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
This is probably not quite the place to ask this, but I'll go for it nonetheless, perhaps someone will read it!
I've been wondering why the Fill-In-The-Middle (FIM) was not more widespread in text models as well (and not just for code models)? It still seems to me to be a great sort of capability for all sorts of writing tasks and processes, where the chat mode/instruct is quite overkill.
Given the claims that it's possible to get FIM "for free" by only modifying a subset of the dataset and adding a few special tokens, it seems peculiar that it is not widespread a practice? Or is it because people are anticipating that chat capabilities have just rendered this moot (and one can now ask the assistant to rewrite part of an existing document at will, given the right interface)?
Thanks for reading, have a great day!
The text was updated successfully, but these errors were encountered:
Hi there,
This is probably not quite the place to ask this, but I'll go for it nonetheless, perhaps someone will read it!
I've been wondering why the Fill-In-The-Middle (FIM) was not more widespread in text models as well (and not just for code models)? It still seems to me to be a great sort of capability for all sorts of writing tasks and processes, where the chat mode/instruct is quite overkill.
Given the claims that it's possible to get FIM "for free" by only modifying a subset of the dataset and adding a few special tokens, it seems peculiar that it is not widespread a practice? Or is it because people are anticipating that chat capabilities have just rendered this moot (and one can now ask the assistant to rewrite part of an existing document at will, given the right interface)?
Thanks for reading, have a great day!
The text was updated successfully, but these errors were encountered: