v0.3.6
[llama] fix training and inference scripts (#5384)
[moe] fix tests
[shardformer]: support gpt-j, falcon, Mistral and add interleaved pipeline for bert (#5088)
[misc] update pre-commit and run all files (#4752)