updata 2.0 API english doc (#28525)

* make Numpy version is below 1.19.3

* fix 2.0 doc

paddle/fluid/operators/unique_op.cu

@@ -177,7 +177,7 @@ static void UniqueFlattendCUDATensor(const framework::ExecutionContext& context,
   thrust::sort_by_key(thrust::device, in_data_hat, in_data_hat + num_input,
                       sorted_indices_data);
 
-  // 1. Calculate op result: 'out'：
+  // 1. Calculate op result: 'out'
   Tensor range;
   range.Resize(framework::make_ddim({num_input + 1}));
   auto range_data_ptr = range.mutable_data<IndexT>(context.GetPlace());

paddle/fluid/pybind/imperative.cc

@@ -685,8 +685,6 @@ void BindImperative(py::module *m_ptr) {
             .. code-block:: python
 
                 import paddle
-                paddle.disable_static()
-
                 linear = Linear(32, 64)
                 data = paddle.uniform(shape=[30, 10, 32], -1, 1)
                 x = linear(data)
@@ -704,19 +702,13 @@ void BindImperative(py::module *m_ptr) {
              .. code-block:: python
 
                 import paddle
-                paddle.disable_static()
-
-                inputs = []
-                for _ in range(10):
-                    tmp = paddle.ones([2, 2])
-                    tmp.stop_gradient=False
-                    inputs.append(tmp)
-                ret = paddle.sums(inputs2)
-                loss = paddle.sum(ret)
-                loss.backward()
-                print("Before clear_gradient {}".format(loss.grad))
-                loss.clear_gradient()
-                print("After clear_gradient {}".format(loss.grad))
+                input = paddle.uniform([10, 2])
+                linear = paddle.nn.Linear(2, 3)
+                out = linear(input)
+                out.backward()
+                print("Before clear_gradient, linear.weight.grad: {}".format(linear.weight.grad))
+                linear.weight.clear_gradient()
+                print("After clear_gradient, linear.weight.grad: {}".format(linear.weight.grad))
       )DOC")
       .def("clone",
            [](std::shared_ptr<imperative::VarBase> &self) {

python/paddle/framework/__init__.py

@@ -18,10 +18,7 @@ __all__ = [
     'get_default_dtype', 'set_default_dtype'
 ]
 
-__all__ += [
-    'grad', 'LayerList', 'load', 'save', 'to_variable', 'no_grad',
-    'DataParallel'
-]
+__all__ += ['grad', 'LayerList', 'load', 'save', 'no_grad', 'DataParallel']
 
 from . import random
 from .random import seed
@@ -39,7 +36,6 @@ from ..fluid.core import VarBase  #DEFINE_ALIAS
 
 from paddle.fluid import core  #DEFINE_ALIAS
 from ..fluid.dygraph.base import no_grad_ as no_grad  #DEFINE_ALIAS
-from ..fluid.dygraph.base import to_variable  #DEFINE_ALIAS
 from ..fluid.dygraph.base import grad  #DEFINE_ALIAS
 from .io import save
 from .io import load

python/paddle/framework/io.py

@@ -225,8 +225,6 @@ def save(obj, path):
 
             import paddle
 
-            paddle.disable_static()
-
             emb = paddle.nn.Embedding(10, 10)
             layer_state_dict = emb.state_dict()
             paddle.save(layer_state_dict, "emb.pdparams")
@@ -319,8 +317,6 @@ def load(path, **configs):
 
             import paddle
 
-            paddle.disable_static()
-
             emb = paddle.nn.Embedding(10, 10)
             layer_state_dict = emb.state_dict()
             paddle.save(layer_state_dict, "emb.pdparams")

python/paddle/optimizer/lr.py

@@ -226,14 +226,15 @@ class NoamDecay(LRScheduler):
             scheduler = paddle.optimizer.lr.NoamDecay(d_model=0.01, warmup_steps=100, verbose=True)
             sgd = paddle.optimizer.SGD(learning_rate=scheduler, parameters=linear.parameters())
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     x = paddle.uniform([10, 10])
                     out = linear(x)
                     loss = paddle.mean(out)
                     loss.backward()
                     sgd.step()
                     sgd.clear_gradients()
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
 
