![[may]-ml-community — highlights-and-achievements](https://prodsens.live/wp-content/uploads/2024/06/24541-may-ml-community-highlights-and-achievements-550x281.png)
[May] 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
ML Developer’s Journey
AI/ML GDE Bhavesh Bhatt (India) became one of the top 10 selected to compete in the Literature category of the Global Prompt Engineering Championship in Dubai — a big competition with over 60,000 applicants. He used Gemini Pro 1.5 to generate 250–300 word passages for the challenge and found it provided well-structured paragraphs quickly. He commented that Gemini’s ability to create coherent and engaging content played a significant role in his success.
https://medium.com/media/879f79722e35543a1a4878f5b6466d19/href
Advanced Colab — How to go Beyond the Basics by AI/ML GDE Sam Witteveen (Singapore) summarizes tips and tricks for people who use Colab regularly. It introduces Code snippets that can import models from sources like Hugging Face; Colab AI where you can ask about your code or let it generate some code instead of you; and how to use custom VMs and so on. This is a follow-up video to the previous one Colab 101: Your Ultimate Beginner’s Guide!, which explained almost everything about Colab from its history to useful features.
Gemini
Recipe Recommender Gemini in Android by AI/ML GDE Juan Guillermo Gomez Torres (Mexico) used Gemini API on Android to recommend recipes with multiple parameters.
Unlock The Gemini 1.5 Pro API (+ File API ) by AI/ML GDE Sam Witteveen (Singapore) shows you how to use Gemini 1.5 Pro via API calls and how to upload files for use in the 1.5 model. His other video Google’s RAG Experiment — NotebookLM shares an overview of NotebookLM and his insight about how Google uses RAG and builds a product around the technique.
Building an AI-Powered chrome extension with Gemini API (repository | Chrome Web Store) by Taha Bouhsine (Morocco) was a talk explaining how he used Gemini API to build a Chrome extension that analyzes the pages you saved in your bookmark.
Using System prompts with Gemini by AI/ML GDE Linda Lawton (US) introduces the system prompts in Gemini API enabling developers to direct the behavior of the model based on their needs.
Function Calling with RAG cache using Gemini by AI/ML GDE Anubhav Singh (India) is a repository for workshops about function calling and RAG based response generation using Gemini. He also participated in TechXcellence — Build with AI Hackathon by GDG Gwalior, the 36-hour hackathon, as a mentor and helped teams through using function calling and RAGs with Gemini on Vertex AI.
Gemini 101: Everything you need to know to build your Gemini apps in scale by AI/ML GDE Rabimba Karanjai (US) focused on building and scaling applications using Gemini. He covered advanced prompt engineering skills leveraging Gemini’s full potential, how to seamlessly integrate Gemini with audio processing tools and external services and so on.
Chat with PDF by AI/ML GDE Suvaditya Mukherjee (India) was a quick solution demo using LangChain, MongoDB, Vertex AI and Gemini Pro to get chat capabilities against a PDF, all in a single Colab Notebook.
Starting to prototype with Gemini and Gema by AI/ML GDE Lesly Zerna (Bolivia) shared her project to help tourists going from Peru to Bolivia who need information about local activities, best routes and local companies. To prototype this idea, she tested the multimodal option for inputs in AI Studio, practiced Gemini’s translation capabilities and prompting. With a demo of the prototype, she explained how AI Studio and the Gemini API worked.
Unleash Generative AI Power in our Hands-on Workshop! by AI/ML GDE Henry Ruiz (US) was a workshop introducing the fundamentals of Gen AI with hands-on demos on Gemini and LangChain APIs. He also gave a speech about Multimodality with Gemini at GDG Bamenda (Cameroon) introducing Gemini and its capabilities as a multimodality model.
Build Custom Retriever using LLamaIndex and Gemini by AI/ML GDE Tarun R Jain (India) described how he implemented a custom retriever that performs hybrid search by combining Vector and Keyword retrievers using LlamaIndex, with the support of Gemini and embeddings. This approach reduced hallucinations to some extent in a typical RAG pipeline.
Gemini : Delving into NLP, LLMs, Function Calling, and RAGs by TFUG Durg introduced students to the latest advancements in GenAI, Gemini, Gemma, and GCP. Three AI/ML GDEs Anubhav Singh, Rishiraj Acharya and Kartikey Rawat participated in the event to share their knowledge.
AI Buildathon with Google Gemini by Deep Tech Stars was a day-long hackathon where developers came together to build Gen AI applications and demo with Gemini.
