April 23: Support for Python and TypeScript SDKs, latest OpenAI, Cohere & Groq models, bug fixes

New Features

  • 💡 Graphlit now supports a native Python SDK, using Pydantic types. The Python SDK is code-generated from the current GraphQL schema, but does not require GraphQL knowledge. You can find the latest PyPi package here. The Streamlit sample applications have been updated to use the new Python SDK.

  • 💡 Graphlit now supports a native Node.js SDK, using TypeScript types. The Node.js SDK is code-generated from the current GraphQL schema, but does not require GraphQL knowledge. You can find the latest NPM package here.

  • 💡 Graphlit now supports the 2024-04-09 models in the OpenAI model service. GPT4_TURBO-128K will give the latest OpenAI GPT-4 model, following this model list. We have added the GPT4_TURBO_128K_2024_04_09 enum to specify the new model.

  • 💡 Graphlit now supports LLaMA3 70b, LLaMA3 8b and Gemma 7b models in the Groq model service.

  • 💡 Graphlit now supports the Command R and Command-R+ models in the Cohere model service.

  • Added support for Jina reranking, using the JINA reranking model service type in the reranking retrieval strategy.

  • Updated the Cohere reranking model to use the latest v3.0 model.

  • Increased the reliability of parsing LLM responses, in cases where they don't follow the JSON schema.

  • Cleaned up nullability of GraphQL parameters, so parameters better reflect if they are required or optional, or allow nulls.

  • Added missing deleteWorkflows and deleteAllCollections mutations.

  • Split out reranking model service type as RetrievalModelServiceTypes enum.

Bugs Fixed

  • GPLA-2114: Adding content to collections not syncing search index

  • GPLA-2511: Failing to render any conversation sources with section retrieval and text content

Last updated