
|
||||
<span class="lineno">200</span><span class="k">def</span> <span class="nf">basic_english</span><span class="p">():</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">204</span><span class="nd">@option</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</span><span class="o">.</span><span class="n">tokenizer</span><span class="p">)</span>
|
||||
<span class="lineno">205</span><span class="k">def</span> <span class="nf">basic_english</span><span class="p">():</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-55'>
|
||||
<div class='section' id='section-56'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-55'>#</a>
|
||||
<a href='#section-56'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">214</span> <span class="kn">from</span> <span class="nn">torchtext.data</span> <span class="kn">import</span> <span class="n">get_tokenizer</span>
|
||||
<span class="lineno">215</span> <span class="k">return</span> <span class="n">get_tokenizer</span><span class="p">(</span><span class="s1">'basic_english'</span><span class="p">)</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">219</span> <span class="kn">from</span> <span class="nn">torchtext.data</span> <span class="kn">import</span> <span class="n">get_tokenizer</span>
|
||||
<span class="lineno">220</span> <span class="k">return</span> <span class="n">get_tokenizer</span><span class="p">(</span><span class="s1">'basic_english'</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-56'>
|
||||
<div class='section' id='section-57'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-56'>#</a>
|
||||
<a href='#section-57'>#</a>
|
||||
</div>
|
||||
<h3>Character level tokenizer</h3>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">218</span><span class="k">def</span> <span class="nf">character_tokenizer</span><span class="p">(</span><span class="n">x</span><span class="p">:</span> <span class="nb">str</span><span class="p">):</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">223</span><span class="k">def</span> <span class="nf">character_tokenizer</span><span class="p">(</span><span class="n">x</span><span class="p">:</span> <span class="nb">str</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-57'>
|
||||
<div class='section' id='section-58'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-57'>#</a>
|
||||
<a href='#section-58'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">222</span> <span class="k">return</span> <span class="nb">list</span><span class="p">(</span><span class="n">x</span><span class="p">)</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">227</span> <span class="k">return</span> <span class="nb">list</span><span class="p">(</span><span class="n">x</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-58'>
|
||||
<div class='section' id='section-59'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-58'>#</a>
|
||||
<a href='#section-59'>#</a>
|
||||
</div>
|
||||
<h3>Character level tokenizer configuration</h3>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">225</span><span class="nd">@option</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</span><span class="o">.</span><span class="n">tokenizer</span><span class="p">)</span>
|
||||
<span class="lineno">226</span><span class="k">def</span> <span class="nf">character</span><span class="p">():</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">230</span><span class="nd">@option</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</span><span class="o">.</span><span class="n">tokenizer</span><span class="p">)</span>
|
||||
<span class="lineno">231</span><span class="k">def</span> <span class="nf">character</span><span class="p">():</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-59'>
|
||||
<div class='section' id='section-60'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-59'>#</a>
|
||||
<a href='#section-60'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">230</span> <span class="k">return</span> <span class="n">character_tokenizer</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">235</span> <span class="k">return</span> <span class="n">character_tokenizer</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-60'>
|
||||
<div class='section' id='section-61'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-60'>#</a>
|
||||
<a href='#section-61'>#</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">233</span><span class="nd">@option</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</span><span class="o">.</span><span class="n">text</span><span class="p">)</span>
|
||||
<span class="lineno">234</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">NLPAutoRegressionConfigs</span><span class="p">):</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">238</span><span class="nd">@option</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</span><span class="o">.</span><span class="n">text</span><span class="p">)</span>
|
||||
<span class="lineno">239</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">NLPAutoRegressionConfigs</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-61'>
|
||||
<div class='section' id='section-62'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-61'>#</a>
|
||||
<a href='#section-62'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">240</span> <span class="k">return</span> <span class="n">TextFileDataset</span><span class="p">(</span>
|
||||
<span class="lineno">241</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="p">,</span>
|
||||
<span class="lineno">242</span> <span class="n">c</span><span class="o">.</span><span class="n">tokenizer</span><span class="p">,</span>
|
||||
<span class="lineno">243</span> <span class="n">url</span><span class="o">=</span><span class="s1">'https://raw.githubusercontent.com/karpathy/char-rnn/master/data/tinyshakespeare/input.txt'</span><span class="p">)</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">245</span> <span class="k">return</span> <span class="n">TextFileDataset</span><span class="p">(</span>
|
||||
<span class="lineno">246</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="p">,</span>
|
||||
<span class="lineno">247</span> <span class="n">c</span><span class="o">.</span><span class="n">tokenizer</span><span class="p">,</span>
|
||||
<span class="lineno">248</span> <span class="n">url</span><span class="o">=</span><span class="s1">'https://raw.githubusercontent.com/karpathy/char-rnn/master/data/tinyshakespeare/input.txt'</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-62'>
|
||||
<div class='section' id='section-63'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-62'>#</a>
|
||||
<a href='#section-63'>#</a>
|
||||
</div>
|
||||
<h3>Sequential training data loader</h3>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">246</span><span class="nd">@option</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</span><span class="o">.</span><span class="n">train_loader</span><span class="p">)</span>
|
||||
<span class="lineno">247</span><span class="k">def</span> <span class="nf">sequential_train_loader</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">NLPAutoRegressionConfigs</span><span class="p">):</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">251</span><span class="nd">@option</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</span><span class="o">.</span><span class="n">train_loader</span><span class="p">)</span>
|
||||
<span class="lineno">252</span><span class="k">def</span> <span class="nf">sequential_train_loader</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">NLPAutoRegressionConfigs</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-63'>
|
||||
<div class='section' id='section-64'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-63'>#</a>
|
||||
<a href='#section-64'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">251</span> <span class="k">return</span> <span class="n">SequentialDataLoader</span><span class="p">(</span><span class="n">text</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="o">.</span><span class="n">train</span><span class="p">,</span>
|
||||
<span class="lineno">252</span> <span class="n">dataset</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="p">,</span>
|
||||
<span class="lineno">253</span> <span class="n">batch_size</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">batch_size</span><span class="p">,</span>
|
||||
<span class="lineno">254</span> <span class="n">seq_len</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">seq_len</span><span class="p">)</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">256</span> <span class="k">return</span> <span class="n">SequentialDataLoader</span><span class="p">(</span><span class="n">text</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="o">.</span><span class="n">train</span><span class="p">,</span>
|
||||
<span class="lineno">257</span> <span class="n">dataset</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="p">,</span>
|
||||
<span class="lineno">258</span> <span class="n">batch_size</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">batch_size</span><span class="p">,</span>
|
||||
<span class="lineno">259</span> <span class="n">seq_len</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">seq_len</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-64'>
|
||||
<div class='section' id='section-65'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-64'>#</a>
|
||||
<a href='#section-65'>#</a>
|
||||
</div>
|
||||
<h3>Sequential validation data loader</h3>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">257</span><span class="nd">@option</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</span><span class="o">.</span><span class="n">valid_loader</span><span class="p">)</span>
|
||||
<span class="lineno">258</span><span class="k">def</span> <span class="nf">sequential_valid_loader</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">NLPAutoRegressionConfigs</span><span class="p">):</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">262</span><span class="nd">@option</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</span><span class="o">.</span><span class="n">valid_loader</span><span class="p">)</span>
|
||||
<span class="lineno">263</span><span class="k">def</span> <span class="nf">sequential_valid_loader</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">NLPAutoRegressionConfigs</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-65'>
|
||||
<div class='section' id='section-66'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-65'>#</a>
|
||||
<a href='#section-66'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">262</span> <span class="k">return</span> <span class="n">SequentialDataLoader</span><span class="p">(</span><span class="n">text</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="o">.</span><span class="n">valid</span><span class="p">,</span>
|
||||
<span class="lineno">263</span> <span class="n">dataset</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="p">,</span>
|
||||
<span class="lineno">264</span> <span class="n">batch_size</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">batch_size</span><span class="p">,</span>
|
||||
<span class="lineno">265</span> <span class="n">seq_len</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">seq_len</span><span class="p">)</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">267</span> <span class="k">return</span> <span class="n">SequentialDataLoader</span><span class="p">(</span><span class="n">text</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="o">.</span><span class="n">valid</span><span class="p">,</span>
|
||||
<span class="lineno">268</span> <span class="n">dataset</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="p">,</span>
|
||||
<span class="lineno">269</span> <span class="n">batch_size</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">batch_size</span><span class="p">,</span>
|
||||
<span class="lineno">270</span> <span class="n">seq_len</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">seq_len</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-66'>
|
||||
<div class='section' id='section-67'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-66'>#</a>
|
||||
<a href='#section-67'>#</a>
|
||||
</div>
|
||||
<h3>Transpose batch</h3>
|
||||
<p><code>DataLoader</code> collects the batches on the first dimension.
|
||||
We need to transpose it to be sequence first.</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">268</span><span class="k">def</span> <span class="nf">transpose_batch</span><span class="p">(</span><span class="n">batch</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-67'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-67'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">276</span> <span class="n">transposed_data</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="nb">zip</span><span class="p">(</span><span class="o">*</span><span class="n">batch</span><span class="p">))</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">273</span><span class="k">def</span> <span class="nf">transpose_batch</span><span class="p">(</span><span class="n">batch</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-68'>
|
||||
@ -877,69 +879,80 @@ We need to transpose it to be sequence first.</p>
|
||||
<div class='section-link'>
|
||||
<a href='#section-68'>#</a>
|
||||
</div>
|
||||
<p>Stack the batch along the second dimension <code>dim=1</code></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">278</span> <span class="n">src</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">stack</span><span class="p">(</span><span class="n">transposed_data</span><span class="p">[</span><span class="mi">0</span><span class="p">],</span> <span class="n">dim</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
|
||||
<span class="lineno">279</span> <span class="n">tgt</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">stack</span><span class="p">(</span><span class="n">transposed_data</span><span class="p">[</span><span class="mi">1</span><span class="p">],</span> <span class="n">dim</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
|
||||
<span class="lineno">280</span>
|
||||
<span class="lineno">281</span> <span class="k">return</span> <span class="n">src</span><span class="p">,</span> <span class="n">tgt</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">281</span> <span class="n">transposed_data</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="nb">zip</span><span class="p">(</span><span class="o">*</span><span class="n">batch</span><span class="p">))</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-69'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-69'>#</a>
|
||||
</div>
|
||||
<p>Stack the batch along the second dimension <code>dim=1</code></p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">283</span> <span class="n">src</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">stack</span><span class="p">(</span><span class="n">transposed_data</span><span class="p">[</span><span class="mi">0</span><span class="p">],</span> <span class="n">dim</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
|
||||
<span class="lineno">284</span> <span class="n">tgt</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">stack</span><span class="p">(</span><span class="n">transposed_data</span><span class="p">[</span><span class="mi">1</span><span class="p">],</span> <span class="n">dim</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>
|
||||
<span class="lineno">285</span>
|
||||
<span class="lineno">286</span> <span class="k">return</span> <span class="n">src</span><span class="p">,</span> <span class="n">tgt</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-70'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-70'>#</a>
|
||||
</div>
|
||||
<h3>Shuffled training data loader</h3>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">284</span><span class="nd">@option</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</span><span class="o">.</span><span class="n">train_loader</span><span class="p">)</span>
|
||||
<span class="lineno">285</span><span class="k">def</span> <span class="nf">shuffled_train_loader</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">NLPAutoRegressionConfigs</span><span class="p">):</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">289</span><span class="nd">@option</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</span><span class="o">.</span><span class="n">train_loader</span><span class="p">)</span>
|
||||
<span class="lineno">290</span><span class="k">def</span> <span class="nf">shuffled_train_loader</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">NLPAutoRegressionConfigs</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-70'>
|
||||
<div class='section' id='section-71'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-70'>#</a>
|
||||
<a href='#section-71'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">289</span> <span class="k">return</span> <span class="n">DataLoader</span><span class="p">(</span><span class="n">SequentialUnBatchedDataset</span><span class="p">(</span><span class="n">text</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="o">.</span><span class="n">train</span><span class="p">,</span>
|
||||
<span class="lineno">290</span> <span class="n">dataset</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="p">,</span>
|
||||
<span class="lineno">291</span> <span class="n">seq_len</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">seq_len</span><span class="p">),</span>
|
||||
<span class="lineno">292</span> <span class="n">batch_size</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">batch_size</span><span class="p">,</span>
|
||||
<span class="lineno">293</span> <span class="n">collate_fn</span><span class="o">=</span><span class="n">transpose_batch</span><span class="p">,</span>
|
||||
<span class="lineno">294</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">294</span> <span class="k">return</span> <span class="n">DataLoader</span><span class="p">(</span><span class="n">SequentialUnBatchedDataset</span><span class="p">(</span><span class="n">text</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="o">.</span><span class="n">train</span><span class="p">,</span>
|
||||
<span class="lineno">295</span> <span class="n">dataset</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="p">,</span>
|
||||
<span class="lineno">296</span> <span class="n">seq_len</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">seq_len</span><span class="p">),</span>
|
||||
<span class="lineno">297</span> <span class="n">batch_size</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">batch_size</span><span class="p">,</span>
|
||||
<span class="lineno">298</span> <span class="n">collate_fn</span><span class="o">=</span><span class="n">transpose_batch</span><span class="p">,</span>
|
||||
<span class="lineno">299</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-71'>
|
||||
<div class='section' id='section-72'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-71'>#</a>
|
||||
<a href='#section-72'>#</a>
|
||||
</div>
|
||||
<h3>Shuffled validation data loader</h3>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">297</span><span class="nd">@option</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</span><span class="o">.</span><span class="n">valid_loader</span><span class="p">)</span>
|
||||
<span class="lineno">298</span><span class="k">def</span> <span class="nf">shuffled_valid_loader</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">NLPAutoRegressionConfigs</span><span class="p">):</span></pre></div>
|
||||
<div class="highlight"><pre><span class="lineno">302</span><span class="nd">@option</span><span class="p">(</span><span class="n">NLPAutoRegressionConfigs</span><span class="o">.</span><span class="n">valid_loader</span><span class="p">)</span>
|
||||
<span class="lineno">303</span><span class="k">def</span> <span class="nf">shuffled_valid_loader</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">NLPAutoRegressionConfigs</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-72'>
|
||||
<div class='section' id='section-73'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-72'>#</a>
|
||||
<a href='#section-73'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">302</span> <span class="k">return</span> <span class="n">DataLoader</span><span class="p">(</span><span class="n">SequentialUnBatchedDataset</span><span class="p">(</span><span class="n">text</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="o">.</span><span class="n">valid</span><span class="p">,</span>
|
||||
<span class="lineno">303</span> <span class="n">dataset</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="p">,</span>
|
||||
<span class="lineno">304</span> <span class="n">seq_len</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">seq_len</span><span class="p">),</span>
|
||||
<span class="lineno">305</span> <span class="n">batch_size</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">batch_size</span><span class="p">,</span>
|
||||
<span class="lineno">306</span> <span class="n">collate_fn</span><span class="o">=</span><span class="n">transpose_batch</span><span class="p">,</span>
|
||||
<span class="lineno">307</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">307</span> <span class="k">return</span> <span class="n">DataLoader</span><span class="p">(</span><span class="n">SequentialUnBatchedDataset</span><span class="p">(</span><span class="n">text</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="o">.</span><span class="n">valid</span><span class="p">,</span>
|
||||
<span class="lineno">308</span> <span class="n">dataset</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="p">,</span>
|
||||
<span class="lineno">309</span> <span class="n">seq_len</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">seq_len</span><span class="p">),</span>
|
||||
<span class="lineno">310</span> <span class="n">batch_size</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">batch_size</span><span class="p">,</span>
