DataBrew

In the rapidly evolving field of natural language processing (NLP), training large and powerful language models often requires significant computational resources that may be out of reach for individuals or small teams.  However, NVIDIA's NEMO toolkit offers an accessible solution for building and training conversational AI models, including large language models (LLMs), without the need for expensive hardware. NEMO, a toolkit designed to simplify the process of building and training AI models for tasks such as language understanding, text generation, and speech recognition. With its modular and extensible architecture, NEMO provides a collection of pre-built components and utilities that can be combined and customized to suit your specific needs. In this article, we'll explore the different components of NEMO and how they can be used to train large language models. We'll cover topics such as data preprocessing with the Data Curator, model definition and configuration, writ...

Tokens: When using text-based models like GPT, input text is split into tokens for processing. For instance, the sentence "I enjoy reading books" would be tokenized into individual units like "I," "enjoy," "reading," and "books." Each token typically consists of 4 English characters. Example: Input Text: "I enjoy reading books" Tokens: ["I", "enjoy", "reading", "books"]Pricing for these models is based on the number of tokens processed. OpenAI charges based on the total input and output tokens generated during interactions with the model.   How to Specify Token Count in API Response: When interacting with an API, you can specify the number of tokens you want in the response using the max_tokens parameter. For example, setting max_tokens=20 would request a response containing up to 20 tokens.   Example: API Request: Generate a summary of a book with max_tokens=20 The API would then provid...

In recent years, containerization has emerged as a game-changer in software development and deployment. While Docker has dominated the containerization landscape in many industries, its suitability for high-performance computing (HPC) environments is debatable. Enter Singularity, a container platform purpose-built for HPC workloads. In this post, we'll explore how Singularity differs from Docker and why it's the go-to solution for HPC users. Key Differentiators Between Singularity and Docker for HPC: Security : Singularity prioritizes security by enabling non-privileged users to create and execute containers, eliminating the need for root access and enhancing user isolation—a crucial aspect in multi-tenant HPC environments. Performance : Singularity leverages the host system's kernel, ensuring seamless integration with HPC infrastructure and minimizing performance overhead compared to Docker's isolated kernel architecture. This optimization translates to improved effici...