updated lecture 6, pulled in ELIZA assignment updates

Author: jeremymanning

assignments/eliza-llm-course

@@ -1041,7 +1041,7 @@ <h1 id="bpe-in-practice-with-huggingface">BPE in practice with HuggingFace</h1>
 </foreignObject></svg><svg data-marpit-svg="" viewBox="0 0 1280 720"><foreignObject width="1280" height="720"><section id="14" data-theme="cdl-theme" lang="en-US" style="--theme:cdl-theme;" data-transition-back="{&quot;name&quot;:&quot;fade&quot;,&quot;duration&quot;:&quot;0.25s&quot;,&quot;builtinFallback&quot;:true}" data-transition="{&quot;name&quot;:&quot;fade&quot;,&quot;duration&quot;:&quot;0.25s&quot;,&quot;builtinFallback&quot;:true}">
 <h1 id="wordpiece-merges-surprisingly-common-pairs">WordPiece merges &quot;surprisingly common&quot; pairs</h1>
 <div class="note-box" data-title="Further reading">
-<p><a href="https://arxiv.org/abs/1609.08144"><strong>Wu et al. (2016, <em>arXiv</em>):</strong></a> Google's Neural Machine Translation System</p>
+<p><a href="https://arxiv.org/abs/1609.08144"><strong>Wu et al. (2016, <em>arXiv</em>):</strong></a> Google's neural machine translation system</p>
 </div>
 <div class="definition-box" data-title="Key difference from BPE">
 <p>Instead of merging most <em>frequent</em> pair, merge pair that maximizes <em>likelihood</em>:</p>
@@ -1080,6 +1080,7 @@ <h1 id="wordpiece-example-with-bert">WordPiece example with BERT</h1>
 <h1 id="sentencepiece-works-for-languages-without-spaces">SentencePiece works for languages without spaces</h1>
 <div class="note-box" data-title="Further reading">
 <p><a href="https://aclanthology.org/D18-2012/"><strong>Kudo &amp; Richardson (2018, <em>EMNLP</em>):</strong></a> SentencePiece: a simple and language independent subword tokenizer</p>
+<p><a href="https://arxiv.org/abs/1910.13267"><strong>Provilkov, Emelianeko, &amp; Voita (2019, <em>arXiv</em>):</strong></a> BPE-Dropout: simple and effective subword regularization</p>
 </div>
 <div class="warning-box" data-title="Problems with BPE/WordPiece">
 <ul>
@@ -1339,7 +1340,7 @@ <h1 id="looking-ahead">Looking ahead</h1>
 <div class="tip-box" data-title="Prepare by...">
 <ul>
 <li>Playing more with the <a href="https://contextlab.github.io/llm-course/demos/tokenization/">Tokenization Explorer Demo</a></li>
-<li>Experimenting with HuggingFace tokenizers</li>
+<li>Experimenting with <a href="https://github.com/huggingface/tokenizers">HuggingFace tokenizers</a></li>
 </ul>
 </div>
 </section>

slides/week2/lecture6.html

@@ -331,7 +331,7 @@ Use the [Tokenization Explorer Demo](https://contextlab.github.io/llm-course/dem
 
 <div class="note-box" data-title="Further reading">
 
-[**Wu et al. (2016, *arXiv*):**](https://arxiv.org/abs/1609.08144) Google's Neural Machine Translation System
+[**Wu et al. (2016, *arXiv*):**](https://arxiv.org/abs/1609.08144) Google's neural machine translation system
 
 </div>
 
@@ -380,6 +380,8 @@ print("Tokens:", tokens2)
 
 [**Kudo & Richardson (2018, *EMNLP*):**](https://aclanthology.org/D18-2012/) SentencePiece: a simple and language independent subword tokenizer
 
+[**Provilkov, Emelianeko, & Voita (2019, *arXiv*):**](https://arxiv.org/abs/1910.13267) BPE-Dropout: simple and effective subword regularization
+
 </div>
 
 <div class="warning-box" data-title="Problems with BPE/WordPiece">
@@ -612,7 +614,7 @@ print("Without special:", clean_decode)  # "hello world"
 <div class="tip-box" data-title="Prepare by...">
 
 - Playing more with the [Tokenization Explorer Demo](https://contextlab.github.io/llm-course/demos/tokenization/)
-- Experimenting with HuggingFace tokenizers
+- Experimenting with [HuggingFace tokenizers](https://github.com/huggingface/tokenizers)
 
 </div>