|
||||
<span class="lineno">311</span> <span class="n">collate_fn</span><span class="o">=</span><span class="n">transpose_batch</span><span class="p">,</span>
|
||||
<span class="lineno">312</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='footer'>
|
||||
|
||||
@ -36,7 +36,7 @@
|
||||
|
||||
<url>
|
||||
<loc>https://nn.labml.ai/gan/original/index.html</loc>
|
||||
<lastmod>2021-08-19T16:30:00+00:00</lastmod>
|
||||
<lastmod>2021-08-25T16:30:00+00:00</lastmod>
|
||||
<priority>1.00</priority>
|
||||
</url>
|
||||
|
||||
@ -50,7 +50,7 @@
|
||||
|
||||
<url>
|
||||
<loc>https://nn.labml.ai/gan/dcgan/index.html</loc>
|
||||
<lastmod>2021-08-19T16:30:00+00:00</lastmod>
|
||||
<lastmod>2021-08-25T16:30:00+00:00</lastmod>
|
||||
<priority>1.00</priority>
|
||||
</url>
|
||||
|
||||
@ -211,7 +211,7 @@
|
||||
|
||||
<url>
|
||||
<loc>https://nn.labml.ai/experiments/nlp_autoregression.html</loc>
|
||||
<lastmod>2021-08-19T16:30:00+00:00</lastmod>
|
||||
<lastmod>2021-08-27T16:30:00+00:00</lastmod>
|
||||
<priority>1.00</priority>
|
||||
</url>
|
||||
|
||||
@ -449,7 +449,7 @@
|
||||
|
||||
<url>
|
||||
<loc>https://nn.labml.ai/transformers/configs.html</loc>
|
||||
<lastmod>2021-08-17T16:30:00+00:00</lastmod>
|
||||
<lastmod>2021-08-27T16:30:00+00:00</lastmod>
|
||||
<priority>1.00</priority>
|
||||
</url>
|
||||
|
||||
|
||||
605
docs/transformers/alibi/experiment.html
Normal file
605
docs/transformers/alibi/experiment.html
Normal file
@ -0,0 +1,605 @@
|
||||
<!DOCTYPE html>
|
||||
<html>
|
||||
<head>
|
||||
<meta http-equiv="content-type" content="text/html;charset=utf-8"/>
|
||||
<meta name="viewport" content="width=device-width, initial-scale=1.0"/>
|
||||
<meta name="description" content="This experiment trains an Attention with Linear Biases (ALiBi) based model on Tiny Shakespeare dataset."/>
|
||||
|
||||
<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="Attention with Linear Biases (ALiBi) Experiment"/>
|
||||
<meta name="twitter:description" content="This experiment trains an Attention with Linear Biases (ALiBi) based model on Tiny Shakespeare dataset."/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/transformers/alibi/experiment.html"/>
|
||||
<meta property="og:title" content="Attention with Linear Biases (ALiBi) Experiment"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="LabML Neural Networks"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="Attention with Linear Biases (ALiBi) Experiment"/>
|
||||
<meta property="og:description" content="This experiment trains an Attention with Linear Biases (ALiBi) based model on Tiny Shakespeare dataset."/>
|
||||
|
||||
<title>Attention with Linear Biases (ALiBi) Experiment</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../../pylit.css">
|
||||
<link rel="canonical" href="https://nn.labml.ai/transformers/alibi/experiment.html"/>
|
||||
<!-- Global site tag (gtag.js) - Google Analytics -->
|
||||
<script async src="https://www.googletagmanager.com/gtag/js?id=G-4V3HC8HBLH"></script>
|
||||
<script>
|
||||
window.dataLayer = window.dataLayer || [];
|
||||
|
||||
function gtag() {
|
||||
dataLayer.push(arguments);
|
||||
}
|
||||
|
||||
gtag('js', new Date());
|
||||
|
||||
gtag('config', 'G-4V3HC8HBLH');
|
||||
</script>
|
||||
</head>
|
||||
<body>
|
||||
<div id='container'>
|
||||
<div id="background"></div>
|
||||
<div class='section'>
|
||||
<div class='docs'>
|
||||
<p>
|
||||
<a class="parent" href="/">home</a>
|
||||
<a class="parent" href="../index.html">transformers</a>
|
||||
<a class="parent" href="index.html">alibi</a>
|
||||
</p>
|
||||
<p>
|
||||
|
||||
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/transformers/alibi/experiment.py">
|
||||
<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">
|
||||
<img alt="Twitter"
|
||||
src="https://img.shields.io/twitter/follow/labmlai?style=social"
|
||||
style="max-width:100%;"/></a>
|
||||
</p>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-0'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-0'>#</a>
|
||||
</div>
|
||||
<h1><a href="index.html">Attention with Linear Biases (ALiBi)</a> Experiment</h1>
|
||||
<p>This is an annotated PyTorch experiment to train a <a href="index.html">ALiBi model</a>.</p>
|
||||
<p>This is based on
|
||||
<a href="../gpt/index.html">our GPT model</a>.</p>
|
||||
<p><a href="https://app.labml.ai/run/e87bec2a074911ec82cdd1759f10c925"><img alt="View Run" src="https://img.shields.io/badge/labml-experiment-brightgreen" /></a></p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">17</span><span></span><span class="kn">import</span> <span class="nn">torch</span>
|
||||
<span class="lineno">18</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="lineno">19</span>
|
||||
<span class="lineno">20</span><span class="kn">from</span> <span class="nn">labml</span> <span class="kn">import</span> <span class="n">experiment</span><span class="p">,</span> <span class="n">tracker</span>
|
||||
<span class="lineno">21</span><span class="kn">from</span> <span class="nn">labml.configs</span> <span class="kn">import</span> <span class="n">option</span><span class="p">,</span> <span class="n">calculate</span>
|
||||
<span class="lineno">22</span><span class="kn">from</span> <span class="nn">labml_helpers.datasets.text</span> <span class="kn">import</span> <span class="n">SequentialUnBatchedDataset</span>
|
||||
<span class="lineno">23</span><span class="kn">from</span> <span class="nn">labml_nn.transformers.alibi</span> <span class="kn">import</span> <span class="n">AlibiMultiHeadAttention</span>
|
||||
<span class="lineno">24</span><span class="kn">from</span> <span class="nn">labml_nn.experiments.nlp_autoregression</span> <span class="kn">import</span> <span class="n">transpose_batch</span>
|
||||
<span class="lineno">25</span><span class="kn">from</span> <span class="nn">labml_nn.transformers</span> <span class="kn">import</span> <span class="n">TransformerConfigs</span>
|
||||
<span class="lineno">26</span><span class="kn">from</span> <span class="nn">labml_nn.transformers.gpt</span> <span class="kn">import</span> <span class="n">Configs</span> <span class="k">as</span> <span class="n">GPTConfigs</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>Configurations</h2>
|
||||
<p>We extend <a href="../gpt/index.html">GPT configurations</a> and change the attention mechanism.</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">29</span><span class="k">class</span> <span class="nc">Configs</span><span class="p">(</span><span class="n">GPTConfigs</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>
|
||||
<p>ALiBi based transformer (defined below)</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">37</span> <span class="n">transformer</span><span class="p">:</span> <span class="n">TransformerConfigs</span> <span class="o">=</span> <span class="s1">'GPT_ALiBi'</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>Longer validation set</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">39</span> <span class="n">valid_seq_len</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">128</span>
|
||||
<span class="lineno">40</span> <span class="n">valid_loader</span> <span class="o">=</span> <span class="s1">'shuffled_longer_valid_loader'</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-4'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-4'>#</a>
|
||||
</div>
|
||||
<p>Log losses at the initial and final tokens</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">42</span> <span class="k">def</span> <span class="nf">other_metrics</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">output</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">target</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-5'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-5'>#</a>
|
||||
</div>
|
||||
<p>If there are more tokens that the training sequence length (during validation),</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">47</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">seq_len</span> <span class="o"><</span> <span class="n">output</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">0</span><span class="p">]:</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-6'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-6'>#</a>
|
||||
</div>
|
||||
<p>Log the loss at training sequence length</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">49</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add</span><span class="p">(</span><span class="sa">f</span><span class="s1">'loss.</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">seq_len</span> <span class="o">-</span> <span class="mi">1</span><span class="si">}</span><span class="s1">.'</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">loss_func</span><span class="p">(</span><span class="n">output</span><span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">seq_len</span> <span class="o">-</span> <span class="mi">1</span><span class="p">],</span> <span class="n">target</span><span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">seq_len</span> <span class="o">-</span> <span class="mi">1</span><span class="p">]))</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-7'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-7'>#</a>
|
||||
</div>
|
||||
<p>Log the loss at the first token</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">51</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add</span><span class="p">(</span><span class="sa">f</span><span class="s1">'loss.0.'</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">loss_func</span><span class="p">(</span><span class="n">output</span><span class="p">[</span><span class="mi">0</span><span class="p">],</span> <span class="n">target</span><span class="p">[</span><span class="mi">0</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>Log the loss at the final token</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">53</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add</span><span class="p">(</span><span class="sa">f</span><span class="s1">'loss.</span><span class="si">{</span><span class="nb">int</span><span class="p">(</span><span class="n">output</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">0</span><span class="p">])</span> <span class="o">-</span> <span class="mi">1</span><span class="si">}</span><span class="s1">.'</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">loss_func</span><span class="p">(</span><span class="n">output</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span> <span class="n">target</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-9'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-9'>#</a>
|
||||
</div>
|
||||
<p>Create an ALiBi attention module</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">56</span><span class="k">def</span> <span class="nf">_alibi_mha</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">TransformerConfigs</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-10'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-10'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">60</span> <span class="k">return</span> <span class="n">AlibiMultiHeadAttention</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">n_heads</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">d_model</span><span class="p">,</span> <span class="n">dropout_prob</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">dropout</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-11'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-11'>#</a>
|
||||
</div>
|
||||
<p>Set all attention mechanisms to ALiBi</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">64</span><span class="n">calculate</span><span class="p">(</span><span class="n">TransformerConfigs</span><span class="o">.</span><span class="n">encoder_attn</span><span class="p">,</span> <span class="s1">'alibi_mha'</span><span class="p">,</span> <span class="n">_alibi_mha</span><span class="p">)</span>
|
||||
<span class="lineno">65</span><span class="n">calculate</span><span class="p">(</span><span class="n">TransformerConfigs</span><span class="o">.</span><span class="n">decoder_attn</span><span class="p">,</span> <span class="s1">'alibi_mha'</span><span class="p">,</span> <span class="n">_alibi_mha</span><span class="p">)</span>
|
||||
<span class="lineno">66</span><span class="n">calculate</span><span class="p">(</span><span class="n">TransformerConfigs</span><span class="o">.</span><span class="n">decoder_mem_attn</span><span class="p">,</span> <span class="s1">'alibi_mha'</span><span class="p">,</span> <span class="n">_alibi_mha</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-12'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-12'>#</a>
|
||||
</div>
|
||||
<p>Shuffled validation data loader with <code>valid_seq_len</code> sequence length</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">69</span><span class="nd">@option</span><span class="p">(</span><span class="n">Configs</span><span class="o">.</span><span class="n">valid_loader</span><span class="p">)</span>
|
||||
<span class="lineno">70</span><span class="k">def</span> <span class="nf">shuffled_longer_valid_loader</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">Configs</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-13'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-13'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">74</span> <span class="k">return</span> <span class="n">DataLoader</span><span class="p">(</span><span class="n">SequentialUnBatchedDataset</span><span class="p">(</span><span class="n">text</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="o">.</span><span class="n">valid</span><span class="p">,</span>
|
||||
<span class="lineno">75</span> <span class="n">dataset</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">text</span><span class="p">,</span>
|
||||
<span class="lineno">76</span> <span class="n">seq_len</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">valid_seq_len</span><span class="p">),</span>
|
||||
<span class="lineno">77</span> <span class="n">batch_size</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">batch_size</span><span class="p">,</span>
|
||||
<span class="lineno">78</span> <span class="n">collate_fn</span><span class="o">=</span><span class="n">transpose_batch</span><span class="p">,</span>
|
||||
<span class="lineno">79</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-14'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-14'>#</a>
|
||||
</div>
|
||||
<h3>ALiBi based Transformer configurations</h3>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">82</span><span class="nd">@option</span><span class="p">(</span><span class="n">Configs</span><span class="o">.</span><span class="n">transformer</span><span class="p">,</span> <span class="s1">'GPT_ALiBi'</span><span class="p">)</span>
|
||||
<span class="lineno">83</span><span class="k">def</span> <span class="nf">_transformer_configs</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">Configs</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-15'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-15'>#</a>
|
||||
</div>
|
||||
<p>We use our
|
||||
<a href="../configs.html#TransformerConfigs">configurable transformer implementation</a></p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">90</span> <span class="n">conf</span> <span class="o">=</span> <span class="n">TransformerConfigs</span><span class="p">()</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-16'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-16'>#</a>
|
||||
</div>
|
||||
<p>Set the vocabulary sizes for embeddings and generating logits</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">92</span> <span class="n">conf</span><span class="o">.</span><span class="n">n_src_vocab</span> <span class="o">=</span> <span class="n">c</span><span class="o">.</span><span class="n">n_tokens</span>
|
||||
<span class="lineno">93</span> <span class="n">conf</span><span class="o">.</span><span class="n">n_tgt_vocab</span> <span class="o">=</span> <span class="n">c</span><span class="o">.</span><span class="n">n_tokens</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 uses GELU activation for position wise feedforward</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">95</span> <span class="n">conf</span><span class="o">.</span><span class="n">ffn</span><span class="o">.</span><span class="n">activation</span> <span class="o">=</span> <span class="s1">'GELU'</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>ALiBi doesn’t use positional embeddings</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">98</span> <span class="n">conf</span><span class="o">.</span><span class="n">src_embed</span> <span class="o">=</span> <span class="s1">'no_pos'</span>
|
||||
<span class="lineno">99</span> <span class="n">conf</span><span class="o">.</span><span class="n">tgt_embed</span> <span class="o">=</span> <span class="s1">'no_pos'</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>Set all attention mechanisms to ALiBi</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">102</span> <span class="n">conf</span><span class="o">.</span><span class="n">encoder_attn</span> <span class="o">=</span> <span class="s1">'alibi_mha'</span>
|
||||
<span class="lineno">103</span> <span class="n">conf</span><span class="o">.</span><span class="n">decoder_attn</span> <span class="o">=</span> <span class="s1">'alibi_mha'</span>
|
||||
<span class="lineno">104</span> <span class="n">conf</span><span class="o">.</span><span class="n">decoder_mem_attn</span> <span class="o">=</span> <span class="s1">'alibi_mha'</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-20'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-20'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">107</span> <span class="k">return</span> <span class="n">conf</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-21'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-21'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">110</span><span class="k">def</span> <span class="nf">main</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>Create experiment</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">112</span> <span class="n">experiment</span><span class="o">.</span><span class="n">create</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s2">"gpt_alibi"</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>Create configs</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">114</span> <span class="n">conf</span> <span class="o">=</span> <span class="n">Configs</span><span class="p">()</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-24'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-24'>#</a>
|
||||
</div>
|
||||
<p>Override configurations</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">116</span> <span class="n">experiment</span><span class="o">.</span><span class="n">configs</span><span class="p">(</span><span class="n">conf</span><span class="p">,</span> <span class="p">{</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-25'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-25'>#</a>
|
||||
</div>
|
||||
