add_Jittor: Passing model tests, Parameter and Module Container test

add_Jittor: Passing model tests, Parameter and Module Container test.

Additional Functionality:
1- TrainOneStep integration.
2- Updated core/train_jt to enable accuracy to be measured.
3- Updated Jittor Optimizer: replaced gradient and apply_gradient function with jittors default functions Zero_grad() and Step(). included a new function Set() to set the trainable_weights paramters for the optimizer.
4- Updated Jittor Metrics for Accuracy, Recall, Precision and AUC.
5- Creating Jittor model tutorial file jittor_module_tutorial.py
6- Module Container and Parameter Container: Updated core_jittor ModuleList and ParameterDict to enable OrderedDict intialization which was not available due to the parent class (Jittor Module) initializing Dict by default which caused integration issues. This issue was handled by updating the function and also excluding the parent Module for these functions.

Areas to optimize integration: Enabling Jittor integration to run large model training as currently it is limited in the complexity of NN layers.

Author: hishambarakat16

examples/basic_tutorials/cifar10_cnn.py

@@ -2,10 +2,11 @@
 # -*- coding: utf-8 -*-
 
 import os
-os.environ['TL_BACKEND'] = 'paddle'
+# os.environ['TL_BACKEND'] = 'paddle'
+# os.environ['TL_BACKEND'] = 'jittor'
 # os.environ['TL_BACKEND'] = 'tensorflow'
 # os.environ['TL_BACKEND'] = 'mindspore'
-# os.environ['TL_BACKEND'] = 'torch'
+os.environ['TL_BACKEND'] = 'torch'
 
 import paddle
 from paddle.distributed import fleet

examples/basic_tutorials/cifar10_cnn_dist.py

@@ -5,12 +5,11 @@
 
 import os
 # os.environ['TL_BACKEND'] = 'paddle'
-
-os.environ['TL_BACKEND'] = 'jittor'
 # os.environ['TL_BACKEND'] = 'tensorflow'
 # os.environ['TL_BACKEND'] = 'mindspore'
+# os.environ['TL_BACKEND'] = 'jittor'
 
-# os.environ['TL_BACKEND'] = 'torch'
+os.environ['TL_BACKEND'] = 'torch'
 
 
 
@@ -76,7 +75,7 @@ def forward(self, x):
 
 # 定义损失函数、优化器等
 loss_fn=tlx.losses.softmax_cross_entropy_with_logits
-optimizer = tlx.optimizers.Adam(net.trainable_weights, lr=learning_rate)
+optimizer = tlx.optimizers.Adam(learning_rate)
 metrics = tlx.metrics.Accuracy()
 
 

examples/basic_tutorials/cifar10_cnn_train.py

@@ -2,9 +2,10 @@
 # -*- coding: utf-8 -*-
 # The tensorlayerx and tensorflow operators can be mixed
 import os
-os.environ['TL_BACKEND'] = 'tensorflow'
+# os.environ['TL_BACKEND'] = 'tensorflow'
 # os.environ['TL_BACKEND'] = 'paddle'
-# os.environ['TL_BACKEND'] = 'torch'
+os.environ['TL_BACKEND'] = 'torch'
+# os.environ['TL_BACKEND'] = 'jittor'
 
 
 import time