Genai_A3

Development of a generative AI model for English-to-Urdu machine translation, leveraging the UMC005 corpus, and evaluating CNN and Vision Transformer architectures for CIFAR-10 image classification.

English-to-Urdu Machine Translation and CIFAR-10 Image Classification

Project Description

This repository focuses on two key areas:

English-to-Urdu Machine Translation:
- Development of a generative AI model for translating English to Urdu using the UMC005 corpus.
- The project incorporates advanced NLP techniques, including preprocessing with SentencePiece Tokenizer and leveraging Transformer-based architectures for high-quality translation.
Image Classification on CIFAR-10:
- Comparative analysis of three deep learning architectures:
  - ResNet with Transfer Learning
  - Vision Transformer (ViT)
  - Hybrid CNN-MLP
- The models are trained and evaluated on the CIFAR-10 dataset, highlighting their performance in terms of accuracy and efficiency.

Key Features

English-to-Urdu Machine Translation

Uses the UMC005 Parallel Corpus for training and evaluation.
Implements preprocessing steps, including tokenization with SentencePiece.
Includes scripts for training, evaluation, and deployment of the Transformer model.

CIFAR-10 Image Classification

Explores performance differences across ResNet, Vision Transformer, and Hybrid CNN-MLP models.
Provides visualization of training loss and accuracy curves for all models.