<p>Use character level tokenizer</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">118</span> <span class="s1">'tokenizer'</span><span class="p">:</span> <span class="s1">'character'</span><span class="p">,</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-26'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-26'>#</a>
|
||||
</div>
|
||||
<p>Prompt separator is blank</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">120</span> <span class="s1">'prompt_separator'</span><span class="p">:</span> <span class="s1">''</span><span class="p">,</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-27'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-27'>#</a>
|
||||
</div>
|
||||
<p>Starting prompt for sampling</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">122</span> <span class="s1">'prompt'</span><span class="p">:</span> <span class="s1">'It is '</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>Use Tiny Shakespeare dataset</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">124</span> <span class="s1">'text'</span><span class="p">:</span> <span class="s1">'tiny_shakespeare'</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>‘text’: ‘tiny_shakespeare_no_split’,</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-30'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-30'>#</a>
|
||||
</div>
|
||||
<p>Use a context size of $128$</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">128</span> <span class="s1">'seq_len'</span><span class="p">:</span> <span class="mi">64</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>Use a context size of $128$</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">130</span> <span class="s1">'valid_seq_len'</span><span class="p">:</span> <span class="mi">80</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>Train for $32$ epochs</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">132</span> <span class="s1">'epochs'</span><span class="p">:</span> <span class="mi">128</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>Batch size $128$</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">134</span> <span class="s1">'batch_size'</span><span class="p">:</span> <span class="mi">128</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>Switch between training and validation for $10$ times
|
||||
per epoch</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">137</span> <span class="s1">'inner_iterations'</span><span class="p">:</span> <span class="mi">10</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>Transformer configurations</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">140</span> <span class="s1">'transformer.d_model'</span><span class="p">:</span> <span class="mi">128</span><span class="p">,</span>
|
||||
<span class="lineno">141</span> <span class="s1">'transformer.ffn.d_ff'</span><span class="p">:</span> <span class="mi">512</span><span class="p">,</span>
|
||||
<span class="lineno">142</span> <span class="s1">'transformer.n_heads'</span><span class="p">:</span> <span class="mi">8</span><span class="p">,</span>
|
||||
<span class="lineno">143</span> <span class="s1">'transformer.n_layers'</span><span class="p">:</span> <span class="mi">4</span><span class="p">,</span>
|
||||
<span class="lineno">144</span> <span class="s1">'transformer.dropout'</span><span class="p">:</span> <span class="mf">0.1</span><span class="p">,</span>
|
||||
<span class="lineno">145</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>
|
||||
<p>Set models for saving and loading</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">148</span> <span class="n">experiment</span><span class="o">.</span><span class="n">add_pytorch_models</span><span class="p">({</span><span class="s1">'model'</span><span class="p">:</span> <span class="n">conf</span><span class="o">.</span><span class="n">model</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>
|
||||
<p>Start the experiment</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">151</span> <span class="k">with</span> <span class="n">experiment</span><span class="o">.</span><span class="n">start</span><span class="p">():</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-38'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-38'>#</a>
|
||||
</div>
|
||||
<p>Run training</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">153</span> <span class="n">conf</span><span class="o">.</span><span class="n">run</span><span class="p">()</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-39'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-39'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">157</span><span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s1">'__main__'</span><span class="p">:</span>
|
||||
<span class="lineno">158</span> <span class="n">main</span><span class="p">()</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='footer'>
|
||||
<a href="https://papers.labml.ai">Trending Research Papers</a>
|
||||
<a href="https://labml.ai">labml.ai</a>
|
||||
</div>
|
||||
</div>
|
||||
<script src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.4/MathJax.js?config=TeX-AMS_HTML">
|
||||
</script>
|
||||
<!-- MathJax configuration -->
|
||||
<script type="text/x-mathjax-config">
|
||||
MathJax.Hub.Config({
|
||||
tex2jax: {
|
||||
inlineMath: [ ['$','$'] ],
|
||||
displayMath: [ ['$$','$$'] ],
|
||||
processEscapes: true,
|
||||
processEnvironments: true
|
||||
},
|
||||
// Center justify equations in code and markdown cells. Elsewhere
|
||||
// we use CSS to left justify single line equations in code cells.
|
||||
displayAlign: 'center',
|
||||
"HTML-CSS": { fonts: ["TeX"] }
|
||||
});
|
||||
|
||||
</script>
|
||||
<script>
|
||||
function handleImages() {
|
||||
var images = document.querySelectorAll('p>img')
|
||||
|
||||
console.log(images);
|
||||
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>
|
||||
383
docs/transformers/alibi/index.html
Normal file
383
docs/transformers/alibi/index.html
Normal file
@ -0,0 +1,383 @@
|
||||
<!DOCTYPE html>
|
||||
<html>
|
||||
<head>
|
||||
<meta http-equiv="content-type" content="text/html;charset=utf-8"/>
|
||||
<meta name="viewport" content="width=device-width, initial-scale=1.0"/>
|
||||
<meta name="description" content="Documented implementation with explanations of Attention with Linear Biases (ALiBi)"/>
|
||||
|
||||
<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="Attention with Linear Biases (ALiBi)"/>
|
||||
<meta name="twitter:description" content="Documented implementation with explanations of Attention with Linear Biases (ALiBi)"/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/transformers/alibi/index.html"/>
|
||||
<meta property="og:title" content="Attention with Linear Biases (ALiBi)"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="LabML Neural Networks"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="Attention with Linear Biases (ALiBi)"/>
|
||||
<meta property="og:description" content="Documented implementation with explanations of Attention with Linear Biases (ALiBi)"/>
|
||||
|
||||
<title>Attention with Linear Biases (ALiBi)</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../../pylit.css">
|
||||
<link rel="canonical" href="https://nn.labml.ai/transformers/alibi/index.html"/>
|
||||
<!-- Global site tag (gtag.js) - Google Analytics -->
|
||||
<script async src="https://www.googletagmanager.com/gtag/js?id=G-4V3HC8HBLH"></script>
|
||||
<script>
|
||||
window.dataLayer = window.dataLayer || [];
|
||||
|
||||
function gtag() {
|
||||
dataLayer.push(arguments);
|
||||
}
|
||||
|
||||
gtag('js', new Date());
|
||||
|
||||
gtag('config', 'G-4V3HC8HBLH');
|
||||
</script>
|
||||
</head>
|
||||
<body>
|
||||
<div id='container'>
|
||||
<div id="background"></div>
|
||||
<div class='section'>
|
||||
<div class='docs'>
|
||||
<p>
|
||||
<a class="parent" href="/">home</a>
|
||||
<a class="parent" href="../index.html">transformers</a>
|
||||
<a class="parent" href="index.html">alibi</a>
|
||||
</p>
|
||||
<p>
|
||||
|
||||
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/transformers/alibi/__init__.py">
|
||||
<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">
|
||||
<img alt="Twitter"
|
||||
src="https://img.shields.io/twitter/follow/labmlai?style=social"
|
||||
style="max-width:100%;"/></a>
|
||||
</p>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-0'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-0'>#</a>
|
||||
</div>
|
||||
<h1>Attention with Linear Biases (ALiBi)</h1>
|
||||
<p>This is an implementation of Attention with Linear Biases (ALiBi) from the paper
|
||||
Train Short, Test Long: Attention with Linear Biases Enables Input Length Extrapolation
|
||||
<a href="https://ofir.io/train_short_test_long.pdf">(pdf)</a>.</p>
|
||||
<p>This replaces positional encodings with biases added to attention scores (attention logits, before the softmax).
|
||||
This is a relative scheme tested on autoregressive tasks, and the bias is higher for closeby tokens
|
||||
and lower for far-away tokens.
|
||||
The biases decrease linearly in the log scale (because it’s before the softmax) and each head has a different slope.</p>
|
||||
<p>Here’s the attention formula for $i$-th token,</p>
|
||||
<p>
|
||||
<script type="math/tex; mode=display">\begin{align}
|
||||
\mathbf{a}_i
|
||||
&= \text{softmax} \bigg( \mathbf{q}_i \mathbf{K}^\top + m \cdot \big[-(i-1), \dots, 1, 0 \big] \bigg) \\
|
||||
&= \text{softmax} \bigg( \mathbf{q}_i \mathbf{K}^\top + m \cdot \big[0, 1, \dots, (i - 1) \big] \bigg)
|
||||
\end{align}</script>
|
||||
</p>
|
||||
<p>where $\mathbf{q}_i \in \mathbb{R}^d$ is the query of the $i$-th token, $K \in \mathbb{R}^{i \times d}$ are the keys
|
||||
up to $i$, and $d$ the number of features per head.
