Add Upcasting for FSDP in Mixed Precision. Add Concept Guide for FSPD and DeepSpeed. (#2674)
* draft fsdp vs ds
* reframe to migration doc
* updated functionality section
* cast to float32
* improvements to float32 casting
* some cleanup
* addressed @pacman100's comments
* Apply some of @muellerz suggestions
Co-authored-by: Zach Mueller <muellerzr@gmail.com>
* change to subsections
* changed the manner upcasting warnings are surfaced
* update document to discuss fsdp and ds plugins. minor fixes.
* @muellerzr's new suggestions
Co-authored-by: Zach Mueller <muellerzr@gmail.com>
* explain all-or-nothing
* add @pacman100's comments
Co-authored-by: Sourab Mangrulkar <13534540+pacman100@users.noreply.github.com>
* minor fix
---------
Co-authored-by: Yu Chin Fabian Lim <flim@sg.ibm.com>
Co-authored-by: Zach Mueller <muellerzr@gmail.com>
Co-authored-by: Sourab Mangrulkar <13534540+pacman100@users.noreply.github.com>