Skip to content

Comment on Sampling and structured outputs in LLMsparent

Comments

One pattern that I've seen develop (in PydanticAI and elsewhere) is to constrain the output but include an escape hatch. If an error happens, that lets it bail out and report the problem rather than be forced to proceed down a doomed path.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.