v1.0.0-publish

Everything seems to be ready

v0.3.2-cpt-configs_and_scripts

• add final data portion of sheared llama
• add gate_network_type, moe_calculator_score_scale_factor, and update prob_map arguments in config
• add exec scripts

v0.3.1-cpt-dynamic_batch_loading: Llama2 CPT with Dynamic Batch Loading

• Llama2 CPT with 4096 context length training.
• Dynamic batch loading from ShearedLlama Implementation.

v0.2.1-cpt-13b: Fix 13B CPT bugs

Merge pull request #31 from pjlab-sys4nlp/scaling_13b

CPT: fix tb logging, fix grad ckpting, faster data loading