'ColossalAI/examples/language/gpt' and 'ColossalAI-Examples/language/gpt' which is the latest demo?

[xinghua3632]

ColoInitContext is more recommended than ZeroInitContext to build model?
In 'ColossalAI/examples/language/gpt', demo in gemini and three demos in experiments which is your recommended? I just want to use all solutions in Colossa for my projects, data parallel, model parallel, and pipeline parallel;
When I set TPDEGREE=2 in run_gemini.sh in 'ColossalAI/examples/language/gpt', the usage of GPU Memory is bigger than TPDEGREE=1, meanwhile the model cannot load pretrained parameters directly ,because the tensons in model are splited to half, but tensors in checkpoint are whole tensors.
Do you have a completed and latest demo that use all parallel mode?

[xinghua3632]

TKS, I have a new question