README.md

---
language: en
license: mit
tags:
- text2text-generation
- multitask
- genre-classification
- rating-prediction
- title-generation
- t5
metrics:
- accuracy
- rmse
- bleu
base_model: google/t5-small
pipeline_tag: text2text-generation
library_name: transformers
---

# T5 Multitask Model for Book Genre, Rating, and Title Tasks

This model was trained on a custom dataset of book descriptions and titles. It supports:

- `genre:` → classify the genre of a book  
- `rating:` → predict the numeric rating  
- `title:` → generate a book title

---

## Usage

```python
from transformers import T5Tokenizer, T5ForConditionalGeneration

model = T5ForConditionalGeneration.from_pretrained("AbrarFahim75/t5-multitask-book")
tokenizer = T5Tokenizer.from_pretrained("AbrarFahim75/t5-multitask-book")

input_text = "genre: A dark and stormy night in an abandoned castle."
inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**inputs)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

---

## Model Details

- **Base model**: [google/t5-small](https://huggingface.co/google/t5-small)
- **Language**: English
- **Model type**: T5 fine-tuned on multi-task dataset (genre, rating, title)
- **License**: MIT
- **Author**: [AbrarFahim75](https://huggingface.co/AbrarFahim75)
- **Repository**: [t5-multitask-book](https://huggingface.co/AbrarFahim75/t5-multitask-book)

---

## Training Details

- **Data source**: Custom CSV with columns: `title`, `description`, `genre`, `rating`
- **Preprocessing**: Merged title and description → formatted prompts like:
  - `"genre: <desc>"`
  - `"rating: <desc>"`
  - `"title: <desc>"`
- **Epochs**: 3
- **Optimizer**: AdamW
- **Batch size**: 8
- **Loss**: Cross-entropy

---

## Evaluation

| Task              | Metric   | Value (sample, dev split) |
|-------------------|----------|----------------------------|
| Genre Classification | Accuracy | ~0.78 (sample set)        |
| Rating Prediction     | RMSE     | ~0.42                     |
| Title Generation      | BLEU     | ~15.3                     |

> ⚠️ These are informal evaluations using validation slices from the dataset.

---

## Intended Use

### Direct Use:
- Classifying book genres from text
- Predicting numeric ratings from descriptions
- Auto-generating book titles

### Out-of-Scope Use:
- Non-book-related input
- Use in high-stakes recommendation without human review

---

## Limitations and Biases

- Trained on a limited dataset of books (genre/bias unknown)
- May underperform on texts outside typical fiction/non-fiction boundaries
- Language is English only

---

## Citation

If you use this model, please cite:

```bibtex
@misc{fahim2025t5bookmultitask,
  title={T5 Multitask for Book Tasks},
  author={Md Abrar Fahim},
  year={2025},
  url={https://huggingface.co/AbrarFahim75/t5-multitask-book}
}
```

---

## Contact

For questions, please reach out at [huggingface.co/AbrarFahim75](https://huggingface.co/AbrarFahim75)