GitHub user zyratlo added a comment to the discussion: Proposal: Statement-level Python-to-Texera workflow compiler MVP
As the developer of the LLM-based tool, here is my interpretation of how these two tools offer different experiences. The LLM tool is non-deterministic, as is the nature of LLMs in general. We can't guarantee that the source code is migrated 100% accurately, but the end result tends to be more understandable for humans since the LLM can group code in UDFs semantically. The user experience we wanted to enable with the LLM tool is to provide a baseline that the user can modify after LLM generation, instead of requiring them to create a new workflow from scratch. The source code is also shown with a mapping to help with the editting process. In contrast, the compiler approach is deterministic and can be checked for equivalence against the source, but at the trade off of human understandability. The main reason I see for having both is that if a user has source code they don't want to expose to an LLM, so they would use the compiler version. GitHub link: https://github.com/apache/texera/discussions/8160#discussioncomment-18585421 ---- This is an automatically sent email for [email protected]. To unsubscribe, please send an email to: [email protected]
