Spaces:

---

nanotron

/

ultrascale-playbook

like 1.9k

Running

Fixed a few typos and merged it with #95:
focus on checkpointing[-the-] expensive {+the+} feed forward computations.
For stability [-reasons-]{+reason+}
to trade [-off-]{+of+} memory

I've tried to push this directly to PR #95 but it was not possible (maybe that's obvious).

Just merge it in #97 thanks a lot (wasn't able to merge this one bc of conflict)

Hi @eliebak , I've just solved the conflicts with #97

I still see some conflict with src/index.html, i think you might need to rebase with the main (pushed the change you suggest in #97 but if there is some new one please modify both src/index.html and distil/index.html and will merge it your PR directly)

Thanks, I'll figure it out

Hi @eliebak , let me know if it's fine now and sorry for the multiple iterations here.

I'm not sure why but there is already some changes that are corrected in the main branch, but still displaying the old version in the diff

Humm, I'll give it another try in a bit

I'm closing this one - I decided to start fresh with #99. If it does not go through, I'll figure out a better way of helping out. Thank you for the patience :)