|
||||
Note that the above equality halts because $\text{softmax}$ is invariant to translations
|
||||
(you can add any constant to all elements without changing the result).</p>
|
||||
<p>Here is <a href="experiment.html">the training code</a> for a ALiBi model.</p>
|
||||
<p><a href="https://app.labml.ai/run/e87bec2a074911ec82cdd1759f10c925"><img alt="View Run" src="https://img.shields.io/badge/labml-experiment-brightgreen" /></a></p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">36</span><span></span><span class="kn">import</span> <span class="nn">math</span>
|
||||
<span class="lineno">37</span>
|
||||
<span class="lineno">38</span><span class="kn">import</span> <span class="nn">torch</span>
|
||||
<span class="lineno">39</span><span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">nn</span>
|
||||
<span class="lineno">40</span>
|
||||
<span class="lineno">41</span><span class="kn">from</span> <span class="nn">labml.logger</span> <span class="kn">import</span> <span class="n">inspect</span>
|
||||
<span class="lineno">42</span><span class="kn">from</span> <span class="nn">labml_nn.transformers.mha</span> <span class="kn">import</span> <span class="n">MultiHeadAttention</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>Get head-specific slope $m$ for each head</h2>
|
||||
<ul>
|
||||
<li><code>n_heads</code> is the number of heads in the attention layer $n$</li>
|
||||
</ul>
|
||||
<p>The slope for first head is</p>
|
||||
<p>
|
||||
<script type="math/tex; mode=display">2^{-2^{-(\log_2 n) - 3}}</script>
|
||||
</p>
|
||||
<p>The slopes for the rest of the heads are in a geometric series with a ratio same as above.</p>
|
||||
<p>For instance when the number of heads is $8$ the slopes are
|
||||
<script type="math/tex; mode=display">\frac{1}{2^1}, \frac{1}{2^2}, \dots, \frac{1}{2^8}</script>
|
||||
</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">45</span><span class="k">def</span> <span class="nf">get_slopes</span><span class="p">(</span><span class="n">n_heads</span><span class="p">:</span> <span class="nb">int</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>
|
||||
<p>
|
||||
<script type="math/tex; mode=display">2^{-2^{-(\log_2 n) - 3}}</script>
|
||||
</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">62</span> <span class="n">s</span> <span class="o">=</span> <span class="p">(</span><span class="mi">2</span> <span class="o">**</span> <span class="p">(</span><span class="o">-</span><span class="mi">2</span> <span class="o">**</span> <span class="o">-</span><span class="p">(</span><span class="n">math</span><span class="o">.</span><span class="n">log2</span><span class="p">(</span><span class="n">n_heads</span><span class="p">)</span> <span class="o">-</span> <span class="mi">3</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>The geometric sequence</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">64</span> <span class="k">return</span> <span class="p">[</span><span class="n">s</span> <span class="o">*</span> <span class="p">(</span><span class="n">s</span> <span class="o">**</span> <span class="n">i</span><span class="p">)</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="n">n_heads</span><span class="p">)]</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-4'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-4'>#</a>
|
||||
</div>
|
||||
<h2>Calculate the attention biases matrix</h2>
|
||||
<ul>
|
||||
<li><code>n_heads</code> is the number of heads in the attention layer</li>
|
||||
<li><code>max_len</code> is the maximum sequence length</li>
|
||||
</ul>
|
||||
<p>This returns a matrix of shape <code>[n_heads, max_len]</code> with attention biases.</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">67</span><span class="k">def</span> <span class="nf">get_biases</span><span class="p">(</span><span class="n">n_heads</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">max_len</span><span class="p">:</span> <span class="nb">int</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-5'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-5'>#</a>
|
||||
</div>
|
||||
<p>Get slopes $m$ for each head</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">78</span> <span class="n">slopes</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">get_slopes</span><span class="p">(</span><span class="n">n_heads</span><span class="p">))</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-6'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-6'>#</a>
|
||||
</div>
|
||||
<p>Calculate distances $[0, 1, \dots, N]$</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">80</span> <span class="n">distance</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">arange</span><span class="p">(</span><span class="n">max_len</span><span class="p">)</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">float</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-7'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-7'>#</a>
|
||||
</div>
|
||||
<p>Multiply them pair-wise to get the bias matrix</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">82</span> <span class="k">return</span> <span class="n">distance</span><span class="p">[:,</span> <span class="kc">None</span><span class="p">]</span> <span class="o">*</span> <span class="n">slopes</span><span class="p">[</span><span class="kc">None</span><span class="p">,</span> <span class="p">:]</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-8'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-8'>#</a>
|
||||
</div>
|
||||
<h2>Attention with Linear Biases (ALiBi)</h2>
|
||||
<p>We override <a href="mha.html">Multi-Head Attention</a> module so we only need to
|
||||
write the <code>get_scores</code> method.</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">85</span><span class="k">class</span> <span class="nc">AlibiMultiHeadAttention</span><span class="p">(</span><span class="n">MultiHeadAttention</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-9'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-9'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">93</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">heads</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">d_model</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">dropout_prob</span><span class="p">:</span> <span class="nb">float</span> <span class="o">=</span> <span class="mf">0.1</span><span class="p">,</span> <span class="n">max_len</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">5_000</span><span class="p">):</span>
|
||||
<span class="lineno">94</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">(</span><span class="n">heads</span><span class="p">,</span> <span class="n">d_model</span><span class="p">,</span> <span class="n">dropout_prob</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-10'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-10'>#</a>
|
||||
</div>
|
||||
<p>Pre-calculate the biases</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">97</span> <span class="bp">self</span><span class="o">.</span><span class="n">bias</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">get_biases</span><span class="p">(</span><span class="n">heads</span><span class="p">,</span> <span class="n">max_len</span><span class="p">),</span> <span class="n">requires_grad</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-11'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-11'>#</a>
|
||||
</div>
|
||||
<h3>Calculate attention scores and add attention biases</h3>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">99</span> <span class="k">def</span> <span class="nf">get_scores</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">query</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">key</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-12'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-12'>#</a>
|
||||
</div>
|
||||
<p>Calculate the standard attention score.
|
||||
It has shape <code>[query_seq_len, key_seq_len, batch_size, head]</code></p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">106</span> <span class="n">scores</span> <span class="o">=</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="n">get_scores</span><span class="p">(</span><span class="n">query</span><span class="p">,</span> <span class="n">key</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>Number of keys</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">109</span> <span class="n">key_seq_len</span> <span class="o">=</span> <span class="n">scores</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">1</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>Add the biases to scores.</p>
|
||||
<p>
|
||||
<script type="math/tex; mode=display">\mathbf{q}_i \mathbf{K}^\top + m \cdot \big[0, 1, \dots, (i - 1) \big]</script>
|
||||
</p>
|
||||
<p>Note that we add biases for all keys (not just upto $i$). We can do this since
|
||||
those extra entries will get removed because of the masking later.</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">116</span> <span class="k">return</span> <span class="n">scores</span> <span class="o">+</span> <span class="bp">self</span><span class="o">.</span><span class="n">bias</span><span class="p">[</span><span class="kc">None</span><span class="p">,</span> <span class="p">:</span><span class="n">key_seq_len</span><span class="p">,</span> <span class="kc">None</span><span class="p">,</span> <span class="p">:]</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-15'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-15'>#</a>
|
||||
</div>
|
||||
<p>Simple test function to see the slopes.</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">119</span><span class="k">def</span> <span class="nf">_test_slopes</span><span class="p">():</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-16'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-16'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">123</span> <span class="n">inspect</span><span class="p">(</span><span class="n">get_slopes</span><span class="p">(</span><span class="mi">8</span><span class="p">))</span>
|
||||
<span class="lineno">124</span> <span class="n">inspect</span><span class="p">(</span><span class="n">get_slopes</span><span class="p">(</span><span class="mi">16</span><span class="p">))</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-17'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-17'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">128</span><span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s1">'__main__'</span><span class="p">:</span>
|
||||
<span class="lineno">129</span> <span class="n">_test_slopes</span><span class="p">()</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='footer'>
|
||||
<a href="https://papers.labml.ai">Trending Research Papers</a>
|
||||
<a href="https://labml.ai">labml.ai</a>
|
||||
</div>
|
||||
</div>
|
||||
<script src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.4/MathJax.js?config=TeX-AMS_HTML">
|
||||
</script>
|
||||
<!-- MathJax configuration -->
|
||||
<script type="text/x-mathjax-config">
|
||||
MathJax.Hub.Config({
|
||||
tex2jax: {
|
||||
inlineMath: [ ['$','$'] ],
|
||||
displayMath: [ ['$$','$$'] ],
|
||||
processEscapes: true,
|
||||
processEnvironments: true
|
||||
},
|
||||
// Center justify equations in code and markdown cells. Elsewhere
|
||||
// we use CSS to left justify single line equations in code cells.
