Improve log, save origin yaml, and fix adan

[SamitHuang]

Thank you for your contribution to the MindOCR repo.
Before submitting this PR, please make sure:

• You have read the Contributing Guidelines on pull requests
• Your code builds clean without any errors or warnings
• You are using approved terminology
• You have added unit tests

Motivation

Improve log info.
Archive original yaml config used in training to better ensure reproduction.
Fix Adan optimizer (sync mindcv)

Test Plan

(How should this PR be tested? Do you require special setup to run the test or repro the fixed bug?)

Related Issues and PRs

(Is this PR part of a group of changes? Link the other relevant PRs and Issues here. Use https://help.github.com/en/articles/closing-issues-using-keywords for help on GitHub syntax)

[hadipash]

if the number of samples in the evaluation set is a prime number, then the batch size will be set to 1. This can significantly increase evaluation time. Can we just set drop_remainder to False for the evaluation set and leave the batch size as it is?

[zhtmike]

agree. The time consumption of run batch size = 1 is usually longer than the time running with two different batch sizes, the compiling time of model due to the different batch size is negligible

[SamitHuang]

Even when we set drop_remainder to False, the last batch will be padded to batch_size, leading to an inaccurate evaluation result.

[zhtmike]

I remember it is not padded, the remainder will be output with different batch size.

from mindspore.dataset import GeneratorDataset

dataset = GeneratorDataset(range(10), 'data').batch(4)
for x in dataset.create_tuple_iterator(num_epochs=1):
    print(x[0].shape)

>>>(4,)
>>>(4,)
>>>(2,)

[hadipash]

this is not shutil. shutil is a built-in python library, so no need to install with pip.

[zhtmike]

The comment out print function can be removed for better readability

[zhtmike]

btw model.eval API indeed will have undefined behaviour like padding at the end of the iteration. Should use it with care

import mindspore as ms
import mindspore.nn as nn
from mindspore.dataset import GeneratorDataset

class Iter:
    def __len__(self):
        return 10
    
    def __getitem__(self, index):
        return index, index 

dataset = GeneratorDataset(Iter(), ['data', 'label'], shuffle=False).batch(4)

class PrintNet(nn.Cell):
    def construct(self, x):
        print(x)
        return x

loss = nn.MSELoss()
net = PrintNet()
model = ms.Model(net, loss, metrics={'mse'})
model.eval(valid_dataset=dataset)
>>>Tensor(shape=[4], dtype=Int64, value=[0 1 2 3])
>>>Tensor(shape=[4], dtype=Int64, value=[4 5 6 7])
>>>Tensor(shape=[4], dtype=Int64, value=[8 9 6 7])

[HaoyangLee]

Update the docstring of function _update_run_op() in line 44 and 45. (0.0, 1.0) -> [0.0, 1.0]