Add other models to RAG implementation on Notebook 3

Updates:

• Notebook 3: adds Cohere Command R+ and Mistral Large from Azure AI Studio Serverless deployments
• Updated to latest versions of OpenAI and Azure AI search API: 2024-03-01-preview

Adding FrontEnd page for FastAPI/LangServe backend

Updates:

• Added: apps/frontend/pages/3_FastAPI_Chat.py as frontend for the FastAPI/LangServe API
• Included instructions to deploy both APIs (BotService and FastAPI) at the same time in the same web app service using slots
• Improved Notebook 6: Better explanations for OpenAI tools API and LLM binding with Tools

Langchain with LCEL and using latest APIs

This is a major release!
It includes the following updates:

• Demo Site is UP again!
• Latest Azure AI Search API (2023-10-01-preview)
• Latest Azure OpenAI API (2023-12-01-preview)
• Uses Integrated chunking and vectorization in Azure AI Search
• Newest version of LangChain which uses LCEL and Runnable protocol
• Everything works with GPT-3.5-Turbo and of course GPT-4-Turbo (1106 versions and newer). This unblocks customers that:
  • Cannot afford GPT4
  • Are not comfortable with the performance (speed) of GPT4 and are willing to exchange quality of the responses for price and speed.
• All agents use the newest OpenAI tools API (preferred over the Functions API that is being deprecated). This enhancement allows for the parallel calling of Functions/Tools.
• New Notebook: "Building our First RAG bot - Skill: talk to Search Engine". Works good with GPT3.5-Turbo-1106
• Dynamic selection of LLM model at runtime
• Streaming and Async works easily using LCEL
• New Notebook: "Using the Bot Service API programmatically." With this notebook now you can access the Bot Service endpoint programmatically in any language. This allows customers to connect the bot to their ReactJS or AngularJS front end.

API Search Capability

This release includes the following updates:

1. API Search Capability (Notebook 09):
   • Introducing an agent adept at querying various types of APIs to retrieve information and effectively answer questions.
2. Updated Versions of OpenAI and Langchain:
   • The software has been upgraded to the latest versions of OpenAI and Langchain.
3. Enhancements in the Brain/Master Agent Notebook:
   • The notebook has been improved, along with more comprehensive explanations.
4. Updated Visual Materials:
   • This includes revisions to images, Visio, and PowerPoint files.
5. Expanded Bot Service API:
   • The Bot Service Bot API now features '@apisearch' as a new skill.

CosmosDB in bot.py

What is in this release:

Based on customer's feedback:

• Improved run_agent() function in utils.py to deal with parsing errors
• Added Cosmos DB to bot.py to avoid memory sharing among concurrent users.

v2.0.0 - Vector Search Capabilities with Azure Cog Search

What's New in this 2.0 release:

• This is a major change , jumping versions from 1.7 to 2.0
• The way documents search work changes from Semantic Search results to Vector Similarity results
• Now two indexes coexist: text-based indexes and its corresponding vector-based indexes
• Vectorization of chunks happens on demand as users discover documents (vs embedding all documents at ingestion)
• Azure Document Intelligence (former Form Recognizer) is added to the repo in a new Notebook 4 when dealing with Complex PDF files. Manual chunking (per page) and push to vector-only index
• Improvements in prompts and functions
• New Architecture image

v1.7.0 - Adding QnA over Websites

What's New:

• Improved Bing Chat system prompt that allows ask questions to any website and the bot will provide a thorough answer.

v1.6.0 - Adding Github Spaces (dev container) and Github Actions

What's New:

1. New Notebook 07 added - Bing Chat Clone, implements callback handlers and mimics Bing Chat functionality
2. Developer tools:

• Dev Container
• Github actions for deployment

v1.5.0 - Addind Bot Framework App

What's new:

• There are 2 apps now: Backend Bot API powered by the Bot Framework and Bot Service, and a Frontend web app power by Streamlit that exposes a Search and a Bot Interface (using the Bot Service Direct Channel)
• Adds CosmosDB as persistent database for conversation history
• Removes HuggingFace BERT models since they are computationally too expensive for any web app
• Other minor improvements.

v1.4.0

What's new:

• New Notebook: 04-Adding_Memory.ipynb - Explains how to add memory to an application using LLMs.
• Adds Chat Interface to the Application
• Improved functions and readability to Notebook 3 and 4