More Related Content
Similar to Certification Study Group - NLP & Recommendation Systems on GCP Session 5
Similar to Certification Study Group - NLP & Recommendation Systems on GCP Session 5 (20)
More from gdgsurrey
More from gdgsurrey (6)
Recently uploaded
Recently uploaded (20)
Certification Study Group - NLP & Recommendation Systems on GCP Session 5
- 1. Session 5 ChengCheng Tan Professional Machine Learning Engineer
- 2. Who am I? ChengCheng Tan - BA Linguistics & Computer Science, UCLA - MS Computer Science, Stanford - LLM + AI Safety FAR AI Communications AISafety.info + chatbot - Google Women Techmakers Ambassador
- 3. Where are we on our journey
- 4. Professional Machine Learning Certification Learning Journey Organized by Google Developer Groups Surrey co hosting with GDG Seattle Session 1 Feb 24, 2024 Virtual Session 2 Mar 2, 2024 Virtual Session 3 Mar 9, 2024 Virtual Session 4 Mar 16, 2024 Virtual Session 5 Mar 23, 2024 Virtual Session 6 Apr 6, 2024 Virtual Review the Professional ML Engineer Exam Guide Review the Professional ML Engineer Sample Questions Go through: Google Cloud Platform Big Data and Machine Learning Fundamentals Hands On Lab Practice: Perform Foundational Data, ML, and AI Tasks in Google Cloud (Skill Badge) - 7hrs Build and Deploy ML Solutions on Vertex AI (Skill Badge) - 8hrs Self study (and potential exam) Lightning talk + Kick-off & Machine Learning Basics + Q&A Lightning talk + GCP- Tensorflow & Feature Engineering + Q&A Lightning talk + Enterprise Machine Learning + Q&A Production ML Systems and Computer Vision with Google Cloud + Q&A Lightning talk + NLP & Recommendation Systems on GCP + Q&A Lightning talk + MOPs & ML Pipelines on GCP + Q&A Complete course: Introduction to AI and Machine Learning on Google Cloud Launching into Machine Learning Complete course: TensorFlow on Google Cloud Feature Engineering Complete course: Machine Learning in the Enterprise Hands On Lab Practice: Production Machine Learning Systems Computer Vision Fundamentals with Google Cloud Complete course: Natural Language Processing on Google Cloud Recommendation Systems on GCP Complete course: ML Ops - Getting Started ML Pipelines on Google Cloud Check Readiness: Professional ML Engineer Sample Questions
- 5. 1 2 4 5 3 LLM Overview Intro to Gemini API Sample Question Review Q&A Session 5 Content Review
- 6. LLM Overview
- 7. Natural Language Processing [NLP]: Computers Understand Human Languages
- 10. J.R. Firth, Linguist You shall know a word by the company it keeps J.R. Firth, Linguist
- 11. 2010s: Rise of Deep Learning and Neural Networks
- 12. Woman Man Queen King V(king) - V(queen) + V(woman) = V(man) 2013: Word2Vec Embeddings
- 13. Analogies Word Pair 1 Word Pair 2 Man-Woman king queen man woman Capital city Athens Greece Oslo Norway City-in-state Chicago Illinois Sacramento California Opposite possibly impossibly ethical unethical Comparative great greater tough tougher Nationality adjective Switzerland Swiss Canada Canadian Past tense walking walked swimming swam Plural nouns mouse mice dollar dollars 2013: Word2Vec Embeddings
- 14. 2010s: Neural Networks RNN, GRU, LSTM
- 15. Early Neural Networks - Slow & Forgetful
- 16. 2017: Transformers - Self-Attention - Algorithm+Data+Compute
- 18. Rise of LLMs >1 Billion Neurons
- 19. Trained for Next Word Prediction
- 24. RLHF: Reinforcement Learning from Human Feedback Fine-tuned - Follow Instructions - Conversations
- 25. Intro to Gemini API
- 28. Get API Key Treat like password
- 29. Create new - Freeform prompt - Structured prompt - Chat prompt
- 30. Freeform Prompt Blog post creator Write a prompt as text and image for the model to auto continue.
