You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Came up in the MF-Workshop at TH Köln 2024-10-08, inspired by C. Marutschke's idea to let an LLM train using flux and fix:
Provide an "opener" that scans an input (let's say, read up to first 30 kB) and probe that with the various openers and handlers (combined with well known parameters (like recordtagname="lido")). Compare the output, if any other than Exceptions. Pass the one with the most output (or/and most deeply nested/complex structure) as a convenient wrapper.
Could be very convenient for newbies.
I think this is not too complicated and wouldn't suck much ressources.
The text was updated successfully, but these errors were encountered:
Hm, involves a little too much "magic" for my taste, I'm afraid... Not to mention that I would contest the assessment that it might not be "too complicated" (for starters, some data are just invalid when arbitrarily truncated).
We could, however, establish conventions for file extensions and MIME types to associate certain input files/URLs with their appropriate readers/handlers (similar to what xdg-open does; see also #96).
Came up in the MF-Workshop at TH Köln 2024-10-08, inspired by C. Marutschke's idea to let an LLM train using
flux
andfix
:Provide an "opener" that scans an input (let's say, read up to first 30 kB) and probe that with the various openers and handlers (combined with well known parameters (like
recordtagname="lido"
)). Compare the output, if any other thanExceptions
. Pass the one with the most output (or/and most deeply nested/complex structure) as a convenient wrapper.Could be very convenient for newbies.
I think this is not too complicated and wouldn't suck much ressources.
The text was updated successfully, but these errors were encountered: