Serving Fastchat - Personal Journey
Serving fastchat for people to experiment with various LLMs. This guide also incluides setting up Vllm to serve multiple models on a single GPU.
Posts by Year
2024
2023
Basics of Transformers and Huggingface - Datasets
A take on trying to help understand LLMs and Transformers - Now the dataset!
Basics of Transformers and Huggingface - Inference
A take on trying to help understand LLMs and Transformers - In a code first approach!
Arch Linux is the Best for Deep Learning (for intermediate users)
Arch linux makes it better to manage deep learning system, and understand the system better.
Elegance of Keras Core and JAX combo
Combining Keras and JAX as a backend, makes JAX to be meant for Humans
Maximizing HPC Potential: The Accelerators [Part 2]
Get those numbers crunching! Hardware accelerators are specialized computing devices designed to perform specific tasks more efficiently than general-purpose...
Maximizing HPC Potential: Unveiling the Best CPU, Motherboard, and Memory [Part 1]
Prepare to harness the immense power of high-performance computing (HPC) nodes. In Part 1 of my comprehensive series, I delve into the art of choosing the ul...
Stuff I learnt doing an end to end project
Deep Learning Lessons: Insights from an End-to-End Project. Gain valuable tips from my personal experiences in deep learning. Discover the power of thorough ...
Unleashing the Power of AI in Data Analysis with Sketch
Explore and Analyze Your Data with Sketch: In this blog post, we will explore Sketch - an AI-powered DataFrame assistant for Python that uses data sketching ...
Distributed Training With Tensorflow
Leveraging Multi Worker Mirrored Strategy in TensorFlow to train models across multiple workers using data parallelism.
Optimizing Deep Learning Models for Inference with Speedster
Nebuly-AI’s Speedster is a tool that can optimize deep learning models for inference on CPUs and GPUs.
Setting up Horovod on cluster
Setting and running Horovod on a PBS managed cluster
2022
Mixed Precision Training
Less memory more speeeeedddd. Training Models with mixed precision for lower memory footprint, and faster training.
LazyPredict
I am just too lazy to compare multiple Machine Learning algorithms.
Introduction to XAI (Explainable AI)
Chapter 01 of Applied Machine Learning Explainability Techniques book by Aditya Bhattacharya
U Net architecture for Image Segmentation
We will go through u-net architecture for image segmentation and its implementation in PyTorch.
Neural Style Transfer
Introduction