identify more robust output format for data synthesis

[cartazio]

the first version uses JSON, which can often be malformed, and theres no good error recovery in that case, need to identify and switch to a more "error tolerant" self aligning format. (meaning we can skip a bad pair and recover useful outputs)

[cartazio]

i've a few ideas for this i'll try.

[mmirman]

XML is the key. Its really the only key. Foundation LLMs have a lot of XML in their outputs so are super-primed to output it. Also you can scrub inputs by ensuring the tags are unlikely to be guessed.

Also this is what LMQL is for!

[cartazio]

Good points. I’ll dig into this direction

[mmirman]

I do want to add regex and lmql to our platform btwBest,Matt==================Dr. Matthew Mirmanhttps://www.mirman.comOn 30 Jun 2023, at 7:15 PM, Carter Tazio Schonwald ***@***.***> wrote: Good points. I’ll dig into this direction —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you commented.Message ID: ***@***.***>

[cartazio]

should be a bit more robust, we can revisit this later