Updated call method in base layer

Added a description about enabling eager mode for debugging purpose.

tensorflow/python/keras/engine/base_layer.py

@@ -477,13 +477,20 @@ class Layer(module.Module, version_utils.LayerVersionSelector):
 
   @doc_controls.for_subclass_implementers
   def call(self, inputs, *args, **kwargs):  # pylint: disable=unused-argument
-    """This is where the layer's logic lives.
+    """This is where the layer's forward pass is defined and logic lives.
+
+    By default, `layer.call()` run in graph mode for higher 
+    performance. In order to run the `layer.call()` body in eager context, 
+    you need to invoke `model.compile(run_eagerly=True)`. Note that this 
+    will have performance slow down due to eager runtime, and should only 
+    be used for debug purpose.
     
     Note here that `call()` method in `tf.keras` is little bit different
     from `keras` API. In `keras` API, you can pass support masking for
     layers as additional arguments. Whereas `tf.keras` has `compute_mask()`
     method to support masking.
     
+
     Args:
       inputs: Input tensor, or dict/list/tuple of input tensors.
         The first positional `inputs` argument is subject to special rules: