https://github.com/mohamedmostafa259/emoji-translator

A fine-tuned BART-Large model that translates English text into expressive emoji sequences using Curriculum Learning and LoRA.
https://github.com/mohamedmostafa259/emoji-translator

bart emoji english fine-tuning huggingface kaggle lora peft

Last synced: about 2 months ago
JSON representation

A fine-tuned BART-Large model that translates English text into expressive emoji sequences using Curriculum Learning and LoRA.

• Host: GitHub
• URL: https://github.com/mohamedmostafa259/emoji-translator
• Owner: MohamedMostafa259
• License: apache-2.0
• Created: 2025-12-12T09:30:43.000Z (6 months ago)
• Default Branch: main
• Last Pushed: 2026-04-22T12:15:14.000Z (about 2 months ago)
• Last Synced: 2026-04-22T14:17:56.591Z (about 2 months ago)
• Topics: bart, emoji, english, fine-tuning, huggingface, kaggle, lora, peft
• Language: Python
• Homepage: https://huggingface.co/spaces/mohamedmostafa259/emoji-translator-demo
• Size: 114 KB
• Stars: 0
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# 🎭 BART Emoji Translator

> **Turn your text into emojis!**  

> A fine-tuned BART-Large model that translates English text into expressive emoji sequences using Curriculum Learning and LoRA.

![Emoji Translator Demo](https://img.shields.io/badge/Demo-HuggingFace%20Space-orange) 

![Model](https://img.shields.io/badge/Model-BART%20Large-blue)

![License](https://img.shields.io/badge/License-Apache%202.0-green)

[![Emoji Translator Demo](https://raw.githubusercontent.com/MohamedMostafa259/emoji-translator/refs/heads/main/app-screenshot.png)](https://huggingface.co/spaces/mohamedmostafa259/emoji-translator-demo)

*Click the image above to try the live demo!*

## Overview

The **Emoji Translator** is a sequence-to-sequence model designed to "translate" English sentences into appropriate emoji representations. Unlike simple keyword matching, this model understands context and sentiment to generate meaningful emoji sequences.

It was fine-tuned using **Curriculum Learning**, starting from simple concept-to-emoji mappings and progressively moving to complex, multi-sentence emotional contexts.

### Key Features

- **Base Model**: `facebook/bart-large`

- **Technique**: Low-Rank Adaptation (LoRA) for efficient fine-tuning.

- **Training Strategy**: 6-Stage Curriculum Learning with memory retention (preventing catastrophic forgetting).

- **Dataset**: Custom dataset generated with **Gemini 3 Pro**, ensuring high-quality and diverse examples.

## Links

- **Live Demo**: [HuggingFace Space](https://huggingface.co/spaces/mohamedmostafa259/emoji-translator-demo)

- **Model Weights**: [HuggingFace Model Hub](https://huggingface.co/mohamedmostafa259/bart-emoji-translator)

- **Dataset**: [Kaggle Dataset](https://www.kaggle.com/datasets/mohamedmostafa259/english-to-emoji)

- **Training Notebook**: [Kaggle Notebook](https://www.kaggle.com/code/mohamedmostafa259/emoji-translator-curriculum-learning)

## Infrastructure

- Trained on **Kaggle** using 2x T4 GPUs.

- **WandB** used for experiment tracking.