DevCoach 171: Machine Learning in Google Cloud | Vertex AI di Google Cloud

Transcript

1. Vertex AI On Google Cloud M Indra Bangsawan Google Cloud Machine Learning
2. Machine Learning Responsible AI at the foundation A rich history of open-source and proprietary innovation across search, hardware, data, algorithms, and models
3. Machine Learning Evolution of AI Capabilities & Tools Predictive AI Regression & Classification Forecasting Sentiment Analysis Entity Extraction Object Detection Generative AI Text, Image & Code Generation Text & Code Rewriting & Formatting Summarization Extractive Q&A Image & Video Descriptions Multimodal Generative AI Natural Image Understanding Video Question Answering Automatic Speech Recognition & Translation Spatial Reasoning and Logic Mathematical Reasoning in Visual Contexts Train Serve MLOps Prompt Tune RAG
4. Machine Learning A unified platform from data to deployment and for all your predictive, generative, and agentic needs Databases AI & ML Data analytics Insights AlloyDB Looker BigQuery Vertex AI Agent Builder Model Builder Model Garden Governance Dataplex MLOps in Vertex AI Infrastructure AI Hypercomputer (GPUs and TPUs) Cloud Ops Gemini for Google Workspace Gemini for Google Cloud AI Agents
5. Machine Learning A machine learning ML) platform that lets you train and deploy ML models and AI applications, and customize large language models LLMs for use in your AI-powered applications. Vertex AI
6. Machine Learning
7. Machine Learning Gemini 1.5 Pro & 1.5 Flash Multimodal reasoning across a wide range of tasks Open Models on Vertex AI Mixtral 8x7B, Image Bind, DITO, Llama 3 and more Hugging Face & Kaggle Models Deployable on Vertex AI Gemma 2B & 7B Family of lightweight, state-of-the-art open models Imagen 2.0 & 3.0 Create and edit images from simple prompts Model Garden 130+ Curated Models Including… Model Builder Agent Builder Claude 3 & 3.5 on Vertex AI Claude 3 Haiku, Sonnet, Opus GA Claude 3.5 Sonnet GA
8. Machine Learning • Choice and flexibility with Google, open source, and third-party foundation models • Multiple modalities to match your use case • Multiple model sizes to match cost and efficacy needs • Domain-specific models for specialized industries • Enterprise ready with safety, security, and responsibility • Decrease time to value with fully integrated platform Google Task Specific Models Google Domain Specific Models Partner & Open Ecosystem Speech-to-Text Text-to-Speech Natural Language Translation Vision Video Intelligence Doc AI OCR Occupancy analytics MedLM Life Science and Healthcare Sec-PaLM Cybersecurity Llama 3.1 Claude 3 and 3.5 Haiku, Sonnet, and Opus Vertex AI Model Garden Gemini Foundation Models 1.0 Pro 1.5 Flash 1.5 Pro Mistral Large 2, Nemo and Codestral Hugging Face Jamba 1.5 Large and Mini Google Foundation Models Imagen 3 PaLM 2 Codey Chirp Embeddings
9. Machine Learning Gemini marks the next phase on our journey to making AI more helpful for everyone State-of-the-art, multimodal capabilities Highly optimized while preserving choice Built with responsibility and safety at the core
10. Machine Learning Highly optimized while preserving choice Gemini efficiently runs on everything from data centers to mobile devices More reliable, scalable & efficient Leveraging Google’s AI-optimized TPUs, Gemini is designed to be substantially more reliable to train, efficient to serve, and faster to run than previous models. Google’s most flexible model yet Gemini’s state-of-the-art capabilities significantly enhance the way developers and enterprise customers build and scale with AI. Gemini comes in different sizes: • Gemini 1.0 Ultra: Most capable and largest model for highly complex tasks. • Gemini 1.5 Pro: Best model for handling tasks with longer prompts, with a context window of up to 1 million tokens. • Gemini 1.0 Pro: Best model for scaling across a wide range of tasks. • Gemini Nano: Most efficient model built for on-device tasks (available via AICore, a new system capability available in Android 14, starting on Pixel 8 Pro devices).
