* update api 1.8

* fix paddlerec readme

* fix

* fix

* fix 2.0

* fix 2.0

* fix 2.0

* fix 2.0

* fix 2.0

* fix

* add pure fp16 training.

* update the usage of cast_parameters_to_fp16.

* use fp32 to initilize parameter values and then cast to fp16.

* enable multi_precision when training with fp16.

* add static loss scaling.

* add the rescale_grad function in the pure fp16 training.

* use cast to complete the pure fp16 training.

* use sum loss for fp16 trainging and mean loss for fp32 training.

* refine codes according reviewers' suggestions

* Calculate and print the average time for video models.

* Calculate and print the average time for dygraph of slowfast.

* Print ips for nextvlad, slowfast, tsm and tsn.

add mso significant commit, because of creep API change, no more steps toward to be a time killer and all of it is crap (#4953)

average reader cost; test=develop

* add reader cost, and change speed to ips; test=develop

* fix inf bug; test=develop

* Create inception_v3.py

* Update __init__.py

* Create InceptionV3.sh

* fix load in metric learning

* fix load in metric learning

* fix typo

* fix typo

* fix video bug and slowfast benchmark bug

* refine bmn doc

* update api 1.8

* fix paddlerec readme

* fix readme

* update 2.0 api

* fix, test=develop

* update videotag, add fine-tune code and doc

* refine datalist

* refine eval.py

* Revert "add fuse_bn_add_act_ops args (#4864)"

* add fuse_bn_add_act_ops args

* update api 1.8

* fix paddlerec readme

* update 20 , test=develop

* fix ptb_dy time print for benchmark, test=develop

* use yiqun(Xreki)'s PR, test=develop

* add empty line, test=develop

* add empty line, test=develop

* update api 1.8

* fix paddlerec readme

* delete some model

* delete some model

* update

* fix ptb_dy time print for benchmark, test=develop

* refine resnet benchmard print, test=develop

* refine benchmark log

test=develop

* refine log

test=develop

* fix ptb_dy time print for benchmark, test=develop

* add ips print for language_model, test=develop

* fix ptb_dy time print for benchmark, test=develop

* add ips print for ptb_lm

* add tokens/sec; test=develop

* change np.array to np.asarray to avoid data copy; test=develop

test=develop