MLOps Assignments
Deep Learning & Machine Learning Operations
Assignments
Deep Learning & SVM Classification
Training ResNet architectures and SVM classifiers on MNIST & FashionMNIST datasets. Includes hyperparameter tuning, CPU vs GPU performance analysis, and FLOPs computation.
CNN Training on CIFAR-10
Training a SimpleCNN model on CIFAR-10 with custom dataloader, FLOPs counting, gradient flow visualization, and weight update tracking using Weights & Biases.
HuggingFace Model Training & Docker Deployment
Fine-tuning DistilBERT on Goodreads book reviews for genre classification. End-to-end workflow: notebook conversion, Trainer API training, HuggingFace Hub deployment, and Docker containerization.
Optimizing Transformer Translation with Ray Tune & Optuna
Refactored the custom English→Hindi Transformer into a reusable tuning pipeline with Ray Tune, Optuna, and ASHA to search faster-converging hyperparameters, beat the notebook BLEU baseline, and export the final report plus model artifacts.
ViT-S LoRA Fine-tuning & Adversarial Attacks with IBM ART
Fine-tuned ViT-Small on CIFAR-100 with LoRA (PEFT) across 9 rank/alpha configurations plus Optuna hyperparameter search. Implemented FGSM attacks from scratch vs IBM ART on CIFAR-10, and trained ResNet-34 adversarial detectors for PGD and BIM attacks.
About
This repository contains my coursework for the MLOps (Machine Learning Operations) course. Each assignment explores different aspects of deploying, optimizing, and managing machine learning models.