Hey everyone! I am beyond excited to publish our 59th Weaviate podcast with Luca Beurer-Kellner, the lead author and creator of LMQL!
LMQL is a programming language for LLMs, a really interesting and unique direction amongst the emerging development of LLM frameworks and tooling. I was really blown away by the elegance of the syntax, and I highly recommend checking out the LMQL playground. Not only is the LMQL playground a great way to learn LMQL particularly, it is one of the world’s best visualizations of complex LLM execution, providing an interactive sandbox to explore!
We discussed many topics on the podcast from Luca’s research background in Programming Languages and how that has shaped his perspectives on Constrained Sampling, the analog of LLM output nil pointer exceptions, and the effort to tame this chaos with LMQL! We also discussed how this fits into existing LLM frameworks such as our friends at LlamaIndex, LangChain, Haystack, MS Semantic Kernel, Jina AI, and others! We also discussed tool use with the Gorilla large language models and the general perspective of a master model such as GPT-4 that routes inferences to cheaper specialized models!
Finally we concluded with discussions on future directions! Luca really opened my eyes about the future of composable models and RETRO-style RAG architectures, can’t wait to see that develop further!
I really hope you enjoy the podcast, as always I am more than happy to answer any questions or discuss any ideas you have related to the content in the podcast!