Skip to content

Latest commit

 

History

History
116 lines (89 loc) · 4.76 KB

CHANGELOG.md

File metadata and controls

116 lines (89 loc) · 4.76 KB
Raw

Releases

v0.4.0

New features:

GENIE ("GenAI for Enterprise" platform on GCP)

  • Multimodal RAG for images using multi-modal embeddings for search
  • Chat file upload in both backend and React app. Upload files or specify URLs to pass to model.
  • Streaming of chat responses in React frontend
  • RBAC for model access. See README for LLM Service for docs on how to manage access to models for users.
  • Added user role management in React frontend, used for RBAC for models and query engines
  • Added documentation on authentication in the platform in docs/AUTH.md
  • Updated to recent releases of FastAPI (0.112.2) and associated libraries
  • Use L4 GPUs with Truss models
  • Chunk size and chunking class are now query engine build params
  • Add chunk size to React Query Admin engine build form
  • Switched to using llama_index.core.node_parser.SentenceSplitter for chunking by default
  • Updated default query generation model to Gemini Flash 1.5 - was set to Palm2
  • Added Microsoft login in React frontend
  • Add webscraping component written in Go for performance. Used for web data query engines with depth > 0.
  • Add timestamps to logging
  • New deploy script for React frontend

Fixes

  • Fixed download of PDFs from scraped sites
  • Fixed deletion of query engines in React frontend; made hard delete the default

v0.3.2

New features:

GENIE ("GenAI for Enterprise" platform on GCP)

  • Remove tools_service from default ingress and install instructions
  • Updated install guide to emphasize that React UX is recommended over Flutterflow and Streamlit
  • Add instructions for deploying Gemma 2B LLM service to GENIE

Fixes

  • Corrected README for React frontend to instruct to build app after setting config

v0.3.1

New features:

GENIE ("GenAI for Enterprise" platform on GCP)

  • Deploy using artifact registry
  • Add Query Engine Admin in React Frontend to create new Query Engines, view build jobs
  • Support basic query filters for GENIE (non-Vertex Search) Query Engines
  • Support basic RBAC on documents in GENIE query engines
  • Add script to install postgres with Cloud SQL

Fixes

  • Update to build and deploy process to fix Container registry deprecation which was causing deploys to fail in new projects.
  • Fix embedding generation bug causing documents to fail to be indexed

v0.3.0

New features:

GENIE ("GenAI for Enterprise" platform on GCP)

  • Add VertexSearch query engine type, using Vertex Search to index data and perform RAG retrieval
  • Add Integrated Search query engine type, allowing RAG retrieval across query engines
  • Add new React-based frontend app (see README to install and deploy)
  • RAG: Significant improvements to chunking; now checking context window length for generation
  • Integrate with LlamaIndex for chunking and data connectors
  • Update langchain libraries (April 2024)
  • Add SharePoint/OneDrive data connector, support SharePoint data sources
  • Improved agent dispatch using langchain LLMRouterChain
  • Background job-based queries, agent dispatch and agent runs
  • Add support for Huggingface embeddings
  • Support multimodal embeddings
  • Support multimodel llm generation with Gemini
  • Include instructions to upgrade/resize llm-service node pool

Core Solution Services

  • Create separate common_ml image to improve build/deploy times for LLM Service
  • Pin version of firestore emulator used in unit tests and deploy. Requires firestore emulator to be started manually for unit tests.
  • Create an environment variable profile on the jumphost during install
  • Add documentation for setting up the CI/CD pipeline in Github

Fixes

  • Vertex embeddings now check token length limit, and use latest guidance on API max chunks and rate limits
  • Allow DbAgent to be called directly via agent run endpoint
  • Do not require OpenAI api key for default install

Known Issues

  • e2e testing not currently enabled
  • Some known install issues (see INSTALL guide)

v0.2.0

Features:

GENIE ("GenAI for Enterprise" platform on GCP)

  • Chat, text generation and embeddings using Vertex models (Feb 2024) and Langchain supported LLMs
  • Basic RAG pipeline using AlloyDB / PGVector (recommended) or Vertex Matching Engine as vector store. Supports web-based datasources and GCS as a data source.
  • Agents using Langchain
  • Agent tool examples: send email using gmail, create spreadsheets, consult a rules engine
  • DB Agents to retrieve data from SQL databases
  • Basic UX in Streamlit and Flutterflow (see READMEs in those components for details)
  • Support for Llama2 via Truss (experimental)

Core Solution Services

  • Authentication
  • User Management
  • Rules Engine
  • Job Service

Known Issues

  • e2e testing not currently enabled
  • Some known install issues (see INSTALL guide)
  • Flutterflow UX not deploying successfully