11. Machine Learning Try Gemini-powered Multimodal experiences on Vertex AI
12. Machine Learning How different capabilities work together to generate content
13. Machine Learning Model Builder Prompt | Serve | Tune | Distill | Eval | Notebooks l Training | Feature Store | Pipelines | Monitoring Agent Builder Model Garden
14. Machine Learning Best of Predictive and Generative AI in one Platform Open Models Task Specific Models Foundation Models Discover Data & Algorithms Prototype Colab & Workbench Notebooks AI Studio Customize Training on GPUs & TPUs Tuning: SFT, RLHF, & Distillation Develop Deploy Orchestrate Pipelines & Metadata Govern FeatureStore & Model Registry Predict Endpoints on GPUs & TPUs Monitoring & Safety Evaluate Experiments & Tensboard AutoMetrics & AutoSxS Augment Grounding & Extensions Embeddings & Vector Search Developer SDKs
15. Machine Learning Training, tuning, and augmentation to customize your data-driven use cases Training Build your own predictive and generative models from scratch with your own proprietary data Tuning Customize foundation models for your specific use cases Augmentation Connect and take action on your data and applications Colab Enterprise Notebooks Vertex AI Training Vertex AI Experiments Vertex AI Prediction Prompt Design Supervised Tuning Reinforcement Learning with Human Feedback Distilling Step-by-Step Model Evaluation Function Calling Extensions Connectors Grounding Customize your models from Google, Partners, and Open Ecosystem
16. Machine Learning Customize & Augment Foundation Models on Vertex AI
17. Machine Learning Model Labeled Prompts Adapter Prompt & Inference Endpoint Generated output Customize Models on Vertex AI 2.Supervised Tuning Adapter User Preference 3.RLHF Tuning Distilled Model 4.Distilling Step-by-Step 1.Out-of-the-box performance
18. Machine Learning Augment Models on Vertex AI Model Prompt & Inference Endpoint Function calling “Where can I watch Hunger Games in Mountain View?ˮ Here are the right key-value pairs for the function find_theatres: { movie: ‘The Hunger Games: The Ballad of Songbirds & Snakes’, location: ‘Mountain, View CA’ } find_theatres(movie, location) Movie & Location DB & API FAQs Instruction Manuals Employee Handbooks “Explain my 401K benefits and contribution limit.” Grounding For 2023, you can contribute up to $22,500 as an individual to your 401(K) plan. Company will match up to half of whatever you contribute to your 401(K) from your paycheck. Grounded on your data
19. ML Ops Learning Path]
20. Machine Learning MLOps A set of standardized processes and technology capabilities for building, deploying, and operationalizing ML systems rapidly and reliably
21. Machine Learning What is it? How is it related to data science? Model deployment CI/CD MLOps
22. Machine Learning Model deployment CI/CD MLOps
23. Machine Learning MLOps maturity levels ML pipeline automation Manual process CI/CD pipeline automation . 0 1 2
24. Machine Learning Manual Process Manual experiment steps Data preparation Model training Model evaluation & validation Model deployment Data extraction & analysis Model API experimentation/dev/test staging/preprod/prod ML Ops Offline data Trained models registry Trained model
25. Machine Learning ML Pipeline Automation Data prep. Model training Model eval. Pipeline deployment Data analysis Model API ML Ops Orchestrated experiment Data validation Model valid. Data prep. Model training Model eval. Data validation Model deploy. Model valid. Automated pipeline Live data Trigger Offline data Model Analysis Performance monitoring Trained models registry Source repo Src code Data extraction Trained model experimentation dev/test/staging/prod Feature store
26. Machine Learning CI/CD Pipeline Automation Data analysis Model API experimentation dev/test/staging/prod MLOps Orchestrated experiment Data prep. Model training Model eval. Data validation Model deployment Model valid. Automated pipeline Live data Trigger Offline data Model analysis Performance monitoring Trained models registry Source repo cod e Src code Data extraction CD: Deploy pipeline CI: Build, test, & package pipeline components 1 Package s Package s Packages 2 3 4 5 6 Feature store Metadata DB Trained model
