v0.2.2
[gemini] update ddp strict mode (#2518)
[example] update gpt example for larger model scale (#2211)
[Gemini] update the non model data record method in runtime memory tracer (#2128)
[hotfix] remove potiential circle import (#1307)
[hotfix] polish chunk import (#1787)
[hotfix] add deconstructor for stateful tensor (#848)
[hotfix] fix zero ddp warmup check (#2545)
[Gemini] fix the convert_to_torch_module bug (#2269)
[gemini] accelerate adjust_layout() (#878)
[Gemini] free and allocate cuda memory by tensor.storage, add grad hook (#2040)