             # train on static graph mode
             paddle.enable_static()
@@ -251,7 +252,7 @@ class NoamDecay(LRScheduler):
             exe = paddle.static.Executor()
             exe.run(start_prog)
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     out = exe.run(
                         main_prog,
                         feed={
@@ -259,7 +260,8 @@ class NoamDecay(LRScheduler):
                             'y': np.random.randn(3, 4, 5).astype('float32')
                         },
                         fetch_list=loss.name)
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
 
     """
 
@@ -322,14 +324,15 @@ class PiecewiseDecay(LRScheduler):
             scheduler = paddle.optimizer.lr.PiecewiseDecay(boundaries=[3, 6, 9], values=[0.1, 0.2, 0.3, 0.4], verbose=True)
             sgd = paddle.optimizer.SGD(learning_rate=scheduler, parameters=linear.parameters())
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     x = paddle.uniform([10, 10])
                     out = linear(x)
                     loss = paddle.mean(out)
                     loss.backward()
                     sgd.step()
                     sgd.clear_gradients()
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
 
             # train on static graph mode
             paddle.enable_static()
@@ -347,7 +350,7 @@ class PiecewiseDecay(LRScheduler):
             exe = paddle.static.Executor()
             exe.run(start_prog)
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     out = exe.run(
                         main_prog,
                         feed={
@@ -355,7 +358,8 @@ class PiecewiseDecay(LRScheduler):
                             'y': np.random.randn(3, 4, 5).astype('float32')
                         },
                         fetch_list=loss.name)
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
     """
 
     def __init__(self, boundaries, values, last_epoch=-1, verbose=False):
@@ -403,14 +407,15 @@ class NaturalExpDecay(LRScheduler):
             scheduler = paddle.optimizer.lr.NaturalExpDecay(learning_rate=0.5, gamma=0.1, verbose=True)
             sgd = paddle.optimizer.SGD(learning_rate=scheduler, parameters=linear.parameters())
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     x = paddle.uniform([10, 10])
                     out = linear(x)
                     loss = paddle.mean(out)
                     loss.backward()
                     sgd.step()
                     sgd.clear_gradients()
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
 
             # train on static graph mode
             paddle.enable_static()
@@ -428,7 +433,7 @@ class NaturalExpDecay(LRScheduler):
             exe = paddle.static.Executor()
             exe.run(start_prog)
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     out = exe.run(
                         main_prog,
                         feed={
@@ -436,7 +441,8 @@ class NaturalExpDecay(LRScheduler):
                             'y': np.random.randn(3, 4, 5).astype('float32')
                         },
                         fetch_list=loss.name)
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
     """
 
     def __init__(self, learning_rate, gamma, last_epoch=-1, verbose=False):
@@ -481,14 +487,15 @@ class InverseTimeDecay(LRScheduler):
             scheduler = paddle.optimizer.lr.InverseTimeDecay(learning_rate=0.5, gamma=0.1, verbose=True)
             sgd = paddle.optimizer.SGD(learning_rate=scheduler, parameters=linear.parameters())
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     x = paddle.uniform([10, 10])
                     out = linear(x)
                     loss = paddle.mean(out)
                     loss.backward()
                     sgd.step()
                     sgd.clear_gradients()
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
 
             # train on static graph mode
             paddle.enable_static()
@@ -506,7 +513,7 @@ class InverseTimeDecay(LRScheduler):
             exe = paddle.static.Executor()
             exe.run(start_prog)
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     out = exe.run(
                         main_prog,
                         feed={
@@ -514,7 +521,8 @@ class InverseTimeDecay(LRScheduler):
                             'y': np.random.randn(3, 4, 5).astype('float32')
                         },
                         fetch_list=loss.name)
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
 
     """
 
@@ -576,14 +584,15 @@ class PolynomialDecay(LRScheduler):
             scheduler = paddle.optimizer.lr.PolynomialDecay(learning_rate=0.5, decay_steps=20, verbose=True)
             sgd = paddle.optimizer.SGD(learning_rate=scheduler, parameters=linear.parameters())
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     x = paddle.uniform([10, 10])
                     out = linear(x)
                     loss = paddle.mean(out)
                     loss.backward()
                     sgd.step()
                     sgd.clear_gradients()
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
 
