Merge pull request #205 from HalberdOfPineapple/wenxuan/mtrain

update expr scripts

apply precommit to current PR

update README and train script

update data prepare script

add README

fix and launch dense

update trainer and data prepare

update 3B model training setting with 32 GPUs

update parallel, setup and 3B training script

refactored some structures

Add Qwen-2.5-0.5B

Refactored the distributed operator testing

fix dependencies and update training settings

Fix(LeanK): fix import path (#189)