Junwei Deng

> When importing `Model` from `bigdl.nano.tf.keras` after creating datasets etc., error `Inter op parallelism cannot be modified after initialization` may occur (for example in the codes [here](https://github.com/intel-analytics/BigDL/blob/dd559a48199219eb9b9ee27c10dccf70d5140562/python/nano/tutorial/notebook/training/tensorflow/accelerate_tensorflow_training_multi_instance.ipynb)):  @TheaperDeng Is...

we need more words to make sure our users to understand how to save/load. But I think the save/load might be refactored recently.

Do we have any bkm to support intel-tensorflow 2.8 in nano before? @MeouSker77

@devWangBin is working on this.

Should we use a setter for this member? since we want to use it as a external API

> > Should we use a setter for this member? since we want to use it as a external API > > Like this? > > ```python > yhat =...

just found this PR......, we definitely need to merge this. @zhentaocc do you have any changes need to do on this PR? I will merge it tomorrow.

@hjzin any comments about this PR? I will merge it today if you are ok with it.

maybe add a collate_fn on users' dataloader could be a helpful feature. but we need to define the api first.

> ``` > # ipex > model_ipex = Trainer.trace(model, accelerator=None, use_ipex=True, ...) > model_ipex(*args) > > # jit > model_jit = Trainer.trace(model, accelerator="jit", input_sample=input_sample, ...) > model_jit(*args) > > #...