🚀 Model API with Flask, BART Base & mBART

A Flask-based API service implementing BART Base and mBART (Multilingual BART) model for online article text summarization with customizable generation parameters. This API provides two different models for text summarization with extensive parameter customization options.

🌟 Features

• Flask REST API for text summarization
• BART Model Integration for test the quality betwen BART base and mBART
• mBART Model Integration for high-quality multilingual summarization
• Customizable Generation Parameters for different summarization styles
• Dynamic Length Control based on input text
• Two Generation Modes for different use cases
• Evaluation Metrics for summary quality assessment
• Batch Processing support for multiple summaries
• Error Handling with detailed error messages

🛠️ API Endpoints

BART Model Endpoints

1. Generate Summary (BART)

POST /bart-summarize

Request Body:

{
  "text": "Your long text to summarize",
  "params": {
    "length_penalty": 2.0,
    "num_beams": 4,
    "no_repeat_ngram_size": 2,
    "temperature": 0.8,
    "top_k": 50,
    "top_p": 0.9
  }
}

Note: params object is optional. Default values will be used if not provided.

Response:

{
  "summary": "Generated summary text"
}

2. Evaluate Summary (BART)

POST /bart-evaluate

Request Body:

{
  "reference_text": "Original reference summary",
  "generated_summary": "Generated summary to evaluate"
}

Response:

{
  "metrics": {
    "precision": 85.5,
    "recall": 78.3,
    "f1": 81.7,
    "bleu": 72.4
  }
}

mBART Model Endpoints

1. Generate Summary (mBART)

POST /mbart-summarize

Request Body:

{
  "text": "Your long text to summarize",
  "params": {
    "num_beams": 4,
    "length_penalty": 1.5,
    "num_beam_groups": 4,
    "diversity_penalty": 0.5
  }
}

2. Evaluate Summary (mBART)

POST /mbart-evaluate

Request Body: Same as BART evaluate endpoint

Batch Processing Endpoints

1. Batch Evaluation (BART)

POST /bart-evaluate-batch

Request Body:

{
  "reference_texts": ["reference1", "reference2", "..."],
  "generated_summaries": ["summary1", "summary2", "..."]
}

2. Batch Evaluation (mBART)

POST /mbart-evaluate-batch

🛠️ Generation Parameters

Basic Parameters

Parameter	Description	Default Value	Valid Range
min_length	Minimum length of summary	Dynamic (1/4 of input)	5-15 words
max_length	Maximum length of summary	Dynamic (1/2 of input)	30-150 words
num_beams	Number of beams for search	4	1-10
early_stopping	Stop when valid output found	True	boolean
temperature	Randomness in generation	0.6-0.8	0.0-1.0
top_k	Top K sampling parameter	50	1-100
top_p	Nucleus sampling parameter	0.9	0.0-1.0
repetition_penalty	Penalty for repeating words	1.5-2.5	1.0-3.0

Advanced Parameters

BART Specific

Parameter	Description	Default Value
length_penalty	Controls summary length	2.0
no_repeat_ngram_size	Prevents repetition	2
do_sample	Enable sampling	True

mBART Specific

Parameter	Description	Default Value
num_beam_groups	Groups for diverse beams	4
diversity_penalty	Penalty for similar beams	0.5
forced_bos_token_id	Beginning token ID	None

🎮 Usage Examples

Basic API Call

import requests

url = "http://your-api-endpoint/bart-summarize"  # or mbart-summarize
text = "Your long text here..."

response = requests.post(url, json={
    "text": text
})

print(response.json()["summary"])

Advanced Parameters

response = requests.post(url, json={
    "text": text,
    "params": {
        "length_penalty": 2.0,
        "num_beams": 6,
        "no_repeat_ngram_size": 3,
        "temperature": 0.8,
        "top_k": 50,
        "top_p": 0.9,
        "repetition_penalty": 2.5
    }
})

Evaluation Example

eval_url = "http://your-api-endpoint/bart-evaluate"
eval_response = requests.post(eval_url, json={
    "reference_text": "Original summary",
    "generated_summary": "Generated summary"
})

print(eval_response.json()["metrics"])

🔧 Installation

1. Clone the repository

git clone https://github.com/Dev-putra12/model-api.git
cd model-api

2. Install dependencies

pip install -r requirements.txt

3. Run the Flask server

python run.py

📝 Notes

• The API automatically adjusts parameters based on input text length
• Two generation modes available for different use cases:
  • Deterministic mode for consistent results
  • Creative mode for more varied outputs
• Parameters can be fine-tuned based on specific needs
• Error handling includes detailed messages for troubleshooting
• Evaluation metrics provide quantitative quality assessment
• Batch processing available for multiple documents

🔍 Error Handling

The API returns appropriate HTTP status codes and error messages:

• 400: Bad Request (invalid parameters or input)
• 500: Internal Server Error (model processing error)
• 404: Endpoint Not Found

Example error response:

{
  "error": "Text cannot be empty",
  "status": 400
}

🤝 Contributing

Feel free to open issues and pull requests for:

• Bug fixes
• New features
• Documentation improvements
• Parameter optimization suggestions
• Test cases and examples

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

📚 References

• BART Paper
• mBART Paper
• Flask Documentation
• Hugging Face Transformers

---

Made with ❤️ for text summarization