rahmanazhar/Travereel-Model-V1

T5 Itinerary Generator

A custom fine-tuned version of FLAN-T5 for generating detailed travel itineraries.

Model Description

This model is fine-tuned from Google's FLAN-T5 to specialize in generating detailed travel itineraries based on user preferences, destinations, duration, and budget constraints.

Intended Use

• Generate detailed day-by-day travel itineraries
• Provide activity suggestions based on preferences
• Consider budget constraints in planning
• Include practical travel details

Training Data

The model is trained on a curated dataset of travel itineraries, including:

• Various destinations worldwide
• Different trip durations
• Various travel preferences and styles
• Different budget ranges

Prerequisites

1. Python 3.8 or higher
2. CUDA-capable GPU (8GB+ VRAM recommended)
3. Hugging Face account and token

Setup

1. Create a virtual environment:

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

2. Install dependencies:

pip install -r requirements.txt

3. Configure your Hugging Face token:

huggingface-cli login

Project Structure

.
├── config/
│   └── config.json         # Training configuration
├── data/
│   └── itineraries.json    # Training data
└── src/
    └── train.py           # Training script

Training Data

The training data in data/itineraries.json contains examples of travel itineraries with the following structure:

• Destination
• Duration
• Preferences
• Budget
• Detailed day-by-day itinerary

Training the Model

1. Review and adjust the configuration in config/config.json if needed.

2. Start training:

python src/train.py

The script will:

• Load the LLaMA-2 base model
• Fine-tune it on the itinerary dataset
• Save checkpoints during training
• Export the final model

Model Details

This model is fine-tuned to generate travel itineraries based on:

• Destination
• Duration of stay
• Travel preferences
• Budget constraints

The model learns to:

• Structure day-by-day itineraries
• Balance activities based on preferences
• Consider budget constraints
• Include practical details like transportation and check-in/out

Output Format

The model generates itineraries in a structured format:

Day 1:
- Activity 1
- Activity 2
...

Day 2:
- Activity 1
- Activity 2
...

Monitoring Training

Training progress can be monitored using TensorBoard:

tensorboard --logdir output/runs

Model Deployment

After training, the model will be saved in the output directory. You can upload it to Hugging Face Hub using:

huggingface-cli upload rahmanazhar/Travereel-Model-V1 output/

License

This project uses FLAN-T5 which is licensed under the Apache 2.0 License.