Categories: AI/ML Research

Pretraining a Llama Model on Your Local GPU

This article is divided into three parts; they are: • Training a Tokenizer with Special Tokens • Preparing the Training Data • Running the Pretraining The model architecture you will use is the same as the one created in the

Preparing Data for BERT Training

This article is divided into four parts; they are: • Preparing Documents • Creating Sentence Pairs from Document • Masking Tokens • Saving the Training Data for Reuse Unlike decoder-only models, BERT's pretraining is more complex.

November 25, 2025

In "AI/ML Research"

Creating a Llama or GPT Model for Next-Token Prediction

This article is divided into three parts; they are: • Understanding the Architecture of Llama or GPT Model • Creating a Llama or GPT Model for Pretraining • Variations in the Architecture The architecture of a Llama or GPT model is simply a stack of transformer blocks.

December 9, 2025

In "AI/ML Research"

Language Models Improve When Pretraining Data Matches Target Tasks

Every data selection method inherently has a target. In practice, these targets often emerge implicitly through benchmark-driven iteration: researchers develop selection strategies, train models, measure benchmark performance, then refine accordingly. This raises a natural question: what happens when we make this optimization explicit? To explore this, we propose benchmark-targeted ranking…

July 19, 2025

In "FAANG"

AI Generated Robotic Content