[Feb 2024] ML Community — Highlights and Achievements

[Feb 2024] ML Community — Highlights and Achievements

Let’s explore highlights and accomplishments of the vast Google Machine Learning communities over the month. We appreciate all the activities and commitment by the community members. Without further ado, here are the key highlights!

Featured Stories

https://medium.com/media/34cc66c6c3734fcb3dc4070bebafccf7/href

Gemini 1.5 Pro for Code — Part 01 by AI/ML GDE Sam Witteveen (Singapore) got 56k+ views! It introduces how to use Gemini 1.5 Pro for code-related tasks, showing its ability to generate code, handle scenarios, and understand code bases.

The third round of the ML Olympiad has been launched! Visit the ML Olympiad page to explore 20+ competitions addressing real-world challenges. 14 teams joined AI Assistants for Data Tasks with Gemma competition and 8 teams joined LLM Prompt Recovery (with Gemma).

Focus Area Highlights

Gemma

Build end-to-end RAG with Gemma and GenAI Stack Studio by AI/ML GDE Tarun R Jain (India) guides how to build an end-to-end RAG application using Gemma without writing a single line of code.

https://medium.com/media/89a4fccce3e6bb2691c8b5f374e9e15f/href

Natural Language to SQL with Google Gemma : A Comprehensive Guide by AI/ML GDE Bhavesh Bhatt (India) shows how to fine-tune Gemma for converting your natural language question to SQL queries.

Cost-Efficient Multi-Agent Collaboration with LangGraph + Gemma for Code Generation

Cost-Efficient Multi-Agent Collaboration with LangGraph + Gemma for Code Generation by AI/ML GDE Rubens Zimbres (Brazil) shows how he made Gemma collaborate with OpenAI’s gpt-3.5-turbo-1106 to generate a graph plot from a simple natural language sentence. This article got more than 290 claps in Google Cloud — Community medium.

Introducing Gemma — 2B 7B 6Trillion Tokens by AI/ML GDE Sam Witteveen (Singapore) is an introduction video with more than 49K views. He introduced the different sizes of Gemma and how you can start using these and train them out.

Gemini

Analyzing Financial Reports with Gemini 1.5 by AI/ML GDE Ertuğrul Demir (Türkiye) evaluates the new context length of Gemini 1.5 using recent data to determine how it will perform against RAG systems in the future. He used Nvidia’s latest financial report as an example.

LangChain Persistent Memory Chatbots with Gemini Pro and Firebase

Building Persistent Memory Chatbots with LangChain, Gemini Pro, and Firebase by AI/ML GDE Daniel Gwerzman (UK) demonstrates the integration of Gemini Pro, LangChain, and Firebase to build chatbots with persistent memory.

Gemini Reshaping the NLP Task for Extracting Knowledge in Text

Gemini Reshaping the NLP Task for Extracting Knowledge in Text by AI/ML GDE Joan Santoso (Indonesia) is a tutorial on how to utilize Gemini to analyze natural language input and produce entities and their relation.

Gemini and LlamaIndex Tutorial by AI/ML GDE Tarun R Jain (India) is a video showing how to implement a multimodal application using Gemini Pro Vision and LlamaIndex.

Gen AI / LLM

Self-Improving LLMs (slides) by AI/ML GDE Martin Andrews (Singapore) explored papers covering self-improvement of LLMs such as Self-Rewarding, AlphaCodium, and Self-Discover.

GenAI video transcription and chat

GenAI Video Knowledge App by AI/ML GDE David Cardozo (Canada) is an AI video knowledge application capable of processing numerous videos to create your own customized knowledge library. This project became a part of the official Docker documentation.

Keras

Neural Network 101 for Dummies () by AI/ML GDE Suvaditya Mukherjee (India) was to help beginners to get started with neural networks — what they are, how they can be applied and used in different domains, and a hands-on short demo on how to get started building your own neural networks using Keras and JAX.

https://medium.com/media/ef5ef95d85704937c53769498d5d4e2f/href

Keras (3) for the Curious and Creative (at PyData) by AI/ML GDE Marvin Ngesa (Kenya) was for people who would like to learn the capabilities of Keras for CV and stable diffusion. He guided how to use Keras CV and Keras 3 together for multi-framework modeling.

Adding support for Object Detection Layers in Keras CV with all the backends by AI/ML GDE Aritra Roy Gosthipaty (India) was a PR adding support to all the object detection layer in KerasCV to work with any backend.

Kaggle

Fine-tuning a large language model on Kaggle Notebooks (or even on your own computer) for solving real-world task by AI/ML GDE Luca Massaron (Italy) is a tour of the step for fine-tuning LLMs in Kaggle notebooks.

Fine Tuning Gemma-7B using Kaggle Models by AI/ML GDE Tarun R Jain discussed how to start Kaggle Models and shared what fine-tuning is and how you can easily fine-tune Gemma on custom data using parameter efficient fine-tuning methods.

Usha Rengaraju (India) shared training and inference notebooks for the competition, The Learning Agency Lab — PII Data Detection. She used Roformer (paper) to solve the problem.

JAX

fit-a-nef by AI/ML GDE Phillip Lippe (Netherlands) is a library for efficient training of neural fields at scale. It is designed to allow users to add their own training task, dataset, and model. Regardless of the frameworks you use, neural fields can be loaded and used in a project.

nerf-keras-jax by AI/ML GDEs, Aritra Roy Gosthipaty (India), Ritwik Raha (India) and Saurav Maheshkar (India), is an implementation of the NeRF (paper) tutorial with Keras and JAX.

On-device ML

Use a tiny Llama model inside android by AI/ML GDE George Soloupis explores the integration of a lightweight Llama model, developed in C language, within an Android application.

Run Gemma directly on your browser client side by by AI/ML GDE Rabimba Karanjai (US) shows how he enabled Gemma to run directly on a device offline in browser client side.

From Data to decision : Exploring The power of cloud vision AI by TFUG Prayagraj aimed to provide participants with a comprehensive understanding of ML fundamentals and hands-on experience with TensorFlow and cloud vision API on Google Cloud. AI/ML GDE Nitin Tiwari (India) participated as a speaker and to share his knowledge.

Unraveling the Magic of Neural Machine Translation with Transformers Pipeline by TFUG Salem covered the evolution of translation technology, from traditional approaches to cutting-edge methodologies. They explored the nuances of designing, training, and fine-tuning NMT models within a Pipeline framework, empowering attendees to leverage these techniques in their own projects.

[Feb 2024] ML Community — Highlights and Achievements was originally published in Google Developer Experts on Medium, where people are continuing the conversation by highlighting and responding to this story.