Merge branch 'master' of https://github.com/vietai/back_translate

colab/Sentiment_Analysis_+_Back_translation.ipynb

@@ -497,12 +497,12 @@
  "source": [
  "# Back-translate\n",
  "\n",
- "You can surely perform beam search decoding one sentence after another until finish. This, however, would take a very long time. Our code helps you with handling all possible speedups you can make:\n",
+ "You can surely perform beam search decoding one sentence after another until finish. This, however, would take a very long time, even with this dataset of only 27K examples. Our code helps you with handling all possible speedups you can make:\n",
  "\n",
- "* Decoding on ML accelerator hardware.\n",
+ "* Decoding on ML accelerator hardware (GPU or TPU).\n",
  "* Batch processing of sentences.\n",
  "* Processing sentences of similar length at the same time.\n",
- "* Saving intermediate results during the process.\n",
+ "* Saving intermediate results during the process to deal with Colab Notebook preemtions.\n",
  "\n",
  "At its core, the `back_translate.py` script essentially packages two calls to `t2t-decode` from vi -> en and then en -> vi. We added the ability to save intermediate results after every single batch of data since back-translating the full dataset can take quite a long time."
  ]
@@ -518,7 +518,7 @@
  "source": [
  "output_file = os.path.join(foody_data_dir, 'foody_paraphrased_train.txt')\n",
  "\n",
- "is_demo = False #@param {type:\"boolean\"}\n",
+ "is_demo = True #@param {type:\"boolean\"}\n",
  "from_ckpt = '/content/vien-translation/checkpoints/translate_vien_iwslt32k_tiny/avg/' # @param{type: \"string\"}\n",
  "to_ckpt = '/content/vien-translation/checkpoints/translate_envi_iwslt32k_tiny/avg/' # @param{type: \"string\"}\n",
  "\n",