Num-Eng Machine Translation

I have trained and fine-tuned the Text-to-Text Transfer Transformer (T5) on custom dataset that I have downloaded from here

Here, I used huggingface's t5-small model (66 million parameters). I set batch_size=128 and learning_rate=0.001 (as they propossed in the literature) without any warmup or lr_schedule.

1. • Input text: 392665

   • Pre-processed text: "number to english: 3 9 2 6 6 5"

   • Predicted output text: "three hundred ninety-two thousand six hundred sixty-five"

2. • Input text: "four hundred ninety-eight thousand one hundred forty-two"

   • Pre-processed text: "english to number: four hundred ninety-eight thousand one hundred forty-two"

   • Predicted output text: "4 9 8 1 4 2"

3. • Input text: 92923

   • Pre-processed text: "number to english: 9 2 9 2 3"

   • Predicted output text: "ninety-two thousand nine hundred twenty-three"

4. • Input text: "eight hundred ninety-four thousand nine hundred sixty-five"

   • Pre-processed text: "english to number: eight hundred ninety-four thousand nine hundred sixty-five"

   • Predicted output text: "8 9 4 9 6 5"

5. • Input text: "four hundred ninety-three"

   • Pre-processed text: "english to number: four hundred ninety-three"

   • Predicted output text: "4 9 3"