- 31. Structured Prompt Marketing description Table-based interface for more complex model priming and prompting
- 32. Chat Prompt Talk to snowman Simulate a back & forth conversation with a model
- 33. Get Code - Choose Language - Open in Colab - Copy to Editor
- 34. Settings Tokens - Words or subwords - Different LLM tokenizers - Training data, context window Temperature - Selected by probability - Between 0 to 1.0 - Diversity or โcreativityโ
- 36. Settings Safety Ratings Harm Categories - Harassment - Hate Speech - Sexually Explicit - Dangerous Content
- 37. Safety Ratings Harm Categories - Harassment - Hate Speech - Sexually Explicit - Dangerous Content Harm Probabilities - NEGLIGIBLE - LOW - MEDIUM - HIGH Settings
- 38. The hottest new programming language is English. Andrej Karpathy OpenAI
- 39. Prompt Engineering - Clear & Specific Instructions - Give Examples - Step by Step
- 40. REST APIs Client libraries for - Python - JavaScript - Android (Kotlin) - Swift - cURL
- 41. Setup Install & import libraries $ pip install google-generativeai import google.generativeai as genai genai.configure(api_key="<YOUR API KEY>")
- 42. Generate Text Text only prompt model = genai.GenerativeModel('gemini-pro') response = model.generate_content("Write a story about a boy and a backpack.") print(response.text)
- 43. Generate Text Text and image prompt model = genai.GenerativeModel('gemini-pro-vision') img = PIL.Image.open('image.jpg') response = model.generate_content("Write a blog based on this photo.", img) print(response.text)
- 44. Chat Conversations For interactive applications model = genai.GenerativeModel('gemini-pro') chat = model.start_chat(history=[]) response = chat.send_message("Hello, how are you?") print(response.text)
- 45. Free for now. Try it out! goo.gle/ai-dev
- 46. Session 5 Content Review
- 47. Session 5 Study Group ML Pipeline Automation & Orchestration - Design pipeline. - Implement training pipeline. - Implement serving pipeline. - Track and audit metadata. - Use CI/CD to test and deploy models. ML Solution Monitoring, Optimization, and Maintenance - Monitor ML solutions. - Troubleshoot ML solutions. - Tune performance of ML solutions for training & serving in production.
- 48. Pipelines automate the training and deployment of models Extract Data Prepare Data Train Model Validate Data Evaluate Model Validate Model Deploy Model
- 49. โ Continuous Integration (CI) is no longer only about testing and validating code and components, but also testing and validating data, data schemas, and models. โ Continuous Deployment (CD) is no longer about a single software package or a service, but a system (ML training pipeline) that should automatically deploy another service (model prediction service). โ Continuous Training is a new property, specific to ML systems, concerning automatically retraining and serving the models. From DevOps to MLOps
- 50. In addition to the actual ML... ML Code
- 51. ...you have to worry about so much more. Configuration Data Collection Data Verification Feature Extraction Process Management Tools Analysis Tools Machine Resource Management Serving Infrastructure Monitoring Source: Sculley et al.: Hidden Technical Debt in Machine Learning Systems ML Code
- 52. Data Ingestion Data Analysis + Validation Data Transformatio n Trainer Model Evaluation and Validation Serving Logging Shared Utilities for Garbage Collection, Data Access Controls Pipeline Storage Tuner Shared Configuration Framework and Job Orchestration Integrated Frontend for Job Management, Monitoring, Debugging, Data/Model/Evaluation Visualization TFX is the solution to this problem
- 53. Data Analysis + Validation Data Transformatio n Trainer Model Evaluation and Validation Serving Tuner Data Ingestion Logging Shared Utilities for Garbage Collection, Data Access Controls Pipeline Storage Shared Configuration Framework and Job Orchestration Integrated Frontend for Job Management, Monitoring, Debugging, Data/Model/Evaluation Visualization TFX is the solution to this problem Tensorflow Data Validation TensorFlow Transform Estimator or Keras Model TensorFlow Model Analysis TensorFlow Serving Keras Tuner
