Update README.md

README.md

@@ -9,4 +9,8 @@ Performance in terms of [sacrebleu](https://huggingface.co/docs/datasets/v1.5.0/
 * learning_rate=2e-5
 * batch_size: 32
 * weight_decay=0.01
-* num_train_epochs=1
+* num_train_epochs=1
+
+# Thoughts
+* Model `Helsinki-NLP/opus-mt-en-vi` is small (around 260MB), and can be easily deployed to a cheap server (e.g., EC2 t2.medium) without CPU
+* Easier and much faster to train compared to t5 or byt5.