Gemma
Understanding PaliGemma in 50 minutes or less by AI/ML GDE Ritwik Raha (India) explores the intricacies of PaliGemma, the latest vision-language model. He covers steps to get PaliGemma up and running, from a simple inference to fine-tuning in details.
Fine-Tuning-Gemma-2b-it-for-Arabic by AI/ML GDE Ruqiya Bin Safi (Saudi Arabia) aims to improve the performance of Gemma-2b-it, specifically for Arabic text. The model has been fine-tuned using the arbml/CIDAR Arabic dataset, a curated collection of Arabic text data. She enhanced the model’s ability to process and generate Arabic text, making it more useful for a wide range of applications in Arabic-speaking contexts.
https://medium.com/media/b4b29d94ff6bfbb9b86f15c49f37f266/href
Mastering Google’s VLM PaliGemma: Tips And Tricks For Success and Fine Tuning by AI/ML GDE Sam Witteveen (Singapore) introduces PaliGemma, the newly launched vision-language model inspired by PaLi-3 and how to fine-tune it.
Demystifying the fine tuning of LLMs in practice: PEFT, LoRA, QLORA and Hamburgers by AI/ML GDE Hugo Zanini (Brazil) is a hands-on article receiving 116+ Claps in Medium. It explains the key concepts of LLMs tuning using Gemma. He used Colab and a dataset of tweets extracted from the Burger King’s X account.
GCCD Extended Kolkata x Flutter Kolkata — QLoRA Finetuning & DPO Aligning Google’s Gemma with Hugging Face by AI/ML GDE Rishiraj Acharya (India) explored the process of fine-tuning Gemma using QLoRA and aligning it with Hugging Face’s platform. He discussed how this method can be applied to Gemma, to tailor it for specific tasks while minimizing the computational resources required.
#Keras Build With Ai in GDG Hangzhou and Build With AI in GDG Wenzhou by AI/ML GDE Yu Chen (China) were Gemma hands-on workshops covering how to use Gemma with KerasNLP & JAX/Flax and how to fine-tune it.
TFUG Islamabad hosted a series of ML Study Jams. Several AI/ML GDEs participated in the event to share their knowledge and experiences with Google ML products. Understanding PaliGemma Google’s Vision Language Model by Ritwik Raha (India); Next-Level Tech: ML, TensorFlow, Computer Vision, & Gen AI by Nitin Tiwari (India), Unlocking AI: Introduction to Gemini Workshop by Karthik Muthuswamy (India) and so on.
Colab
Step-by-Step Tutorial to Build a RAG using Google Gemma and MongoDB Atlas by AI/ML GDE Ashmi Banerjee (India) is a Colab tutorial using a simple RAG with Gemma to help it answer travel-related queries.
Pure Python RAG with In-Memory Vector DB by AI/ML GDE Anubhav Singh (India) is a repository built on Colab and contains a pure Python implementation of a RAG system with an in-memory vector database. The project showcases a basic yet effective approach to store, query, and retrieve information based on vector similarity. As an educational tool, it demonstrates the foundational aspects of information retrieval, NLP and their application in building intelligent systems. The repo also includes an example of using the in-memory DB with Gemini to generate responses.
Clustering with unsupervised models: Kmeans & collective clustering & Practical workshop on unsupervised models (Insurance data) by AI/ML GDE Adonai Vera (India) are notebooks for workshops on unsupervised models for university students.
Kaggle
Retrieval Augmented Generation (RAG) using Gemma to Explain Basic Data Science Concepts (Notebook) by AI/ML GDE Ruqiya Bin Safi (Saudi Arabia) introduces how to build an AI assistant which is ready to explain complex Data science concepts in simple terms and to guide you through the learning process. She used Gemma and RAG to make it possible, and guides you in the writing step by step.
More Feature Ideas for GBDT’s (discussion) by AI/ML GDE Ertuğrul Demir (Türkiye) is a notebook and discussion on the Kaggle featured code competition, Automated Essay Scoring 2.0. In this work, he created a new set of features to spark some ideas for the competition using various techniques.
Data Science AI Assistant with Gemma 2b-it: a RAG 101 (his notebook) by AI/ML GDE Luca Massaron (Italy) shares knowledge and experiences he learned by participating in the Kaggle featured competition, Google — AI Assistants for Data Tasks with Gemma. This 25 min long writing goes into the details about how he leveraged the capabilities of Gemma and a RAG system.
