Enable gauntlet training (#501)
* add subset num batches
* add subset num batches
* remove tiktoken
* remove openai import
* remove bad line
* foo
* add training callback
* modify yamls
* implement train
* fix indexing to get most recent eval result
* finish
* finish
* finish
* finish
* finish
* foo
* foo
* working on debugging changeS
* [wip] removing logger dependency from model gauntlet
* remove logger from eval
* remove logger from eval
* remove logger from eval
* debug
* debug
* debug
* debug
* fix
* finish?
* fix bug
* merge main
* fix bug
* minor nits
* add unit test
* add unit test
* fixing unit test [wip]
* trying to unit test :(
* trying to unit test :(
* test not added on github
* fix all errors
* lint
* ignore pyright
* change yaml
* fix testing
* remove unnecessary line
* fix typing in test
* fix typing
* fix typing
* ignore pyright
* done :)
* change name to eval gauntlet
* merge main
* merge main
* fix broken test
* shrink test
* try commenting out test_train
* rm redundant test
* remove broken test
* trying to fix test
* done :))))))
* reduce training to 1ba
* address daniel's concerns about back compat
* address daniel's concerns about back compat
* fix license
---------
Co-authored-by: Cody Blakeney <cody@mosaicml.com>