27. Machine learning Development Operations • Data processing • Model development • Metric evaluation • CI/CD building • Model integration • Latency calculations • Continuous monitoring • User feedback loop • Model delivery Summarizing MLOps - One quick explanation Machine Learning
28. MLOps on Vertex AI Trained model Model service Model Registry Vertex Models CT pipeline Vertex Pipelines (Vertex Training) Pipeline components Container Registry ML Data assets Vertex Datasets & Feature Store Code Repository Cloud Source Repository Experimentation and development Notebooks Serving logs Code & config changes Pipeline artifacts CI/CD for Training Pipeline Cloud Build Prediction serving Vertex Prediction Monitoring Vertex Model Monitoring training datasets CI/CD for Model Serving Cloud Build ML metadata Vertex Metadata Logged experiment Vertex Experiments Visualizations Vertex TensorBoard Resource sizing GCE machine types Accelerators Cloud TPUs & GPUs serving features Alert Trigger
29. Machine Learning Workbench and Colab Enterprise on Vertex AI Custom model development via notebooks Collaboration & Productivity IAM based notebook sharing Generative AI powered code completion and generation Automatic Versioning Zero-Config & Flexible Compute Provides both zero-config compute options, as well as access to a wide range of machine-types and compute Enterprise Ready Will support a wide range of security and management capabilities including: • VPCSC • CMEK • Regionalization • Cloud Logging & Monitoring Available across Google Cloud Available in BigQuery and Vertex AI Dataproc and Dataflow coming soon), making it easy to work across data and AI workloads
30. Upskilling in AI/ML Machine Learning
31. ML Engineer, Data Scientist ML and Artificial Intelligence Overview A Machine Learning Engineer designs, builds, productionizes, optimizes, operates, and maintains ML systems. What will you learn • TensorFlow, AI Platform Notebooks, Cloud Dataflow, Cloud DataFusion, AI Platform; • BigQuery, BigQuery ML, and more. CO SB DS CO SB CO SB CO SB SB CO SB DS CO SB DS OR Google Cloud Big Data and ML Fundamentals | ⧗ 1 DAY Machine learning on Google Cloud | ⧗ 5 DAYS How Google Does Machine Learning Launching into Machine Learning TensorFlow on Google Cloud Feature Engineering Machine Learning in the Enterprise Advanced Solutions Lab (ASL) | ⧗ 5 WEEKS Advanced Machine Learning with TensorFlow on Google Cloud | ⧗ 5 WEEKS Production Machine Learning Systems Computer Vision Fundamentals with Google Cloud Natural Language Processing on Google Cloud Recommendation Systems on Google Cloud ML Ops Fundamentals | ⧗ 3 WEEKS ML Pipelines on Google Cloud | ⧗ 4 DAYS Perform Foundational Data, ML, AI Tasks in Google Cloud Build and Deploy ML Solutions on Vertex AI Professional Machine Learning Engineer Exam Guide and Sample Questions Professional Machine Learning Engineer SB SB CO SB DS CO SB CO SB CO SB CO SB SB CO SB DS CO  Resource  On-demand  Virtual and in-person Classroom  Self-paced Labs  Skill badge CO SB DS  Data Sheet  Coursera  Skills Boost
32. Generative AI (GenAI) ML and Artificial Intelligence Overview This learning path guides you through a curated collection of content on Generative AI products and technologies, from the fundamentals of Large Language Models to how to create and deploy generative AI solutions on Google Cloud. LINKS ↓  Resource  On-demand  Virtual and in-person Classroom  Self-paced Labs  Skill badge CO SB DS  Data Sheet  Coursera  Skills Boost Introduction to Generative AI | ⧗ 1 DAY Introduction to Large Language Models | ⧗ 1 DAY Introduction to Responsible AI | ⧗ 1 DAY Generative AI Fundamentals Introduction to Image Generation | ⧗ 1 DAY Encoder-Decoder Architecture | ⧗ 1 DAY Attention Mechanism | ⧗ 1 DAY Transformer Models and BERT Model | ⧗ 1 DAY Create Image Captioning Models | ⧗ 1 DAY SB SB SB SB SB SB SB SB SB
33. Thank You indra@bangkit.academy Indra Bangsawan Machine Learning