Use rope_theta from config & correct llama3 freqs

Llama3 dev

Fixed a lot of grammar/spelling mistakes in all the documentation files.

fix the option to support 'streaming=false'

fix: fix bug not found config

fix bug on config convert_checkpoint

Added back optimizer

added BPT for training very long sequence

FIx convert script

Fix missing backslash typo in pretrain_llama_7b.sh