[zero] prevent poor configs from running w. zero-offload (microsoft#2971

)

Add a runner for MosaicML cloud (#44)

* add mosaic launcher string constant

* first attempt at mosaic multinode runner for gptneox

* typo

* actually add mosaic runner

* string cast

* debug

* fix

* actually fix?

* strip extra space

* debugging

* debugging

* correctly set env vars

* drop cd

* add env vars via env instead of export commands

* fix

* try using slurms arg parsing

* debug print

* debug print

* print debug

* cleanup

* try getting world info from the hostfile

* add missing init arg

* more cleanup

* remove more prints

version bump to 0.3.10

Elastic training support (microsoft#602)

Co-authored-by: Samyam Rajbhandari <[email protected]>

bump to 0.3.8

calculate grad norm wrt sub partitions

bump to 0.3.7

bump to 0.3.6 and fix manifest to include reqs (microsoft#561)

bump to 0.3.5

bump version 0.3.4