support auto parallel in dit and largedit

[jeff41404]

trafficstars

support auto parallel in model of Dit and LargeDit, include:

• support model of Dit and LargeDit to be trained using auto parallel in dynamic graph mode (hereinafter referred to as auto).
• support model of Dit and LargeDit to be trained using auto parallel in dynamic to static graph mode with pir (hereinafter referred to as auto-static).
• support hybrid parallel(DP、Sharding、MP) in model of Dit and LargeDit, PP under verification.

compare loss of manual parallel(present, hereinafter referred to as manual), auto and auto-static

compare loss diff of manual and auto

compare loss diff of manual and auto-static

compare loss diff of auto and auto-static

[paddle-bot[bot]]

Thanks for your contribution!

[CLAassistant]

All committers have signed the CLA.

[jeff41404]

https://github.com/PaddlePaddle/PaddleMIX/tree/develop/ppdiffusers/examples/class_conditional_image_generation/DiT 在这里新增自动并行的相关文档吧

感谢提醒，已增加使用自动并行进行训练的文档说明