Trained Tiny Tales GPT (30 million parameter) model from scratch and deployed it in production for $15
Tiny Tales GPT is a 30 million-parameter language model trained on 1 billion tokens from scratch. The training was done using Distributed Data-Parallel on two A-100 GPUs
After the training is done, an inference script is created to predict the tokens from the trained model given the input context vector.
Developed REST-based API service using Flask framework to interact with the inference service to the end user and deployed the web service in Google Cloud Platform
AWS-Hosted Flask Microservice Ecosystem
Engineered a Flask-driven microservice ecosystem on AWS, leveraging EC2 with AMI for rapid deployment, alongside S3, RDS, ELB, Lambda, SES, and SNS for a resilient, scalable infrastructure.
Image Recognition Model for American Sign Language
Image Recognition Model for American Sign Language: Developed a hybrid CNN and ResNet50 model, optimized for ASL gesture classification, leveraging PyTorch's DDP for multi-GPU training.
Image Processing application with GUI
Developed an advanced, interactive GUI and command-line application using Java SDK and JUnit, supporting extensive image enhancement and manipulation. Integrated command design pattern and object-oriented architecture for scalable, efficient code structure.
Context-based Question Answering model
Build a seq-seq automatic answer-generating model given the context and question on a collection of 100k datapoint using Bidirectional GRUs and fine-tuning.
Implemented Convolutional and Maxpool layer using Numpy
Build these layers from ground up. The results we got from custom operations were the same as Tensorflow's and Pytorch's output of Conv2d layer.
Parallel Image Classification on multiple GPUs using Pytorch
Performed cleaning, hyperparameter tuning and training of 90k images parallely on 4 GPUs of high-performance computer using Pytorch.
Data Structures and Algorithms- LeetCode top 100 questions
A GitHub repo for the top 100 liked questions . The solutions are written in C++ and Python language. The approach is described in comments.
Kunal Mishra
kunalmishra78@gmail.com