- 54. TFX Components Data Ingestion Logging Tensorflow Data Validation TensorFlow Transform Estimator or Keras Model TensorFlow Model Analysis Validation Outcomes TensorFlow Serving ExampleGen StatisticsGen SchemaGen Example Validator Transform Trainer Evaluator Pusher Model Server Data Ingestion Tensorflow Data Validation TensorFlow Transform Estimator or Keras Model TensorFlow Model Analysis Validation Outcomes TensorFlow Serving
- 56. Data Ingestion Data Analysis + Validation Data Transformatio n Trainer Model Evaluation and Validation Serving Logging Shared Utilities for Garbage Collection, Data Access Controls Pipeline Storage Tuner Shared Configuration Framework and Job Orchestration Integrated Frontend for Job Management, Monitoring, Debugging, Data/Model/Evaluation Visualization Estimator or Keras Model TensorFlow Model Analysis TensorFlow Serving Keras Tuner
- 59. Rolling Update v1 v2 v1 v1 v2 v2 v2 - Slowly increasing number of NEW - Slowly decreasing number of OLD
- 60. Canary Deployment v1 v2 v1 v1 v2 v2 v2 - Test on small SUBSET in production - After testing, traffic shifts to NEW
- 61. Blue-Green Deployment v1 v2 v1 v2 - 2 SEPARATE deployments - After testing, switch traffic from OLD Blue to NEW Green.
- 63. Which of the following are benefit(s) of running TFX on Google Cloud? Select all that apply? A. Vertex Pipelines is the only supported orchestrator for TFX pipelines B. Simplify scaling of TFX pipeline data processing as your data grows C. Automate your ML operational processes for individual and multiple ML pipelines. D. Increase your pipeline development and experimentation velocity.
- 64. Which of the following are benefit(s) of running TFX on Google Cloud? Select all that apply? A. Vertex Pipelines is the only supported orchestrator for TFX pipelines B. Simplify scaling of TFX pipeline data processing as your data grows C. Automate your ML operational processes for individual and multiple ML pipelines. D. Increase your pipeline development and experimentation velocity.
- 65. In addition to CI/CD practiced by DevOps teams, MLOps introduces: A. Continuous classification B. Continuous regression C. Continuous training D. All of the above.
- 66. In addition to CI/CD practiced by DevOps teams, MLOps introduces: A. Continuous classification B. Continuous regression C. Continuous training D. All of the above.
- 67. In what order are the following phases executed in a machine learning project? I - Selection of ML algorithm II - Data Exploration III - Definition of the business use case IV - Model monitoring V - Model operationalization VI - Model Development A. I, II, III, IV, V, VI B. III, II, I, VI, V, IV C. II, III, I, VI, IV, V D. II, I, III, VI, IV, V
- 68. In what order are the following phases executed in a machine learning project? I - Selection of ML algorithm II - Data Exploration III - Definition of the business use case IV - Model monitoring V - Model operationalization VI - Model Development A. I, II, III, IV, V, VI B. III, II, I, VI, V, IV C. II, III, I, VI, IV, V D. II, I, III, VI, IV, V III - Definition of the business use case II - Data Exploration I - Selection of ML algorithm VI - Model Development V - Model operationalization IV - Model monitoring Pipeline
- 69. You want to have two versions of your application in production, but be able to direct a small percentage of traffic to the newer version as a gradual test. This is an example of which deployment strategy? A. Canary deployment B. Blue-green deployment C. Rolling updates
- 70. You want to have two versions of your application in production, but be able to direct a small percentage of traffic to the newer version as a gradual test. This is an example of which deployment strategy? A. Canary deployment B. Blue-green deployment C. Rolling updates Canary Deployment Blue-Green Deployment Rolling Updates
- 71. Q&A