Update documentation

_sources/llm-defined.md

@@ -22,13 +22,6 @@ A `Language Model` (LM) is a model that assigns probabilities to sequences of wo
 </video>
 :::
 
-<!-- ```{figure} ./images/animated-transfomer.mp4
----
-width: 600px
-name: animated-transfomer
----
-[Animated Transfomer](https://prvnsmpth.github.io/animated-transformer/)
-``` -->
 
 ```{figure} ./images/lm-hist.png
 ---
@@ -40,9 +33,9 @@ name: lm-hist
 
 :::{admonition} **[Large](https://en.wikipedia.org/wiki/Large_language_model#List)**
 :class: tip
-We call a language model `large` when it has 
-- Many parameters (billions)
-- And has been trained on large quantities of language data (billions of words/tokens)
+We call a language model `large` when:
+- The model has billions of parameters
+- The model has been trained on billions of words/tokens
 :::
 
 ```{figure} ./images/wikipedia-list.png
@@ -55,7 +48,7 @@ name: wikipedia-list
 
 :::{admonition} **Transformer**
 :class: tip
-Transformers were the key innovation that allowed language models to get large. They are a deep learning architecture that allow massive parallelization of training and inference on GPUs
+`Transformers` were the key innovation that allowed language models to get large. They are a deep learning architecture that allow massive parallelization of training and inference on GPUs
 :::
 
 ```{figure} ./images/ai-2-transformer.png
@@ -76,12 +69,12 @@ name: attention
 
 :::{admonition} **Pre-trained**
 :class: tip
-Pre-trained language models have been trained via self-supervision on vast quantities of text. These are also called [`foundation`](https://en.wikipedia.org/wiki/Foundation_models) models. They are not typically useful until...
+`Pre-trained` language models have been trained via self-supervision on vast quantities of text. These are also called [`foundation`](https://en.wikipedia.org/wiki/Foundation_models) models. They are not typically useful until...
 :::
 
 :::{admonition} **Generative**
 :class: tip
-Generative models are foundation models that have been further trained via supervised fine-tuning and reinforcement learning from human feedback (RLHF) to behave in a useful and safe manner, for example by responding to questions with answers like a chat assistant.
+`Generative` models are foundation models that have been further trained via supervised fine-tuning and reinforcement learning from human feedback (RLHF) to behave in a useful and safe manner, for example by responding to questions with answers like a chat assistant.
 :::
 
 :::{card} [OpenAI](https://en.wikipedia.org/wiki/OpenAI):

llm-defined.html

@@ -411,13 +411,6 @@ <h1>Generative Pre-Trained Transformer (GPT)<a class="headerlink" href="#generat
 </video>
 </div>
 </div>
-<!-- ```{figure} ./images/animated-transfomer.mp4
----
-width: 600px
-name: animated-transfomer
----
-[Animated Transfomer](https://prvnsmpth.github.io/animated-transformer/)
-``` -->
 <figure class="align-default" id="lm-hist">
 <a class="reference internal image-reference" href="_images/lm-hist.png"><img alt="_images/lm-hist.png" src="_images/lm-hist.png" style="width: 600px;" /></a>
 <figcaption>
@@ -426,10 +419,10 @@ <h1>Generative Pre-Trained Transformer (GPT)<a class="headerlink" href="#generat
 </figure>
 <div class="tip admonition">
 <p class="admonition-title"><strong><a class="reference external" href="https://en.wikipedia.org/wiki/Large_language_model#List">Large</a></strong></p>
-<p>We call a language model <code class="docutils literal notranslate"><span class="pre">large</span></code> when it has</p>
+<p>We call a language model <code class="docutils literal notranslate"><span class="pre">large</span></code> when:</p>
 <ul class="simple">
-<li><p>Many parameters (billions)</p></li>
-<li><p>And has been trained on large quantities of language data (billions of words/tokens)</p></li>
+<li><p>The model has billions of parameters</p></li>
+<li><p>The model has been trained on billions of words/tokens</p></li>
 </ul>
 </div>
 <figure class="align-default" id="wikipedia-list">
@@ -440,7 +433,7 @@ <h1>Generative Pre-Trained Transformer (GPT)<a class="headerlink" href="#generat
 </figure>
 <div class="tip admonition">
 <p class="admonition-title"><strong>Transformer</strong></p>
-<p>Transformers were the key innovation that allowed language models to get large. They are a deep learning architecture that allow massive parallelization of training and inference on GPUs</p>
+<p><code class="docutils literal notranslate"><span class="pre">Transformers</span></code> were the key innovation that allowed language models to get large. They are a deep learning architecture that allow massive parallelization of training and inference on GPUs</p>
 </div>
 <figure class="align-default" id="trans-subset">
 <a class="reference internal image-reference" href="_images/ai-2-transformer.png"><img alt="_images/ai-2-transformer.png" src="_images/ai-2-transformer.png" style="width: 600px;" /></a>
@@ -454,11 +447,11 @@ <h1>Generative Pre-Trained Transformer (GPT)<a class="headerlink" href="#generat
 </figure>
 <div class="tip admonition">
 <p class="admonition-title"><strong>Pre-trained</strong></p>
-<p>Pre-trained language models have been trained via self-supervision on vast quantities of text. These are also called <a class="reference external" href="https://en.wikipedia.org/wiki/Foundation_models"><code class="docutils literal notranslate"><span class="pre">foundation</span></code></a> models. They are not typically useful until…</p>
+<p><code class="docutils literal notranslate"><span class="pre">Pre-trained</span></code> language models have been trained via self-supervision on vast quantities of text. These are also called <a class="reference external" href="https://en.wikipedia.org/wiki/Foundation_models"><code class="docutils literal notranslate"><span class="pre">foundation</span></code></a> models. They are not typically useful until…</p>
 </div>
 <div class="tip admonition">
 <p class="admonition-title"><strong>Generative</strong></p>
-<p>Generative models are foundation models that have been further trained via supervised fine-tuning and reinforcement learning from human feedback (RLHF) to behave in a useful and safe manner, for example by responding to questions with answers like a chat assistant.</p>
+<p><code class="docutils literal notranslate"><span class="pre">Generative</span></code> models are foundation models that have been further trained via supervised fine-tuning and reinforcement learning from human feedback (RLHF) to behave in a useful and safe manner, for example by responding to questions with answers like a chat assistant.</p>
 </div>
 <div class="sd-card sd-sphinx-override sd-mb-3 sd-shadow-sm docutils">
 <div class="sd-card-body docutils">