README.md

---
language: en
tags:
- google/mt5-small
- machine_translation
license: apache-2.0
---

# Hindi-English Translation Model

Based on the "google/mt5-small" pre-trained model. Fine-tuned it on Hindi to English dataset. 

### Parameters
- number of epochs = 8
- batch size = 16
- learning rate = 5e-4
- number of batches = int(np.ceil(len(dataset) / batch size))
- n_warmup_steps = int(number of epochs * number of batches * 0.01)

### Training Loss

![loss.png](./h2e_loss.png)

### Examples

![example_1.png](./h2e_example1.png)
![example_2.png](./h2e_example2.png)
![example_3.png](./h2e_example3.png)