             # train on static graph mode
             paddle.enable_static()
@@ -601,7 +610,7 @@ class PolynomialDecay(LRScheduler):
             exe = paddle.static.Executor()
             exe.run(start_prog)
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     out = exe.run(
                         main_prog,
                         feed={
@@ -609,7 +618,8 @@ class PolynomialDecay(LRScheduler):
                             'y': np.random.randn(3, 4, 5).astype('float32')
                         },
                         fetch_list=loss.name)
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
     """
 
     def __init__(self,
@@ -691,14 +701,15 @@ class LinearWarmup(LRScheduler):
                     learning_rate=0.5, warmup_steps=20, start_lr=0, end_lr=0.5, verbose=True)
             sgd = paddle.optimizer.SGD(learning_rate=scheduler, parameters=linear.parameters())
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     x = paddle.uniform([10, 10])
                     out = linear(x)
                     loss = paddle.mean(out)
                     loss.backward()
                     sgd.step()
                     sgd.clear_gradients()
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
 
             # train on static graph mode
             paddle.enable_static()
@@ -717,7 +728,7 @@ class LinearWarmup(LRScheduler):
             exe = paddle.static.Executor()
             exe.run(start_prog)
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     out = exe.run(
                         main_prog,
                         feed={
@@ -725,7 +736,8 @@ class LinearWarmup(LRScheduler):
                             'y': np.random.randn(3, 4, 5).astype('float32')
                         },
                         fetch_list=loss.name)
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
     """
 
     def __init__(self,
@@ -814,14 +826,15 @@ class ExponentialDecay(LRScheduler):
             scheduler = paddle.optimizer.lr.ExponentialDecay(learning_rate=0.5, gamma=0.9, verbose=True)
             sgd = paddle.optimizer.SGD(learning_rate=scheduler, parameters=linear.parameters())
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     x = paddle.uniform([10, 10])
                     out = linear(x)
                     loss = paddle.mean(out)
                     loss.backward()
                     sgd.step()
                     sgd.clear_gradients()
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
 
             # train on static graph mode
             paddle.enable_static()
@@ -839,7 +852,7 @@ class ExponentialDecay(LRScheduler):
             exe = paddle.static.Executor()
             exe.run(start_prog)
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     out = exe.run(
                         main_prog,
                         feed={
@@ -847,7 +860,8 @@ class ExponentialDecay(LRScheduler):
                             'y': np.random.randn(3, 4, 5).astype('float32')
                         },
                         fetch_list=loss.name)
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
     """
 
     def __init__(self, learning_rate, gamma, last_epoch=-1, verbose=False):
@@ -901,14 +915,15 @@ class MultiStepDecay(LRScheduler):
             scheduler = paddle.optimizer.lr.MultiStepDecay(learning_rate=0.5, milestones=[2, 4, 6], gamma=0.8, verbose=True)
             sgd = paddle.optimizer.SGD(learning_rate=scheduler, parameters=linear.parameters())
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     x = paddle.uniform([10, 10])
                     out = linear(x)
                     loss = paddle.mean(out)
                     loss.backward()
                     sgd.step()
                     sgd.clear_gradients()
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
 
             # train on static graph mode
             paddle.enable_static()
@@ -926,7 +941,7 @@ class MultiStepDecay(LRScheduler):
             exe = paddle.static.Executor()
             exe.run(start_prog)
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     out = exe.run(
                         main_prog,
                         feed={
@@ -934,7 +949,8 @@ class MultiStepDecay(LRScheduler):
                             'y': np.random.randn(3, 4, 5).astype('float32')
                         },
                         fetch_list=loss.name)
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
     """
 
     def __init__(self,
@@ -1008,14 +1024,15 @@ class StepDecay(LRScheduler):
             scheduler = paddle.optimizer.lr.StepDecay(learning_rate=0.5, step_size=5, gamma=0.8, verbose=True)
             sgd = paddle.optimizer.SGD(learning_rate=scheduler, parameters=linear.parameters())
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     x = paddle.uniform([10, 10])
                     out = linear(x)
                     loss = paddle.mean(out)
                     loss.backward()
                     sgd.step()
                     sgd.clear_gradients()
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
 