ODML
MobileLlama3: Run Llama3 locally on mobile (repository) by AI/ML GDE Nitin Tiwari (India) is an end-to-end tutorial on how to quantize, convert, and deploy the Llama3–8B-Instruct model on mobile devices for offline inference. Additionally, it also covers the steps to set up an environment on GCP to compile the model weights for Android compatibility.
Pytorch classification model to .tflite with Google-AI-Edge library by AI/ML GDE George Soloupis (Greece) explores how to convert a PyTorch model to .tflite using the AI Edge library. Traditionally, this conversion involved using the ONNX library and several conversion steps from PyTorch to TensorFlow Lite. However, the AI Edge library significantly simplified this procedure, making the conversion process much more straightforward.
How was Gemma trained and how was she deployed? by AI/ML GDE Ngoc Ba (Vietnam) was an introduction to Gemma, how it was trained, and the techniques used to deploy the Gemma model on Android.
Walkthrough of Gemini and Gemma on Cloud and On-device by AI/ML GDE Rabimba Karanjai (US) shared all you need to build and rapidly prototype an AI application using Gemma, right on your own machine, and scale it up to the cloud within 20 minutes.
JAX
Lecture on Finetunning Gemma (slides) by AI/ML GDE David Cardozo (Canada) provided a guide to fine-tuning the Gemma language model within the GCP environment. It covers the necessary steps and considerations for adapting Gemma to specific use cases, from Cloud Build as a code repo to using advanced features of Vertex AI Pipelines. He used Flax, since the course uses JAX and an intro to generative models.
Fine Tuning Pali Gemma by AI/ML GDEs Aritra Roy Gosthipaty (India) and Ritwik Raha (India) is a bare bone PyTorch repository using JAX as a backend to fine-tune PaliGemma.
GCCD Extended Kolkata x GDSC FIEM — Intro to JAX by building a VectorDB to perform RAG with Gemini by AI/ML GDE Rishiraj Acharya (India) introduced the JAX library by building a Vector Database from scratch. He explored core JAX features and how they can be used to efficiently implement a VectorDB. He also demonstrated how to use the VectorDB to perform RAG with the Gemini.
Gemma Inference using JAX and Flax & using JAX and KerasNLP by AI/ML GDE Kartikey Rawat (India) was about JAX/Flax architecture, intro to XLA, Gemini and Gemma and inference demo of Gemma.
At TensorFlow week by TFUG Uyo, AI/ML GDE Kartikey Rawat (India) provided an in-depth understanding of the architecture of JAX and demonstrated Gemma Inference using KerasNLP.
https://medium.com/media/bc64213e86caaf7fc5d4742d81f072e0/href
#Firebase Python Firebase Functions Using Gemini — Part 1 & Part 2 by AI/ML GDE Juan Guillermo Gomez Torres (Mexico) delivered how to create Cloud functions from scratch with Python and how to use Gemini through Vertex AI to create a poem.
#VertexAI The (Hidden?) Costs of Vertex AI Resource Pools: A Cautionary Tale by AI/ML GDE Paolo Galeone (Italy) shares his experience with a sudden price increase in Vertex AI and how he handled it on the way of training a custom model.
#VertexAI Taking a walk in the Model Garden by AI/ML GDE Suvaditya Mukherjee (India) was a session about how to get started with generative AI, Vertex AI on GCP. He also explained usage of the APIs to directly interface with large models easily.
#Cloud #GKE Deploy a Generative AI application with Terraform in less than 30 minutes by AI/ML GDE Rubens Zimbres (Brazil) explored how to host data and embeddings in a SQL instance using pgvector and the generative AI application will run on GKE. It is a related writing to Deploy a Generative AI application with Terraform in less than 30 minutes he published about a month ago.
#Cloud Build with AI Qatar — Gemini Assist with GCP by AI/ML GDE Kartikey Rawat (India) introduced how to use Gemini Assist (formerly called Duet AI) and showed a live demo on GCP.
Others
https://medium.com/media/9d53008226572e3ca055eda41fe54524/href
Episode 194 — Google AI/O 2024 by AI/ML GDEs Allen Firstenberg (US) and Roya Kandalan (US) discussed the major AI announcements from Google I/O 2024. From Gemini updates and new models to responsible AI and projects like Astra, this episode dives into the future of AI development.
Artificial Intelligence and Machine Learning Malaysia reached 10K Members Milestone since the group was founded in 2017!
[May] 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.
