Update README.md
Browse files
README.md
CHANGED
@@ -9,4 +9,8 @@ Performance in terms of [sacrebleu](https://huggingface.co/docs/datasets/v1.5.0/
|
|
9 |
* learning_rate=2e-5
|
10 |
* batch_size: 32
|
11 |
* weight_decay=0.01
|
12 |
-
* num_train_epochs=1
|
|
|
|
|
|
|
|
|
|
9 |
* learning_rate=2e-5
|
10 |
* batch_size: 32
|
11 |
* weight_decay=0.01
|
12 |
+
* num_train_epochs=1
|
13 |
+
|
14 |
+
# Thoughts
|
15 |
+
* 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
|
16 |
+
* Easier and much faster to train compared to t5 or byt5.
|