Open nidhoggr-nil opened 1 month ago
Many agent and llm frameworks are starting to use structured output to make integration and parsing output work consistently.
I could not find any documentation if turbomind supports this or if it is in the works, but it seems very relevant.
Something like lm-format-enforcer would be very nice to have.
This should then be made to support tool calls via the openai API, so that turbomind becomes openai REST compatible with any framework.
LMDeploy api_server is compatible with openai API Is https://lmdeploy.readthedocs.io/en/latest/serving/api_server.html#integrate-with-openai what you are looking for?
@AllentDan is working on this feature
Motivation
Many agent and llm frameworks are starting to use structured output to make integration and parsing output work consistently.
I could not find any documentation if turbomind supports this or if it is in the works, but it seems very relevant.
Something like lm-format-enforcer would be very nice to have.
This should then be made to support tool calls via the openai API, so that turbomind becomes openai REST compatible with any framework.