undo cache

.gitignore

@@ -1,6 +1,7 @@
 _candidate
 .sass-cache
 .jekyll-metadata
+.jekyll-cache
 node_modules
 Gemfile.lock
 

Gemfile

@@ -15,4 +15,5 @@ group :jekyll_plugins do
     gem 'jekyll-sitemap'
     gem 'jekyll-paginate'
     gem 'jekyll-seo-tag'
-end
+end
+gem "webrick", "~> 1.7"

_config.yml

@@ -54,10 +54,15 @@ defaults:
       layout: page
 
 # Syntax
-markdown:
-  kramdown:
-    input: GFM
-    syntax_highlighter: rouge
+markdown: kramdown
+highlighter: rouge
+kramdown:
+  input: GFM
+  syntax_highlighter_opts:
+    default_lang: html
+    css_class   : 'syntax'    
+
+
 
 # Paginate
 paginate: 10

_layouts/default.html

@@ -22,10 +22,6 @@
     <!-- Theme Stylesheet -->
     <link rel="stylesheet" href="{{site.baseurl}}/assets/css/theme.css">
 
-    <!-- Syntax Highlither, rouge Stylesheet -->
-
-    <link href="{{site.baseurl}}/assets/css/syntax.css" rel="stylesheet" >
-
     <!-- Jquery on header to make sure everything works, the rest  of the scripts in footer for fast loading -->
     <script
     src="https://code.jquery.com/jquery-3.3.1.min.js"

_posts/part2/2021-12-22-RF-transformer.md

@@ -33,18 +33,16 @@ most of the code and model architecture comes from the paper [Attention is all y
 
 A1.
 
-<pre>
-    <code class="python">
-        class PositionalEncoding(nn.Module):
-            def __init__(self, d_model:int):
-                super().__init__()
-                self.register_buffer('freq', 1/(10000 ** (torch.arange(0, d_model, 2.))))
-            def __call__(self, pos):
-                inp = torch.outer(pos, self.freq)
-                enc = torch.cat([inp.sin(), inp.cos()], dim=-1)
-                return enc
-    </code>
-</pre>
+{% highlight python %}
+class PositionalEncoding(nn.Module):
+    def __init__(self, d_model:int):
+        super().__init__()
+        self.register_buffer('freq', 1/(10000 ** (torch.arange(0, d_model, 2.))))
+    def __call__(self, pos):
+        inp = torch.outer(pos, self.freq)
+        enc = torch.cat([inp.sin(), inp.cos()], dim=-1)
+        return enc
+{% endhighlight %}
 
 [^5]
 
@@ -53,7 +51,7 @@ A1.
 However, as of Jan 2022, many paper reported that positional does not affect the performance. Of course it depends on specifics 😉.
 
 
-```python
+{% highlight python %}
 positions = torch.arange(0, 100).float(); positions[:10]
 d_model = 26  # original paper set this value to 512
 
@@ -67,8 +65,7 @@ for i in range(0,3):
     ax[0].legend()
     ax[0].set_xlabel("relative posotion")
     ax[1].plot(res[:,int(d_model/2+i)], label=f"cos, cur pos:{i}"); ax[1].legend()
-
-```
+{% endhighlight %}
 
 A2