TLA+ for maths-based modeling, Software2, Fine-tuning with H2O LLM Studio & Model Optimization with TensorFlow

AWS Trainium for Hugging Face Transformers [Open in app]( or [online]() [TLA+ for maths-based modeling, Software², Fine-tuning with H2O LLM Studio & Model Optimization with TensorFlow]( AWS Trainium for Hugging Face Transformers Apr 28 [Share]( 👋 Hey, "The true power of large language models lies not in their size but in their ability to be fine-tuned on a wide range of tasks." - [Jeremy Howard, Data Scientist & Founding Researcher at fast.ai]( Want to take your language processing to the next level? Customizing large language models can unleash their true potential and revolutionize language processing. Tweaking these models for specific needs can achieve remarkable results and discover amazing possibilities. Let's work together to fine-tune and reveal their full power! Get ready to dive deep into [DataPro#41]( where you'll find an array of interesting and compelling content. We'll be delving into the world of [fine-tuning an LLM model using H2O LLM Studio to generate Cypher statements]( and [visual blocks for machine learning](. Plus, we'll also be discussing the exciting topic of [applying large language models to tabular data to identify drift](. And that's not all! We've got a [special serialized module]( lined up for you, where we'll be discussing the insights of [Jason Tamara Widjaja, the director of AI at MSD]( on analytics maturity models. It's time to fasten your seatbelts and prepare for a fun-filled learning experience! If you’re interested in sharing ideas to foster the growth of the data community, then this survey is for you. Consider sharing your thoughts and get a FREE bestselling Packt book, The Applied Artificial Intelligence Workshop as PDF. Jump on in! [TELL US WHAT YOU THINK]( Cheers, Merlyn Shelley Editor in Chief, Packt This Week’s Key Highlights: - [TLA+ Foundation brings math-based software modeling to the mainstream]( - [Setting up AWS Trainium for Hugging Face Transformers]( - [Software²: A new generation of AI that produces their own training data]( - [Model Optimization with TensorFlow]( - [Linear Regression to GPT in Seven Steps]( Latest Forks on GitHub - [facebookresearch]( DINOv2 models generate robust visual features that can be used in computer vision tasks without fine-tuning. - [Anything-of-anything]( Project combines [Segment Anything]( with 3D models for a demo; ongoing improvements and more demos planned. - [Stability-AI]( This repository contains [Stability AI]( ongoing development of the StableLM series of language models with continuous updates of new checkpoints. - [Vision-CAIR]( This includes a pretrained MiniGPT-4 aligned with [Vicuna-7B]( The demo GPU memory consumption now can be as low as 12GB. - [lupantech]( Chameleon is a plug-and-play [compositional reasoning framework]( that augments LLMs with various types of tools. - [OpenBMB]( BMTools extends language models via an open-source repository that eases building, sharing plugins, and using ChatGPT-Plugins externally. - [sanchit-gandhi]( Optimised JAX code for OpenAI's [Whisper Model]( largely built on the Hugging Face Transformers Whisper implementation. - [FreedomIntelligence]( LLM Zoo is a project that provides data, models, and evaluation benchmark for large language models. - [allenai]( MultimodalC4 is a multimodal extension of c4 that interleaves millions of images with text. - [suno-ai]( "Bark" by [Suno]( is a transformer-based text-to-audio model, that generates multilingual speech, music, background noise, and sound effects. [Pledge your support]( Industry Insights AWS ML - [Amazon SageMaker Data Wrangler for dimensionality reduction:]( Amazon SageMaker Data Wrangler is a tool for [ML data preparation]( and feature engineering. It offers over 300 preconfigured data transformations and allows users to write custom transformation in PySpark, SQL, and pandas. In addition, it supports dimensionality reduction techniques to help reduce data size and improve training times. - [Setting up AWS Trainium for Hugging Face Transformers:]( Hugging Face and AWS are [partnering]( to make AI more accessible by developing tools that simplify the use of [AWS Trainium]( and [AWS Inferentia]( for training and deploying Transformer and Diffusion models on AWS. AWS Trainium is a purpose-built machine learning chip for deep learning training, and a hands-on post shows how to set it up with the [Hugging Face Neuron Deep Learning AMI]( for text classification fine-tuning. Microsoft Research - [TLA+ Foundation aims to bring math-based software modeling to the mainstream:]( TLA+ is a math-based language for modeling computer programs and systems, especially concurrent and distributed ones. It has tools to help eliminate fundamental design errors that can be hard to find and costly to fix. [TLA+]( has been adopted by semiconductor makers, tech companies, and mainstream applications like payment systems. - [Unifying learning from preferences and demonstration via a ranking game for imitation learning:]( Reinforcement learning uses signals to teach robots how to perform tasks through exploration. Imitation learning reduces trial and error by providing demonstrations for the robot to learn from. Future work can automate the learning of reward functions. While effective policies were learned, reusable robust reward functions were not, and this remains a challenge for the field. Google Research - [Responsible AI at Google Research: Technology, AI, Society and Culture:]( Google considers [AI as a foundational and transformational technology](. Recent advances in generative AI technologies are being incorporated into their [products](. They acknowledge the responsibility to be culturally inclusive, equitable, accessible, and reflect the needs of impacted communities. They will take on these challenges through inter- and multidisciplinary research that centers on impacted communities to discover new ways to develop and evaluate AI technologies. - [Visual Blocks for ML: Accelerating machine learning prototyping with interactive tools:]( Deep learning has enabled real-time multimedia applications based on ML, such as [human