| # Text-to-Text Transfer Transformer Quantized Model for Drug Reports Summarization | |
| This repository hosts a quantized version of the T5 model, fine-tuned for text summarization tasks. The model has been optimized for efficient deployment while maintaining high accuracy, making it suitable for resource-constrained environments. | |
| ## Model Details | |
| - **Model Architecture:** T5 | |
| - **Task:** Drug Report Summarization | |
| - **Dataset:** Hugging Face's `cnn_dailymail' | |
| - **Quantization:** Float16 | |
| - **Fine-tuning Framework:** Hugging Face Transformers | |
| ## Usage | |
| ### Installation | |
| ```sh | |
| pip install transformers torch | |
| ``` | |
| ### Loading the Model | |
| ```python | |
| from transformers import T5Tokenizer, T5ForConditionalGeneration | |
| import torch | |
| device = "cuda" if torch.cuda.is_available() else "cpu" | |
| model_name = "AventIQ-AI/t5-summarization-for-drug-reports" | |
| tokenizer = T5Tokenizer.from_pretrained(model_name) | |
| model = T5ForConditionalGeneration.from_pretrained(model_name).to(device) | |
| def test_summarization(model, tokenizer): | |
| user_text = input("\nEnter your text for summarization:\n") | |
| input_text = "summarize: " + user_text | |
| inputs = tokenizer(input_text, return_tensors="pt", truncation=True, max_length=512).to(device) | |
| output = model.generate( | |
| **inputs, | |
| max_new_tokens=100, | |
| num_beams=5, | |
| length_penalty=0.8, | |
| early_stopping=True | |
| ) | |
| summary = tokenizer.decode(output[0], skip_special_tokens=True) | |
| return summary | |
| print("\n📝 **Model Summary:**") | |
| print(test_summarization(model, tokenizer)) | |
| ``` | |
| # 📊 ROUGE Evaluation Results | |
| After fine-tuning the **T5-Small** model for text summarization, we obtained the following **ROUGE** scores: | |
| | **Metric** | **Score** | **Meaning** | | |
| |-------------|-----------|-------------| | |
| | **ROUGE-1** | **0.3061** (~30%) | Measures overlap of **unigrams (single words)** between the reference and generated summary. | | |
| | **ROUGE-2** | **0.1241** (~12%) | Measures overlap of **bigrams (two-word phrases)**, indicating coherence and fluency. | | |
| | **ROUGE-L** | **0.2233** (~22%) | Measures **longest matching word sequences**, testing sentence structure preservation. | | |
| | **ROUGE-Lsum** | **0.2620** (~26%) | Similar to ROUGE-L but optimized for summarization tasks. | | |
| ## Fine-Tuning Details | |
| ### Dataset | |
| The Hugging Face's `cnn_dailymail` dataset was used, containing the text and their summarization examples. | |
| ### Training | |
| - Number of epochs: 3 | |
| - Batch size: 4 | |
| - Evaluation strategy: epoch | |
| - Learning rate: 3e-5 | |
| ### Quantization | |
| Post-training quantization was applied using PyTorch's built-in quantization framework to reduce the model size and improve inference efficiency. | |
| ## Repository Structure | |
| ``` | |
| . | |
| ├── model/ # Contains the quantized model files | |
| ├── tokenizer_config/ # Tokenizer configuration and vocabulary files | |
| ├── model.safetensors/ # Quantized Model | |
| ├── README.md # Model documentation | |
| ``` | |
| ## Limitations | |
| - The model may not generalize well to domains outside the fine-tuning dataset. | |
| - Quantization may result in minor accuracy degradation compared to full-precision models. | |
| ## Contributing | |
| Contributions are welcome! Feel free to open an issue or submit a pull request if you have suggestions or improvements. | |