Tiny fix

Hvass-Labs · Hvass-Labs · commit aa0d6796c6bb · 2019-02-03T16:21:48.000+01:00
diff --git a/21_Machine_Translation.ipynb b/21_Machine_Translation.ipynb
@@ -1719,7 +1719,7 @@
    "source": [
     "Now we can train the model. One epoch of training took about 1 hour on a GTX 1070 GPU. You probably need to run 10 epochs or more during training. After 10 epochs the loss was about 1.10 on the training-set and about 1.15 on the validation-set.\n",
     "\n",
-    "Note the strange batch-size of 640 (512 + 128) which was chosen because it kept the GPU running at nearly 100% while being within the memory limits of 8GB for this GPU."
+    "Note the batch-size of 512 which was chosen because it kept the GPU running at nearly 100% while being within the memory limits of 8GB for this GPU."
    ]
   },
   {
@@ -1732,7 +1732,7 @@
    "source": [
     "model_train.fit(x=x_data,\n",
     "                y=y_data,\n",
-    "                batch_size=640,\n",
+    "                batch_size=512,\n",
     "                epochs=10,\n",
     "                validation_split=validation_split,\n",
     "                callbacks=callbacks)"
@@ -2162,7 +2162,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.6.1"
+   "version": "3.6.6"
   }
  },
  "nbformat": 4,
