DSPy Assertions: Computational Constraints for Self-Refining Language Model Pipelines

Explore the concept of LM Assertions in this 38-minute talk by Arnav Singhvi, a recent CS graduate from UC Berkeley and researcher with the Stanford NLP Group. Dive into the challenges of ensuring language models adhere to important constraints when chaining them as composable modules. Learn about a new programming construct for expressing computational constraints that LMs should satisfy, integrated into the DSPy programming model. Discover strategies for compiling programs with arbitrary LM Assertions to create more reliable and accurate systems. Understand how these assertions can be integrated at compile time through automatic prompt optimization and at inference-time via automatic self-refinement and backtracking. Gain insights into the DSPy framework and its potential to revolutionize LM programming. This talk, presented by MLOps.community, offers valuable knowledge for those interested in advancing language model capabilities and constraints in AI development.

DSPy Assertions: Computational Constraints for Self-Refining LM Pipelines // Arnav Singhvi // Talk