https://github.com/asrot0/polyglottransformer
A fully modular, notebook-powered Transformer model for translating across languages—including English, Bengali, Japanese, German, and more. Built for flexibility and global communication.
https://github.com/asrot0/polyglottransformer
ai deep-learning multilingual transformer
Last synced: 12 months ago
JSON representation
A fully modular, notebook-powered Transformer model for translating across languages—including English, Bengali, Japanese, German, and more. Built for flexibility and global communication.
- Host: GitHub
- URL: https://github.com/asrot0/polyglottransformer
- Owner: asRot0
- Created: 2025-04-10T12:00:21.000Z (about 1 year ago)
- Default Branch: main
- Last Pushed: 2025-05-18T12:04:20.000Z (about 1 year ago)
- Last Synced: 2025-06-07T03:07:26.075Z (about 1 year ago)
- Topics: ai, deep-learning, multilingual, transformer
- Language: Jupyter Notebook
- Homepage:
- Size: 14.1 MB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# PolyglotTransformer
> A multilingual Transformer-based Neural Machine Translation (NMT) system built with TensorFlow/Keras — modular, extendable, and production-ready.
**PolyglotTransformer** is a fully modular implementation of the Transformer architecture for sequence-to-sequence tasks, such as language translation. It supports language-specific notebooks and is designed for easy extensibility and experimentation.
- Built on TensorFlow 2 & Keras
- Clean modular codebase under `models/`
- Notebook-ready workflows for multilingual translation
- Supports English to Bengali, German, and more
```bash
PolyglotTransformer/
│
├── data/
├── models/
│ ├── __init__.py
│ ├── transformer_encoder.py # Transformer Encoder Block
│ ├── transformer_decoder.py # Transformer Decoder Block
│ ├── multi_head_attention.py # Multi-Head Self-Attention
│ ├── feed_forward.py # Position-wise Feed Forward Layer
│ ├── positional_encoding.py # Sinusoidal Positional Encoding
│ ├── positional_embedding.py # Token + Positional Embedding
│ └── transformer.py # Full Encoder-Decoder Transformer
│
├── eng_to_ban.ipynb # English ➜ Bengali notebook
├── eng_to_japanese.ipynb # English ➜ Japanese notebook
├── README.md
```
## Model Architecture
- Multi-head attention
- Position-wise feed-forward networks
- Layer normalization & residual connections
- Sinusoidal or learned positional embeddings
- Encoder–Decoder structure (as proposed in "Attention Is All You Need")
### 🗾 Translation Demo (English ➜ Japanese)
| 🌐 English Sentence | 🇯🇵 Japanese Translation |
|--------------------------------------------|-------------------------------------------------|
| Do you have any plans for today? | 今日の予定はありますか？ |
| I forgot to bring my umbrella. | 傘を持ってくるのを忘れました。 |
| Let's meet at the cafe at 3 p.m. | 午後3時にカフェで会いましょう。 |
| This movie was really interesting. | この映画は本当に面白かったです。 |
| I want to visit Japan someday. | いつか日本を訪れたいです。 |
### 🌍 Translation Demo (English ➜ Bengali)
| 🌐 English Sentence | 🇧🇩 Bengali Translation |
|--------------------------------------------|---------------------------------------------|
| What are you doing? | তুমি কী করছো? |
| I love learning new languages. | আমি নতুন ভাষা শিখতে ভালোবাসি। |
| Can you help me, please? | তুমি কি আমাকে সাহায্য করতে পারো? |
| This is my favorite book. | এটি আমার প্রিয় বই। |
| We need to leave early tomorrow morning. | আমাদের আগামীকাল সকালেই রওনা দিতে হবে। |
> 📝 *Translations generated using the trained Transformer model.*
## Extend to More Languages
Just copy an existing notebook like `eng_to_ban.ipynb` and change:
- The parallel corpus (`.txt`) file
- `source_vectorization` / `target_vectorization` setup
- Save as `eng_to_german.ipynb` or similar!
#### Recommended Dataset Source
You can download free English-to-X language sentence pairs from:
🔗 [https://www.manythings.org/anki/](https://www.manythings.org/anki/)
Example files:
- `eng-ger.txt` → English to German
- `eng-jpn.txt` → English to Japanese
- `eng-ita.txt` → English to Italian
- `eng-spa.txt` → English to Spanish
Just place the downloaded `.txt` file into your project and update the `load_data()` function accordingly!