---
license: apache-2.0
language:
- uk
- en
---

# ukr-t5-small

A compact T5-small model fine-tuned for Ukrainian language tasks, with base English understanding.

## Model Description

* **Base Model:** mT5-small 
* **Fine-tuning Data:** Leipzig Corpora Collection (English & Ukrainian news from 2023)
* **Tasks:**  
    * Text summarization (Ukrainian)
    * Text generation (Ukrainian)
    * Other Ukrainian-centric NLP tasks

## Technical Details
* **Model Size:** 300 MB 
* **Framework:** Transformers (Hugging Face)

## Usage 

**Installation**

```bash
pip install transformers
```

**Loading the Model**

```python
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

tokenizer = AutoTokenizer.from_pretrained("path/to/ukr-t5-small")
model = AutoModelForSeq2SeqLM.from_pretrained("path/to/ukr-t5-small")
```

**Example: Machine Translation**

```python
text = "(Text in Ukrainian here)"

# Tokenize and translate
inputs = tokenizer("summarize: " + text, return_tensors="pt", max_length=512, truncation=True)
summary_ids = model.generate(inputs["input_ids"], num_beams=4, max_length=128)

# Decode output 
summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True)
print(summary)
```

## Limitations

* The model's focus is on Ukrainian text processing, so performance on purely English tasks may be below that of general T5-small models.
* Further fine-tuning may be required for optimal results on specific NLP tasks.

## Dataset Credits

This model was fine-tuned on the Leipzig Corpora Collection (specify if there's a particular subset within the collection that you used). For full licensing and usage information of the original dataset, please refer to [Leipzig Corpora Collection website](https://wortschatz.uni-leipzig.de/en/download)

## Ethical Considerations

* NLP models can reflect biases present in their training data. Be mindful of this when using this model for applications that have real-world impact.
* It's important to test this model thoroughly across a variety of Ukrainian language samples to evaluate its reliability and fairness.