logan-markewich.bsky.social
@logan-markewich.bsky.social
Head of Open-Source @llamaindex
Working on adding this actually. The idea is that structured output becomes a tool, and then we set tool_choice="required" on the LLM. If the LLM calls that "tool" it ends the agent loop

The longer story is we need to update some interfaces to support this across LLM integrations, but nearly there
May 27, 2025 at 10:57 PM
github.com/run-llama/ll...

This is definitely more specific feedback than "workflows are broken", thanks! Easy enough fix -- it was already using generics in the base class, some integrations incorrectly overrode it

The nice thing about open-source software is anyone can make a PR 😁
type safe structured predict in overriden methods by logan-markewich · Pull Request #18290 · run-llama/llama_index
Some LLMs overwrote the structured_predict methods and did not use proper generics, this PR remedies that
github.com
March 27, 2025 at 1:05 AM
New repo to distance from Microsoft?
November 17, 2024 at 5:09 AM