microsoft/semantic-kernel python-1.20.0

on GitHub

Release Notes

New Features

• Chat History Reducer #10190 by @moonbox3

  • Example: Chat history reducer with agents.
  • Example: Using a chat history reducer in chat completion.
  • Example: Chat history reducer including FunctionCallContent and FunctionResultContent.

• Deepseek Service in Concept Samples #10306 by @TaoChenOSU

• Prompt Template Config and Kernel Args Support for Agents #10298 by @moonbox3

  • Example: Using a prompt template config and kernel arguments with a ChatCompletionAgent.
  • Example: Using a prompt template config and kernel arguments with Assistant Agents.

Note: The OpenAIChatCompletion and AzureChatCompletion classes now support an instruction_role keyword argument. This is particularly useful when using reasoning models like o1. When set, any AuthorRole.SYSTEM messages are converted to AuthorRole.DEVELOPER before being sent to the model.

Example usage:

chat_service = OpenAIChatCompletion(service_id=service_id, instruction_role="developer")

Python Package Updates

• Bump Python version to 1.20.0 for a release. #10331 by @moonbox3

Enhancements and Fixes

• Improve handling for kernel plugin from file. #10286 by @moonbox3

• Improve hashing of CMC and SCMC items. Add tests. #10332 by @moonbox3

• Improve agent samples and chat history handling. #10301 by @eavanvalkenburg

• Allow enums to be passed into on_function_result. Improve handling of FRC result so it can be hashed. #10316 by @moonbox3

Bug Fixes and Improvements

• Bug fixes from GraphRAG insights. #10270 by @eavanvalkenburg

Full Changelog: python-1.19.0...python-1.20.0