This workshop is designed to help you operationalize an open source Large Language Model (LLM), and use the LLM to build a generative AI Q&A bot. Basically, this workshop is broken down into the following parts:
-
Register a base LLM model using SageMaker Model Registry
-
Finetune a Llama2 model with custom dataset using SageMaker Processing, Training and Inference using SageMaker Hosting.
-
Create an automated LLM training pipeline that orchestrates the LLM finetuning jobs using SageMaker Pipelines. Triggers an LLM deployment to production through CICD pipeline supported via CodePipeline.
-
Model Monitoring for LLM.
-
Deploy an Embedding Model through SageMaker Jumpstart. The embedding model will be used for creating embedding for the content to be stored in a vector database.
-
Build a vector database using Amazon OpenSearch serverless. Ingest content (news articles) into the vector database using langchain library. These content will be used in a retrieval Q&A bot to provide accurate answer based on user queries.
-
Build a fully functional Q&A chatbot using open source components. In particular, 1/ Amazon OpenSearch Serverless as a vector database for knowledge base repository. 2/ Open Sorce LLM (Llama2) to handle user queries, answers using natural language. 3/ Langchain framework for orchestrating the chat application. 4/ Streamlit framework to build a fully working chat user interface.
The following diagram depicts the building blocks used in the workshop:
This repository contains the materials and source codes required to complete the workshop. For instruction on how to get started, please refer to this link: