Changelog for genaiPlatform Version 3.0

Copy link to this section

[3.0.0]

Copy link to this section

Added

Copy link to this section

Tools

• GenaiCore.Tool.NativePy.Spec#allowCrossRuntimeCalls
  • Added support for cross-runtime native py tool calls.
• GenaiCore.Tool.NativePy#initialize
  • Added support for initialization in native py tools.
• GenaiCore.Toolkit#publishToolkit
  • Added support for publishing toolkits to the application package.

Copy link to this section

Vector Store

• GenaiCore.Vector
  • Added optional base type for vector definitions. Standardizes key fields for vectors in the vector store.
• GenaiCore.VectorStore.Pg
  • Added GenaiCore.VectorStore.Pg as an alias to GenaiCore.PgVector. GenaiCore.PgVector is now deprecated and will be removed in a future release, so use GenaiCore.VectorStore.Pg instead.
  • Added support for providing BatchImportSpec and BatchExportSpec to GenaiCore.VectorStore.Pg#importFromFiles and GenaiCore.VectorStore.Pg#exportToFiles APIs, respectively.
• GenaiCore.VectorStore#addVectors
  • Added abstract API for adding vectors to a vector store + implemented for GenaiCore.VectorStore.Pg.

Copy link to this section

Unstructured Document Processing

• GenaiCore.Unstructured
  • Added types and APIs under the GenaiCore.Unstructured namespace to support processing of unstructured documents.
  • Added support for defining serial parsing/chunking pipelines with GenaiCore.Unstructured.Processor
  • Added support for defining scalable parsing/chunking jobs with GenaiCore.Unstructured.Job
  • Added tutorial for unstructured document processing.
• GenaiCore.Mew3
  • Added support for C3-maintained mew3 document extraction library as an out-of-the-box parser/chunker.
  • Added tutorial for using mew3 with GenaiCore.Unstructured.Processor and GenaiCore.Unstructured.Job.

Copy link to this section

Observability

• GenaiCore.Observability.Tracer
  • Added support for custom trace-level aggregations by sub-typing Observability.Tracer:
    • input/output (i.e. query/response)
    • LLM token sum/max
    • LLM latency sum/max

Copy link to this section

LLM

• GenaiCore.Llm.Router
  • Added support for an out-of-the-box router for LLM calls.
  • GenaiCore.Llm.Router.Random routes to a randomly selected auth (or model) on the first request and then again to a randomly selected auth (or model) after any retry-able error is encountered, selecting from a user-provided list.
  • GenaiCore.Llm.Router.LinearFallback routes to a new auth (or model) when a retry-able error is encountered, selecting in order from a user-provided list.

Copy link to this section

Changed

• py-llm and related runtimes upgraded to update the version of litellm from 1.63.11 to 1.67.1:
  • py-litellm
  • py-vertexai_litellm
  • py-bedrock_litellm
• updated existing entity types to also be SeedData:
  • GenaiCore.Tool
  • GenaiCore.Toolkit
  • GenaiCore.VectorStore
  • GenaiCore.SessionStore.Kv
  • GenaiCore.Embedder.Hf

Copy link to this section

Fixed

• GenaiCore.Embedder.Hf to utilize all available GPUs by default.