Jupiter Notebook
Basics of Triton
Triton is a language and a compiler for parallel programming in python. This helps in writing kernel level code to execute in parallel for Deep Neural Networks
As a Engineer
Personal blogging
Creating a piece of what I have learned so far in software industry. Listing down few blogs I wanted to share with people
I will be posting content here regularly. Please visit for more content later.
Jupiter Notebook
Custom dataloader from scratch
In the world of machine learning coming late has a few disadvantages and few advantages. Advantages is you have different approaches of people designing models in there own way. There are a lot of duplicate libraries across tools and platforms.
As a Data Uploader
Jupiter Notebook
Transformer++ (as per 2024)
This is based in the changes which have happened over time on the current architecture of a transformer. For me this is an improvement over the older architecture
As a Engineer
Jupiter Notebook
Python Code Generation From Scratch
I am trying to build an LLM in python by just letting it go through few github repos. This is just feeding raw python code into an DECODER of a TRANSFORMER.
As a Engineer
Jupiter Notebook
Mini-Unet
Unet is one of the milestones of vision modelling. This brought revolution in the world of CNN modelling. It works on the basis of downsampling the parameters as any CNN network but upsampling with a crop and concatenated layer of the downsampled parameters together with the upsampled parameters at each step of upsampling.
As a Probabilist
Jupiter Notebook
Sequential Monte Carlo
When we have data coming in a form of sequence, we try to evaluate the next T steps of the data. This help us see the future across multiple paths created using the samplers.
As a Probabilist
Jupiter Notebook
Mini-AlexNet
Alexnet was created for ImageNet images of size 3x244x244(cropped & rescaled) which gave output of size 1000. It could classify images of 1000 types. It used relu & dropouts.
As a CNN Modeller
Jupiter Notebook
Truncated SVD
To understand Truncated SVD. We need to understand the EVD and SVD.
As a Linear Algebraian
Jupiter Notebook
Basics of ML
Probability
Probability is a very huge topic to dicuss in a single blog. Here I am just going to name few things to remember while dicussing models using probablistic approach.
As a Probabilist
Jupiter Notebook
Probablistic Approach to Modelling
Lets start with something we are sure about. We know that when we train a model with dataset D we try to minimize some parameter suppose E. So if we write it in equation form, this becomes
As a Probablistic Modeller
