AventIQ-AI/t5-paraphrase-generation

Paraphrase Generation with Text-to-Text Transfer Transformer

📌 Overview

This repository hosts the quantized version of the T5 model fine-tuned for Paraphrase Generation. The model has been trained on the chatgpt-paraphrases dataset from HF中国镜像站 to enhance grammatical accuracy in given text inputs. The model is quantized to Float16 (FP16) to optimize inference speed and efficiency while maintaining high performance.

🏗 Model Details

• Model Architecture: t5-small
• Task: Paraphrase Generation
• Dataset: HF中国镜像站's chatgpt-paraphrases
• Quantization: Float16 (FP16) for optimized inference
• Fine-tuning Framework: HF中国镜像站 Transformers

🚀 Usage

Installation

pip install transformers torch

Loading the Model

from transformers import T5Tokenizer, T5ForConditionalGeneration, pipeline
import torch

device = "cuda" if torch.cuda.is_available() else "cpu"

model_name = "AventIQ-AI/t5-paraphrase-generation"
model = T5ForConditionalGeneration.from_pretrained(model_name).to(device)
tokenizer = T5Tokenizer.from_pretrained(model_name)

Grammar Correction Inference

paraphrase_pipeline = pipeline("text2text-generation", model=model, tokenizer=tokenizer)
test_text = "The quick brown fox jumps over the lazy dog"

# Generate paraphrases
results = paraphrase_pipeline(
    test_text,
    max_length=256,
    truncation=True,
    num_return_sequences=5,
    do_sample=True,
    top_k=50,
    temperature=0.7
)

print("Original Text:", test_text)
print("\nParaphrased Outputs:")

for i, output in enumerate(results):
    generated_text = output["generated_text"] if isinstance(output, dict) else str(output)
    print(f"{i+1}. {generated_text.strip()}")

📊 ROUGE Evaluation Results

After fine-tuning the T5-Small model for paraphrase generation, we obtained the following ROUGE scores:

Metric	Score	Meaning
ROUGE-1	0.7777 (~78%)	Measures overlap of unigrams (single words) between the reference and generated summary.
ROUGE-2	0.5 (~50%)	Measures overlap of bigrams (two-word phrases), indicating coherence and fluency.
ROUGE-L	0.7777 (~78%)	Measures longest matching word sequences, testing sentence structure preservation.
ROUGE-Lsum	0.7777 (~78%)	Similar to ROUGE-L but optimized for summarization tasks.

⚡ Quantization Details

Post-training quantization was applied using PyTorch's built-in quantization framework. The model was quantized to Float16 (FP16) to reduce model size and improve inference efficiency while balancing accuracy.

📂 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 struggle with highly ambiguous sentences.
• Quantization may lead to slight degradation in accuracy compared to full-precision models.
• Performance may vary across different writing styles and sentence structures.

🤝 Contributing

Contributions are welcome! Feel free to open an issue or submit a pull request if you have suggestions or improvements.