Huawei modelarts training alexnet model

One 、 Reference material

ModelArts Web Training

Two 、 Key steps

2.1 Upload Source code To obs

.
|-- dataset  #  Data sets 
|   |-- train
|   `-- val `-- train
    |-- data.py
    |-- model.py
    `-- train_npu.py  #  Startup file 

2.2 obs Create directory

obs Directory structure

log： Log storage path 
output： Training output path 

2.3 Create algorithm

 Algorithm management 
	---》 establish 

 name ： Customize 
 describe ： Customize 
 How it was created ： Customize 
	AI engine ：tensorflow_1.15-cann_5.0.3-py_3.7-euler_2.8.3-aarch64
	 Code directory ：/xxx/alexnet/train/
	 Startup file ：/xxx/alexnet/train/train_npu.py
 Input data configuration ： Dataset path 
	 Mapping name ： Customize （ The default can be ）
	 Code path parameters ：data_url
 Output data configuration ： Training output path 
	 Mapping name ： Customize （ The default can be ）
	 Code path parameters ：train_url
 other ： The default can be 

2.4 Create training assignments

 Training management 
	---》 Training assignment 
		---》 establish 

 name ： Customize 
 describe ： Customize 
 Algorithm ：
	---》 My algorithm 
		---》 Check the algorithm just created 
 Training input ：
	---》data_url：/xxx/alexnet/dataset/
 Training output ：
	---》train_url：/xxx/alexnet/output/
 Resource pool ： According to the actual demand 
 The resource type ：Ascend
 specifications ： According to the actual demand 
 Calculate the number of nodes ： According to the actual demand 
 Job log path ：/xxx/alexnet/log/

2.5 Training done

3、 ... and 、FAQ

Q：The input shape of GeOp5_0 is dynamic

【 Nanjing University 】【MEMNET】【ID1085】 Model npu Migration error : The input shape of GeOp5_0 is dynamic

File "/home/ma-user/modelarts/user-job-dir/train/train_npu.py", line 114, in main
    test_loss, test_acc, summary = sess.run([cost, accuracy, summary_op], feed_dict=val_feed)
  File "/home/ma-user/anaconda/lib/python3.7/site-packages/tensorflow_core/python/client/session.py", line 956, in run
    run_metadata_ptr)
  File "/home/ma-user/anaconda/lib/python3.7/site-packages/tensorflow_core/python/client/session.py", line 1180, in _run
    feed_dict_tensor, options, run_metadata)
  File "/home/ma-user/anaconda/lib/python3.7/site-packages/tensorflow_core/python/client/session.py", line 1359, in _do_run
    run_metadata)
  File "/home/ma-user/anaconda/lib/python3.7/site-packages/tensorflow_core/python/client/session.py", line 1384, in _do_call
    raise type(e)(node_def, op, message)
tensorflow.python.framework.errors_impl.InternalError: The input shape of GeOp5_0 is dynamic, please ensure that npu option[dynamic_input] is set correctly, for more details please refer to the migration guide.
	 [[{
    {
    node GeOp5_0}}]]

 The reason for the error ：
 There are inputs during model training shape Problem of change 

 terms of settlement ：
train_npu.py Add 
custom_op.parameter_map["dynamic_input"].b = True
custom_op.parameter_map["dynamic_graph_execute_mode"].s = tf.compat.as_bytes("lazy_recompile")