Artificical Intelligence

LLM Inferencing Basics

LLM inference is an iterative process, where each new forward pass of the model generates one additional completion token. The initial ingestion (prefill) of the prompt takes as much time as generating each subsequent token. During the prefill phase, the model pre-computes certain inputs of the attention mechanism, specifically the key (k) and value (v) matrices, which remain constant for the remainder of the inference process.

Read more →

October 1, 2024

LLM Evaluation

LLM Benchmarks Coding Capabilities Knowledge and Language Understanding Reasoning Capabilities Multi-Turn Ended Conversations Content Moderations and Narrative Control

Read more →

September 1, 2024

LLM Basics

The key characteristic of Large Language Models (LLMs) is their ability to understand and generate human language. LLMs based on transformer architecture are mainly of three types i.e., Mask Language Model (predict the masked words based on the surrounding context), Causal Language Model (predict the next word in a sequence given the preceding words) and Seq-to-Seq Model (translation, summarization etc.)

Read more →

October 27, 2023

Generative Adversarial Networks

In this post, we’ll delve deep into Generative Adversarial Networks (GANs) architecture, its loss function and evaluation metrics. We’ll also explore the diverse types of GANs and its alternatives that have been pushing the boundaries of what’s possible in AI.

Read more →

October 24, 2023

LLM Verse

In this post, we’ll provide an overview of different types of Large Language Models (LLMs), focusing on encoder, decoder, and encoder-decoder architectures.

Read more →

October 1, 2023

Hypothesis Evaluation

Before making claims about the accuracy of a particular trained model or any hypothesis, it is crucial to conduct statistical significance tests. This ensures that the claimed accuracy is reliable and not just a result of chance. It is essential not only in the context of machine learning or deep learning models but also in various other experiments.

Read more →

April 1, 2023

Mining Massive Datasets

Data Downloading Data Preprocessing Data Postprocessing Data Deduplication Open Source Datasets C4 MC4 Refined Web Red Pajama Slim Pajama OSCAR Roots NLLB Pile Massive Dataset Dolma Dataset

Read more →

April 1, 2023

Reinforcement Learning : Tabular Solution Methods

Multi-armed Bandits The multi-armed bandit problem serves as an introductory concept in reinforcement learning, much like how the house pricing problem acts as a starting point in machine learning. In the K-armed bandit problem, an agent is presented with a set of ‘k’ actions to choose from, and the agent’s decision results in a corresponding reward. Value of an action is the expected or mean reward given when that action is selected.

Read more →

September 2, 2020

Reinforcement Learning : Fundamentals

Reinforcement learning is the process of discovering what to do in order to maximise a numerical reward signal. A reinforcement learning system’s primary sub-components are policy, reward signal, value function, agent, and environment.

Read more →

September 1, 2020

Tokenizers

Fast Tokenizers Slow Tokenizers Byte Pair Encodings Word Piece Unigram Sentence Piece

Read more →

April 1, 2020

Machine Learning Basics

Machine Learning is the field of study that gives computer the ability to learn without being explicitly programmed. A computer program is said to learn from experience E with respect ot some task T and some performance measure P, if its performance on T , as measured by P, improves with experience E. Machine Learning Algorithms: Supervised Learning Regression (continuous output) Classification (discrete valued output) Unsupervised Learning Linear Regression Linear Regression with one variable

Read more →

July 1, 2017