ONNX-NLP-DNN-Perf

🚀 Project Overview

This project demonstrates the importance of integrating tokenization directly into ONNX-based Deep Neural Network (DNN) models for NLP tasks. By doing so, we improve inference speed, optimize memory usage, and eliminate the need for separate preprocessing steps.

📌 Why This Project?

• Traditional DNN models (like OpenCV's DNN module) do not handle textual data.
• ONNX bridges the gap, enabling NLP models to work efficiently within a DNN pipeline.
• By integrating tokenization inside the model, we achieve faster inference and better memory optimization.

🛠️ Required Libraries

Ensure you have the following dependencies installed before running the project:

pip install onnxruntime transformers datasets numpy opencv-python

📂 How to Set Up & Run

1. Clone the repository:

   [git clone https://github.com/your-username/ONNX-NLP-DNN-Perf.git](https://github.com/PramodBasavarajMenasi/ONNX-NLP-DNN-Perf-.git)
   cd ONNX-NLP-DNN-Perf

📊 Results

Scenario	Execution Time	Memory Usage
Before Tokenization Integration	2.5495 seconds	53.31 KB
After Tokenization Integration	0.3809 seconds	46.17 KB

📢 Key Takeaways

• Without tokenization inside the model: Higher execution time and memory overhead.
• With tokenization inside the model: Faster inference and reduced memory usage.