Smaller LitGPT like configs

[carmocca]

What does this PR do?

Alternative to #219

Weight memory usage before:

gpt-neox-like 6532864 26.131456 MB
llama1-like 602432 2.409728 MB
long-context-like 4301120 17.20448 MB
llama2-like 602432 2.409728 MB
falcon-7b-like 4359040 17.43616 MB
falcon-40b-like 1589760 6.35904 MB
codellama2-like 819520 3.27808 MB
mixtral-like 3308800 13.2352 MB

After:

gpt-neox-like 141056 0.564224 MB
llama1-like 602432 2.409728 MB
long-context-like 602432 2.409728 MB
llama2-like 602432 2.409728 MB
falcon-7b-like 77700 0.3108 MB
falcon-40b-like 398080 1.59232 MB
codellama2-like 401568 1.606272 MB
mixtral-like 257856 1.031424 MB

[t-vi]

Thank you @carmocca ,I don't think we need to go smaller than we currently have. Anything < 500MB GPU memory should be more than fine.

[carmocca]

Okay. We can always revisit this