Use in-place multiplication and new_empty. (Atcold#756)

To speed up computation use in-place tensor multiplication. Use new_empty to create a new tensor, eps, based off of tensor std's properties.
hayaalsh · Jun 29, 2021 · dc26995 · dc26995
1 parent ab9eb59
commit dc26995
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/11-VAE.ipynb b/11-VAE.ipynb
@@ -115,8 +115,8 @@
     "    def reparameterise(self, mu, logvar):\n",
     "        if self.training:\n",
     "            std = logvar.mul(0.5).exp_()\n",
-    "            eps = std.data.new(std.size()).normal_()\n",
-    "            return eps.mul(std).add_(mu)\n",
+    "            eps = std.new_empty(std.size()).normal_()\n",
+    "            return eps.mul_(std).add_(mu)\n",
     "        else:\n",
     "            return mu\n",
     "\n",