stabilityai

/

StableBeluga2

like 884

try alphawave-py (https://tuuyi.io) - python package that provides OpenAI like function calling as well as syntax validation/repair in JSON, TOML, and python, and prompt templating for dozens of OS LLMs.
pip3 install alphawave available on pypi (works with gpt-3.5/4 as well!)

Thank you! I'll take a look!

docs are kinda sparse, but it is finally pretty stable. feel free to ask questions. (where? maybe on https://github.com/stevenic/alphawave-py discussions page?)

yeah, thank you. Function calling is a little different service, you are defining what kind of response the model would give you. The model has to be fine tuned to provide this.

No, it does provide function calling, but in a slightly difference way. You add 'commands' to the agent, and it provides a prompt to the llm informing it of functions available and syntax to use. For example, it includes a search function/command built in, so if I type 'what is the weather tomorrow for berkeley?' the llm will likely decide to invoke the search command, and return a json form aw-py will recognize and use to invoke the searchCommand (you can add whatever commands to the system). It will then return the final result to the user (for search, a digestion of various web pages and their urls.) The system, when it does see a command in the LLM return, validates the syntax and typing of the command arguments and attempts to auto-repair (via feedback calls to the llm). Not sure we're communicating well, if this seems like what you need, or are interested in, happy to continue, but if not, sorry to waste your time.

Thank you so much for your advice. I'll check it for sure, everything is valuable and we are all learning and many things emerges all the time. My intention is to inspire stabilityai team to provide function calling fine tuning into their models, so it would be easily implementable into code. I'll look into your solution as well, I just need some dedicated time for this....