Update_pretrain_benchmarks (#543)

* changed loc to look for yamls

* ex.py submits run ok, needs correct data loading commands

* modified to work w/ new sdk features

* generates some results, off from what Vitaly calculated

* update some metrics, scripts

* fixed some bugs, currently using this script

* delete old prints \& ex.py

* add compile settings

* added some scripts

* printing compile config deleted

* update benchmarks

* modify script

* update ARGS to run locally

* add local script

* fix submit_benchmarks LOCAL

* update

* reee

* ;|

* ;|

* :|

* :|

* wip

* merge xformers

* revert the attention commit

* undo attn on this branch

* cleanup files

* remove h100 shell script

* remove random things

* use dict

* use mosaicml llmf

* updated a100 numbers

* added back old numbers that I didn't reproduce

* Update submit_benchmarks.py -- enable fp8 deps iff h100 and fp8

* Added TFLOP column

* whoops -- fix

* Correct TFLOP units

* Punt H100 350m numbers

* calculate TFLOPS not FLOPS

* lint & pyright

* daniel review v1

* rm compile support in profiling

---------

Co-authored-by: Chris Rinard <[email protected]>
Co-authored-by: Vitaliy Chiley <[email protected]>
Co-authored-by: Vitaliy Chiley <[email protected]>