Pratik's Blog

Sign in Subscribe

Showboating: Extracting Training Data from LLMs

Showboating: Extracting Training Data from LLMs

Competition is for losers.

Distributing Key Values

Distributed Systems

Distributing Key Values

Consistency is a virtue of mules?

"Serverless" GPU

Machine Learning

"Serverless" GPU

Need...more.....power

Two Birds

Night had begun to envelop Dholavira, as we trudged back from the site of an ancient settlement, weary from a day's worth of travel and exploration. An interesting question loomed before us - one we had been staving off for later. Find a place to stay. There was

PipeDream: The Dreamweaver of Distributed Training

PipeDream: The Dreamweaver of Distributed Training

The authors of PipeDream set out to solve an ambitious problem - how do you train a colossal model on a distributed cluster? Dividing the vast dataset required to support model training is certainly needed. With huge models, we may even need to segment the model itself into various stages.

MapReduce: Distributed Computing For All

MapReduce: Distributed Computing For All

Computer science is the science of abstractions. So much simply happens because a group of developers somewhere set out to abstract out a complex process for developers everywhere. But there's still a balancing act. Too many abstractions introduces too many overheads - too few and we're

Google File System: Chunk and Conquer

Google File System: Chunk and Conquer

The Google File System is a widely studied distributed file system. Its ability to provide a scalabily and fault tolerance on inexpensive commodity hardware made it a platform of choice within Google. The design goals of this file system was similar to its predecessors - performance, scalability, reliability and availability

Simplify Law: Using RAGs to Unleash Legal Discovery

Simplify Law: Using RAGs to Unleash Legal Discovery

Diving into the rich tapestry of India's legislative history, we can see a clear vision that our leaders have for our nation. In an attempt to promote transparency, this project aims pave way for LLMs to answer questions on Indian law.

Langchain Unchained: From Prompt to Possibilities

Langchain Unchained: From Prompt to Possibilities

Redefining LLM interactions with LangChain

That which unites and divides

Machine Learning

That which unites and divides

National anthems reflect collective identity and history. In this blog lets examine similarities and differences between them

We Are The World

We Are The World

You are not a drop in the ocean. You are the entire ocean in a drop Rumi Quick history lesson: When did the very concept of a nation state and all its trappings come to be? That would be the Peace of Westphalia in 1648. This landmark event formalized political

ThreadCraft: Building a Proxy Server

ThreadCraft: Building a Proxy Server

Designing a multi-threaded proxy server that prioritizes requests and ensures a robust, scalable system for serving static web pages

Shell We C: Building Terminals

Shell We C: Building Terminals

Explore the journey of building a powerful C-based shell, from process handling to executing commands with piping.

Git Magic - From Chaos to Ordered Collaboration

Git Magic - From Chaos to Ordered Collaboration

Why Git? Once the why is clear, the how is easy This is easy. We want to be able to take snapshots of our code at critical junctures of the development cycle (read as undoing the mess we'll eventually create later on). We'd also like collaborating

The Battle for Ethical Intelligence⚖️🤖

Machine Learning

The Battle for Ethical Intelligence⚖️🤖

AI's future splits: colossal models from tech giants vs. sustainable pathways with large language models. Societal challenges abound, from wealth shifts to research migration. Amidst this, several issues guide our journey through technological advancement.

Transformers Explained: Part I

Transformers Explained: Part I

Transformers-the quintessential panacea to sequence models.

Multi-Label Genre Classification-Part III

Multi-Label Genre Classification-Part III

Deploying the saved movie classifier

Multi-Label Genre Classification-Part II

Multi-Label Genre Classification-Part II

In this part, we'll be looking at some more preprocessing and BERT model training using the Huggingface library

Multi-Label Genre Classification-Part I

Multi-Label Genre Classification-Part I

Unleash the power of data preprocessing for accurate multi-label genre classification in NLP. From data wrangling to feature engineering, discover the essential steps to optimize your models.

Multi-Label Genre Classification

Multi-Label Genre Classification

In this 3 part series, we'll be looking to clean, preprocess, train and deploy a model for identifying the genres from a movie summary. We'll keep this one short.

RISC V Pipelined Processor

computer architecture

RISC V Pipelined Processor

In this blog post, we will explore the implementation of a pipelined RISC-V processor using Verilog, a hardware description language.

RISC V Single Cycle Processor

RISC V Single Cycle Processor

In the world of computer architecture, the single-cycle processor is a testament to simplicity and efficiency. We'll be looking at the datapath and the control unit for a single cycle processor with two classes of instructions Immediate Type Example li, r1, constant ➡️loads immediate signed value specified in