|
||||
displayAlign: 'center',
|
||||
"HTML-CSS": { fonts: ["TeX"] }
|
||||
});
|
||||
|
||||
</script>
|
||||
<script>
|
||||
function handleImages() {
|
||||
var images = document.querySelectorAll('p>img')
|
||||
|
||||
console.log(images);
|
||||
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>
|
||||
@ -620,7 +620,7 @@ are calculated.</p>
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">201</span><span class="k">def</span> <span class="nf">_mha</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">TransformerConfigs</span><span class="p">):</span>
|
||||
<span class="lineno">202</span> <span class="k">return</span> <span class="n">MultiHeadAttention</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">n_heads</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">d_model</span><span class="p">)</span>
|
||||
<span class="lineno">202</span> <span class="k">return</span> <span class="n">MultiHeadAttention</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">n_heads</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">d_model</span><span class="p">,</span> <span class="n">dropout_prob</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">dropout</span><span class="p">)</span>
|
||||
<span class="lineno">203</span>
|
||||
<span class="lineno">204</span>
|
||||
<span class="lineno">205</span><span class="n">calculate</span><span class="p">(</span><span class="n">TransformerConfigs</span><span class="o">.</span><span class="n">encoder_attn</span><span class="p">,</span> <span class="s1">'mha'</span><span class="p">,</span> <span class="n">_mha</span><span class="p">)</span>
|
||||
|
||||
@ -1,71 +0,0 @@
|
||||
"""
|
||||
---
|
||||
title: Attention with Linear Biases (ALiBi)
|
||||
summary: >
|
||||
Documented implementation with explanations of Attention with Linear Biases (ALiBi)
|
||||
---
|
||||
|
||||
# Attention with Linear Biases (ALiBi)
|
||||
|
||||
This is an implementation of Attention with Linear Biases (ALiBi).
|
||||
"""
|
||||
import math
|
||||
|
||||
import torch
|
||||
from torch import nn
|
||||
|
||||
from labml.logger import inspect
|
||||
from labml_nn.transformers.mha import MultiHeadAttention
|
||||
|
||||
|
||||
def get_slopes(n_heads: int):
|
||||
"""
|
||||
## Get head-specific slope $m$ for each head
|
||||
"""
|
||||
|
||||
assert math.log2(n_heads).is_integer()
|
||||
|
||||
s = (2 ** (-2 ** -(math.log2(n_heads) - 3)))
|
||||
r = s
|
||||
return [s * (r ** i) for i in range(n_heads)]
|
||||
|
||||
|
||||
class AlibiMultiHeadAttention(MultiHeadAttention):
|
||||
"""
|
||||
## Attention with Linear Biases (ALiBi)
|
||||
|
||||
We override [Multi-Head Attention](mha.html) module so we only need to
|
||||
write the `get_scores` method.
|
||||
"""
|
||||
|
||||
def __init__(self, heads: int, d_model: int, dropout_prob: float = 0.1):
|
||||
# The linear transformations do not need a bias since we
|
||||
# explicitly include it when calculating scores.
|
||||
# However having a bias for `value` might make sense.
|
||||
super().__init__(heads, d_model, dropout_prob)
|
||||
|
||||
self.slopes = nn.Parameter(torch.tensor(get_slopes(heads)), requires_grad=False)
|
||||
|
||||
def get_scores(self, query: torch.Tensor, key: torch.Tensor):
|
||||
r"""
|
||||
### Calculate attention scores and add attention biases
|
||||
"""
|
||||
|
||||
# scores has shape `[query_seq_len, key_seq_len, batch_size, head]`
|
||||
scores = super().get_scores(query, key)
|
||||
|
||||
distance = torch.arange(scores.shape[1]).to(scores.device, scores.dtype)
|
||||
bias = distance[None, :, None, None] * self.slopes[None, None, None, :]
|
||||
# add to scores
|
||||
scores = scores + bias
|
||||
|
||||
return scores
|
||||
|
||||
|
||||
def _test_slopes():
|
||||
inspect(get_slopes(8))
|
||||
inspect(get_slopes(16))
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
_test_slopes()
|
||||
129
labml_nn/transformers/alibi/__init__.py
Normal file
129
labml_nn/transformers/alibi/__init__.py
Normal file
@ -0,0 +1,129 @@
|
||||
"""
|
||||
---
|
||||
title: Attention with Linear Biases (ALiBi)
|
||||
summary: >
|
||||
Documented implementation with explanations of Attention with Linear Biases (ALiBi)
|
||||
---
|
||||
|
||||
# Attention with Linear Biases (ALiBi)
|
||||
|
||||
This is an implementation of Attention with Linear Biases (ALiBi) from the paper
|
||||
Train Short, Test Long: Attention with Linear Biases Enables Input Length Extrapolation
|
||||
[(pdf)](https://ofir.io/train_short_test_long.pdf).
|
||||
|
||||
This replaces positional encodings with biases added to attention scores (attention logits, before the softmax).
|
||||
This is a relative scheme tested on autoregressive tasks, and the bias is higher for closeby tokens
|
||||
and lower for far-away tokens.
|
||||
The biases decrease linearly in the log scale (because it's before the softmax) and each head has a different slope.
|
||||
|
||||
Here's the attention formula for $i$-th token,
|
||||
|
||||
\begin{align}
|
||||
\mathbf{a}_i
|
||||
&= \text{softmax} \bigg( \mathbf{q}_i \mathbf{K}^\top + m \cdot \big[-(i-1), \dots, 1, 0 \big] \bigg) \\
|
||||
&= \text{softmax} \bigg( \mathbf{q}_i \mathbf{K}^\top + m \cdot \big[0, 1, \dots, (i - 1) \big] \bigg)
|
||||
\end{align}
|
||||
|
||||
where $\mathbf{q}_i \in \mathbb{R}^d$ is the query of the $i$-th token, $K \in \mathbb{R}^{i \times d}$ are the keys
|
||||
up to $i$, and $d$ the number of features per head.
|
||||
Note that the above equality halts because $\text{softmax}$ is invariant to translations
|
||||
(you can add any constant to all elements without changing the result).
|
||||
|
||||
Here is [the training code](experiment.html) for a ALiBi model.
|
||||
|
||||
[](https://app.labml.ai/run/e87bec2a074911ec82cdd1759f10c925)
|
||||
"""
|
||||
import math
|
||||
|
||||
import torch
|
||||
from torch import nn
|
||||
|
||||
from labml.logger import inspect
|
||||
from labml_nn.transformers.mha import MultiHeadAttention
|
||||
|
||||
|
||||
def get_slopes(n_heads: int):
|
||||
"""
|
||||
## Get head-specific slope $m$ for each head
|
||||
|
||||
* `n_heads` is the number of heads in the attention layer $n$
|
||||
|
||||
The slope for first head is
|
||||
|
||||
$$2^{-2^{-(\log_2 n) - 3}}$$
|
||||
|
||||
The slopes for the rest of the heads are in a geometric series with a ratio same as above.
|
||||
|
||||
For instance when the number of heads is $8$ the slopes are
|
||||
$$\frac{1}{2^1}, \frac{1}{2^2}, \dots, \frac{1}{2^8}$$
|
||||
"""
|
||||
|
||||
# $$2^{-2^{-(\log_2 n) - 3}}$$
|
||||
s = (2 ** (-2 ** -(math.log2(n_heads) - 3)))
|
||||
# The geometric sequence
|
||||
return [s * (s ** i) for i in range(n_heads)]
|
||||
|
||||
|
||||
def get_biases(n_heads: int, max_len: int):
|
||||
"""
|
||||
## Calculate the attention biases matrix
|
||||
|
||||
* `n_heads` is the number of heads in the attention layer
|
||||
* `max_len` is the maximum sequence length
|
||||
|
||||
This returns a matrix of shape `[n_heads, max_len]` with attention biases.
