Prompt Priming

[distant maple]

The type of query would be:
Why is this query not valid?

-10m | plat == windows | DNS_REQUEST event/DOMAIN_NAME contains 'google' | event/DOMAIN_NAME as domain COUNT_UNIQUE(routing/sid) as count GROUP BY(domain)

And result is:
This query is not valid because it is missing a pipe (|) character between the DNS_REQUEST clause and the event/DOMAIN_NAME clause.

I know I could generate 100s of examples of query that cover all cases and operators available and then train a lesser model based on that. But to me the attraction is to be able to just give the model the documentation to the QL and have it validate queries in a human language.

That's why I was hoping to find a way to "prime" the -003 model with my documentation, and then just query with the user's query, instead of feeding the whole doc page on every single query.

And yes @unkempt island I think my use case falls between the DaVinci3 and older models.

[mighty oak]

You wont be able to use fine-tuning to teach it a manual

[faint lantern]

/imagin