better browser plugins

Author: Eduardo Leao

dist/index.js

@@ -1780,12 +1780,12 @@ class Embedding extends Module {
   /**
    * Embedding class, turns indexes into vectors.
    *
-   * @param {number} in_size - number of different indexes (vocabulary size).
-   * @param {number} out_size - size of the embedding vector generated.
+   * @param {number} vocab_size - number of different indexes (vocabulary size).
+   * @param {number} embed_size - size of the embedding vector generated.
    */
-  constructor(in_size, embed_size) {
+  constructor(vocab_size, embed_size) {
     super();
-    this.E = randn([in_size, embed_size], true, "cpu", false);
+    this.E = randn([vocab_size, embed_size], true, "cpu", false);
   }
   /**
    * Extracts embedding from rows in "idx":
@@ -1802,14 +1802,14 @@ class Embedding extends Module {
 class PositionalEmbedding extends Module {
   E;
   /**
-   * Embedding class, turns indexes into vectors.
+   * Embedding class, turns indexes into vectors based on it's position through an optimized lookup table.
    *
-   * @param {number} n_timesteps - number of different embeddings (number of timesteps in each instance in batch).
+   * @param {number} input_size - number of different embeddings (size of the input).
    * @param {number} embed_size - size of the embedding vector generated.
    */
-  constructor(n_timesteps, embed_size) {
+  constructor(input_size, embed_size) {
     super();
-    this.E = randn([n_timesteps, embed_size], true, "cpu", false);
+    this.E = randn([input_size, embed_size], true, "cpu", false);
   }
   /**
    * Gets embedding for timesteps in "idx" array.

dist/index.mjs

@@ -1780,12 +1780,12 @@ class Embedding extends Module {
   /**
    * Embedding class, turns indexes into vectors.
    *
-   * @param {number} in_size - number of different indexes (vocabulary size).
-   * @param {number} out_size - size of the embedding vector generated.
+   * @param {number} vocab_size - number of different indexes (vocabulary size).
+   * @param {number} embed_size - size of the embedding vector generated.
    */
-  constructor(in_size, embed_size) {
+  constructor(vocab_size, embed_size) {
     super();
-    this.E = randn([in_size, embed_size], true, "cpu", false);
+    this.E = randn([vocab_size, embed_size], true, "cpu", false);
   }
   /**
    * Extracts embedding from rows in "idx":
@@ -1802,14 +1802,14 @@ class Embedding extends Module {
 class PositionalEmbedding extends Module {
   E;
   /**
-   * Embedding class, turns indexes into vectors.
+   * Embedding class, turns indexes into vectors based on it's position through an optimized lookup table.
    *
-   * @param {number} n_timesteps - number of different embeddings (number of timesteps in each instance in batch).
+   * @param {number} input_size - number of different embeddings (size of the input).
    * @param {number} embed_size - size of the embedding vector generated.
    */
-  constructor(n_timesteps, embed_size) {
+  constructor(input_size, embed_size) {
     super();
-    this.E = randn([n_timesteps, embed_size], true, "cpu", false);
+    this.E = randn([input_size, embed_size], true, "cpu", false);
   }
   /**
    * Gets embedding for timesteps in "idx" array.

dist/js-pytorch-browser.js

@@ -753,14 +753,16 @@ <h2 id="torchtril">torch.tril</h2>
 <h2 id="torchrandn">torch.randn</h2>
 <pre><code>torch.randn(*shape,
             requires_grad=false,
-            device='cpu') → Tensor
+            device='cpu',
+            xavier=false) → Tensor
 </code></pre>
 <p>Returns a tensor filled with randomly sampled data with dimensions like <code>shape</code>. The sample is from a normal distribution.</p>
 <p>Parameters</p>
 <ul>
 <li><strong>shape (Array)</strong> - Javascript Array containing the shape of the Tensor.</li>
 <li><strong>requires_grad (boolean)</strong> - Whether to keep track of this tensor's gradients. Set this to true if you want to <strong>learn</strong> this parameter in your model. Default: <code>false</code>.</li>
 <li><strong>device (string)</strong> - Device to store Tensor. Either "gpu" or "cpu". If your device has a gpu, large models will train faster on it.</li>
+<li><strong>xavier (boolean)</strong> - Whether to use <a target="_blank" href="https://prateekvishnu.medium.com/xavier-and-he-normal-he-et-al-initialization-8e3d7a087528">Xavier Initialization</a> on this tensor. Default: <code>false</code>.</li>
 </ul>
 <p>Example</p>
 <pre><code class="language-javascript">&gt;&gt;&gt; let a = torch.randn([3,2], false, 'gpu');