|
||||
"""
|
||||
|
||||
# Get slopes $m$ for each head
|
||||
slopes = torch.tensor(get_slopes(n_heads))
|
||||
# Calculate distances $[0, 1, \dots, N]$
|
||||
distance = torch.arange(max_len).to(torch.float)
|
||||
# Multiply them pair-wise to get the bias matrix
|
||||
return distance[:, None] * slopes[None, :]
|
||||
|
||||
|
||||
class AlibiMultiHeadAttention(MultiHeadAttention):
|
||||
"""
|
||||
## Attention with Linear Biases (ALiBi)
|
||||
|
||||
We override [Multi-Head Attention](mha.html) module so we only need to
|
||||
write the `get_scores` method.
|
||||
"""
|
||||
|
||||
def __init__(self, heads: int, d_model: int, dropout_prob: float = 0.1, max_len: int = 5_000):
|
||||
super().__init__(heads, d_model, dropout_prob)
|
||||
|
||||
# Pre-calculate the biases
|
||||
self.bias = nn.Parameter(get_biases(heads, max_len), requires_grad=False)
|
||||
|
||||
def get_scores(self, query: torch.Tensor, key: torch.Tensor):
|
||||
r"""
|
||||
### Calculate attention scores and add attention biases
|
||||
"""
|
||||
|
||||
# Calculate the standard attention score.
|
||||
# It has shape `[query_seq_len, key_seq_len, batch_size, head]`
|
||||
scores = super().get_scores(query, key)
|
||||
|
||||
# Number of keys
|
||||
key_seq_len = scores.shape[1]
|
||||
# Add the biases to scores.
|
||||
#
|
||||
# $$\mathbf{q}_i \mathbf{K}^\top + m \cdot \big[0, 1, \dots, (i - 1) \big]$$
|
||||
#
|
||||
# Note that we add biases for all keys (not just upto $i$). We can do this since
|
||||
# those extra entries will get removed because of the masking later.
|
||||
return scores + self.bias[None, :key_seq_len, None, :]
|
||||
|
||||
|
||||
def _test_slopes():
|
||||
"""
|
||||
Simple test function to see the slopes.
|
||||
"""
|
||||
inspect(get_slopes(8))
|
||||
inspect(get_slopes(16))
|
||||
|
||||
|
||||
#
|
||||
if __name__ == '__main__':
|
||||
_test_slopes()
|
||||
@ -1,32 +1,66 @@
|
||||
"""
|
||||
---
|
||||
title: Attention with Linear Biases (ALiBi) Experiment
|
||||
summary: This experiment trains an Attention with Linear Biases (ALiBi) based model on Tiny Shakespeare dataset.
|
||||
---
|
||||
|
||||
# [Attention with Linear Biases (ALiBi)](index.html) Experiment
|
||||
|
||||
This is an annotated PyTorch experiment to train a [ALiBi model](index.html).
|
||||
|
||||
This is based on
|
||||
[our GPT model](../gpt/index.html).
|
||||
|
||||
[](https://app.labml.ai/run/e87bec2a074911ec82cdd1759f10c925)
|
||||
"""
|
||||
|
||||
import torch
|
||||
from torch.utils.data import DataLoader
|
||||
|
||||
from labml import experiment, tracker
|
||||
from labml.configs import option, calculate
|
||||
from labml_helpers.datasets.text import SequentialUnBatchedDataset
|
||||
from labml_nn.alibi import AlibiMultiHeadAttention
|
||||
from labml_nn.transformers.alibi import AlibiMultiHeadAttention
|
||||
from labml_nn.experiments.nlp_autoregression import transpose_batch
|
||||
from labml_nn.transformers import TransformerConfigs
|
||||
from labml_nn.transformers.gpt import Configs as GPTConfigs
|
||||
|
||||
|
||||
class Configs(GPTConfigs):
|
||||
"""
|
||||
## Configurations
|
||||
|
||||
We extend [GPT configurations](../gpt/index.html) and change the attention mechanism.
|
||||
"""
|
||||
|
||||
# ALiBi based transformer (defined below)
|
||||
transformer: TransformerConfigs = 'GPT_ALiBi'
|
||||
# Longer validation set
|
||||
valid_seq_len: int = 128
|
||||
valid_loader = 'shuffled_longer_valid_loader'
|
||||
|
||||
def other_metrics(self, output: torch.Tensor, target: torch.Tensor):
|
||||
"""
|
||||
Log losses at the initial and final tokens
|
||||
"""
|
||||
# If there are more tokens that the training sequence length (during validation),
|
||||
if self.seq_len < output.shape[0]:
|
||||
# Log the loss at training sequence length
|
||||
tracker.add(f'loss.{self.seq_len - 1}.', self.loss_func(output[self.seq_len - 1], target[self.seq_len - 1]))
|
||||
# Log the loss at the first token
|
||||
tracker.add(f'loss.0.', self.loss_func(output[0], target[0]))
|
||||
# Log the loss at the final token
|
||||
tracker.add(f'loss.{int(output.shape[0]) - 1}.', self.loss_func(output[-1], target[-1]))
|
||||
|
||||
|
||||
# ### Multi-head Attention
|
||||
def _alibi_mha(c: TransformerConfigs):
|
||||
"""
|
||||
Create an ALiBi attention module
|
||||
"""
|
||||
return AlibiMultiHeadAttention(c.n_heads, c.d_model, dropout_prob=c.dropout)
|
||||
|
||||
|
||||
# Set all attention mechanisms to ALiBi
|
||||
calculate(TransformerConfigs.encoder_attn, 'alibi_mha', _alibi_mha)
|
||||
calculate(TransformerConfigs.decoder_attn, 'alibi_mha', _alibi_mha)
|
||||
calculate(TransformerConfigs.decoder_mem_attn, 'alibi_mha', _alibi_mha)
|
||||
@ -35,7 +69,7 @@ calculate(TransformerConfigs.decoder_mem_attn, 'alibi_mha', _alibi_mha)
|
||||
@option(Configs.valid_loader)
|
||||
def shuffled_longer_valid_loader(c: Configs):
|
||||
"""
|
||||
### Shuffled validation data loader
|
||||
Shuffled validation data loader with `valid_seq_len` sequence length
|
||||
"""
|
||||
return DataLoader(SequentialUnBatchedDataset(text=c.text.valid,
|
||||
dataset=c.text,
|
||||
@ -48,7 +82,7 @@ def shuffled_longer_valid_loader(c: Configs):
|
||||
@option(Configs.transformer, 'GPT_ALiBi')
|
||||
def _transformer_configs(c: Configs):
|
||||
"""
|
||||
### Transformer configurations
|
||||
### ALiBi based Transformer configurations
|
||||
"""
|
||||
|
||||
# We use our
|
||||
@ -60,9 +94,11 @@ def _transformer_configs(c: Configs):
|
||||
# GPT uses GELU activation for position wise feedforward
|
||||
conf.ffn.activation = 'GELU'
|
||||
|
||||
# ALiBi doesn't use positional embeddings
|
||||
conf.src_embed = 'no_pos'
|
||||
conf.tgt_embed = 'no_pos'
|
||||
|
||||
# Set all attention mechanisms to ALiBi
|
||||
conf.encoder_attn = 'alibi_mha'
|
||||
conf.decoder_attn = 'alibi_mha'
|
||||
conf.decoder_mem_attn = 'alibi_mha'
|
||||
@ -105,7 +141,6 @@ def main():
|
||||
'transformer.ffn.d_ff': 512,
|
||||
'transformer.n_heads': 8,
|
||||
'transformer.n_layers': 4,
|
||||
|
||||
'transformer.dropout': 0.1,
|
||||
})
|
||||
|
||||
Reference in New Issue
Block a user