body segmentation]( estimation]( [hand and body tracking]( and [audio processing](. Visual Blocks is a framework that lowers development barriers for ML-based multimedia applications, empowering users to experiment without coding and facilitating collaboration. In the future, it will be open for community contributions and expected to be a common interface across ML tooling. Just for Laughs! Why was the LLM model nervous about being fine-tuned? Because it didn't want to get lost in translation! Streamline your Data Strategy with Experts [Creators of Intelligence]( – By [Dr. Alex Antic]( We appreciate all the feedback you have shared about the book so far! As promised, we're excited to provide you with a sneak peek of insights from [Jason Tamara Widjaja, the director of AI at MSD]( a research-intensive biopharmaceutical company. Alex Antic: Why shouldn’t organizations entering this for the first time, in particular, get caught up in the often-quoted maturity models, such as Gartner? I know you’ve written a piece recently about this, so I thought we could talk a bit about that. I think that with some of the insights you provided, you hit the nail on the head. Jason Tamara Widjaja: When we think about maturity models, a couple of them come to mind, and for one of them, the narrative goes like this. You get all your data in shape, then you get all the insights from it, and then you get all the value from the insights. That's one very simplified version of the narrative. But it's also almost impossible for a business. If you start with, “Let's get all our data in place,” that's an impossible task because the amount of data increases exponentially. [Find out more about the book here!]( Also check out, [How analytics maturity models are stunting data science teams,]( by Jason Tamara Widjaja on Towards Data Science. This exclusively curated content is extracted from the upcoming book “[Creators of Intelligence]( by [Dr. Alex Antic.]( [EXPLORE NEW IDEAS & READ ON!]( Find Out What’s New? - [Software²: A new generation of AIs that become increasingly general by producing their own training data:]( The emergence of Software², a data-centric paradigm for developing self-improving programs based on modern deep learning, holds great promise. This article highlights the need to carefully consider the data we feed learning algorithms and design self-instructing systems that can generate such data. The future of software systems is likely to be influenced by Software², and further [exploration of its potential is recommended](. - [Linear Regression to GPT in Seven Steps:]( The article explains that generative AI is a form of prediction, using the example of Large Language Models (LLM). It breaks down LLMs into seven simple steps, from linear regression to next word prediction, and introduces important terms such as prompt, embeddings, attention, and transformers. These resources are aimed at individuals interested in machine learning who wish to gain a fundamental understanding of LLMs. - [Democratizing Machine Learning with AWS SageMaker AutoML:]( AutoML is gaining popularity as it allows businesses to leverage ML and AI without requiring expertise in data science. This article focuses on AWS SageMaker AutoML, a popular AutoML tool, and how it can be used to solve complex ML use cases. Here they also compare the results of training a credit card fraud detection model using the manual approach and AWS SageMaker AutoML. You can find the dataset [here](. - [Applying Large Language Models to Tabular Data to Identify Drift:]( LLM-based models can serve as an additional predictive tool for data science with little effort involved, scoring in the mid-eighties' percentile of several competition entries. Drift and anomaly detection are common challenges in data science and machine learning workflows. This article showcases the use of pre-trained LLMs to identify drift and anomalies in tabular data, detecting anomalous regions with as few as 2% of the data centered within five centiles from the median of the variables’ values. - [Rust: The Next Big Thing in Data Science:]( Rust's performance and security features make it a practical choice for data science, especially when handling large datasets. While lacking some features of Python, Rust has libraries designed for data analysis. The article explores Rust's tools and their application in analyzing the [iris dataset]( showcasing Rust's potential in data science beyond conventional means. - [Fine-tuning an LLM model with H2O LLM Studio to generate Cypher statements:]( Open-source LLM models have a basic understanding of Cypher syntax, but none of them could reliably generate Cypher statements based on examples or graph schema. To address this, the author fine-tuned an open-source LLM model using H2O's newly released LLM Studio tool, which provides a graphical interface for fine-tuning without requiring extensive coding or commands. - [Model Optimization with TensorFlow:]( Machine learning models have been getting bigger, but they are also being deployed onto smaller devices. Model optimization is the process of compressing and reducing latency, [allowing models to be deployed]( on smaller devices. Optimizing models can improve efficiency, reduce memory usage, and speed up inference times. Popular optimization methods include quantization and pruning, which can be implemented in TensorFlow. See you next time! As a GDPR-compliant company, we want you to know why you’re getting this email. The _datapro team, as a part of Packt Publishing, believes that you have a legitimate interest in our newsletter and its products. Our research shows that you opted-in for email communication with Packt Publishing in the past and we think your previous interest warrants our appropriate communication. If you do not feel that you should have received this or are no longer interested in _datapro, you can opt out of our emails by clicking the link below. [Like]( [Comment]( [Restack]( © 2023 Copyright © 2022 Packt Publishing, All rights reserved. Our mailing address is:, Packt Publishing Livery Place, 35 Livery Street, Birmingham, West Midlands B3 2PB United Kingdom [Unsubscribe]() [Start writing]()