             # train on static graph mode
             paddle.enable_static()
@@ -1033,7 +1050,7 @@ class StepDecay(LRScheduler):
             exe = paddle.static.Executor()
             exe.run(start_prog)
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     out = exe.run(
                         main_prog,
                         feed={
@@ -1041,7 +1058,8 @@ class StepDecay(LRScheduler):
                             'y': np.random.randn(3, 4, 5).astype('float32')
                         },
                         fetch_list=loss.name)
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
     """
 
     def __init__(self,
@@ -1102,14 +1120,15 @@ class LambdaDecay(LRScheduler):
             scheduler = paddle.optimizer.lr.LambdaDecay(learning_rate=0.5, lr_lambda=lambda x:0.95**x, verbose=True)
             sgd = paddle.optimizer.SGD(learning_rate=scheduler, parameters=linear.parameters())
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     x = paddle.uniform([10, 10])
                     out = linear(x)
                     loss = paddle.mean(out)
                     loss.backward()
                     sgd.step()
                     sgd.clear_gradients()
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
 
             # train on static graph mode
             paddle.enable_static()
@@ -1127,7 +1146,7 @@ class LambdaDecay(LRScheduler):
             exe = paddle.static.Executor()
             exe.run(start_prog)
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     out = exe.run(
                         main_prog,
                         feed={
@@ -1135,7 +1154,8 @@ class LambdaDecay(LRScheduler):
                             'y': np.random.randn(3, 4, 5).astype('float32')
                         },
                         fetch_list=loss.name)
-                scheduler.step()
+                    scheduler.step()    # If you update learning rate each step
+              # scheduler.step()        # If you update learning rate each epoch
 
     """
 
@@ -1200,14 +1220,15 @@ class ReduceOnPlateau(LRScheduler):
             scheduler = paddle.optimizer.lr.ReduceOnPlateau(learning_rate=1.0, factor=0.5, patience=5, verbose=True)
             sgd = paddle.optimizer.SGD(learning_rate=scheduler, parameters=linear.parameters())
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     x = paddle.uniform([10, 10])
                     out = linear(x)
                     loss = paddle.mean(out)
                     loss.backward()
                     sgd.step()
                     sgd.clear_gradients()
-                scheduler.step(loss)
+                    scheduler.step(loss)    # If you update learning rate each step
+              # scheduler.step(loss)        # If you update learning rate each epoch
 
             # train on static graph mode
             paddle.enable_static()
@@ -1225,7 +1246,7 @@ class ReduceOnPlateau(LRScheduler):
             exe = paddle.static.Executor()
             exe.run(start_prog)
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     out = exe.run(
                         main_prog,
                         feed={
@@ -1233,7 +1254,8 @@ class ReduceOnPlateau(LRScheduler):
                             'y': np.random.randn(3, 4, 5).astype('float32')
                         },
                         fetch_list=loss.name)
-                scheduler.step(out[0])
+                    scheduler.step(out[0])    # If you update learning rate each step
+              # scheduler.step(out[0])        # If you update learning rate each epoch
 
     """
 
@@ -1268,7 +1290,6 @@ class ReduceOnPlateau(LRScheduler):
                 "The type of 'learning_rate' in 'ReduceOnPlateau' must be 'float', but received %s."
                 % type(learning_rate))
 
-        self.verbose = verbose
         self.patience = patience
         self.threshold = threshold
         self.threshold_mode = threshold_mode
@@ -1406,7 +1427,7 @@ class CosineAnnealingDecay(LRScheduler):
             scheduler = paddle.optimizer.lr.CosineAnnealingDecay(learning_rate=0.5, T_max=10, verbose=True)
             sgd = paddle.optimizer.SGD(learning_rate=scheduler, parameters=linear.parameters())
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     x = paddle.uniform([10, 10])
                     out = linear(x)
                     loss = paddle.mean(out)
@@ -1431,7 +1452,7 @@ class CosineAnnealingDecay(LRScheduler):
             exe = paddle.static.Executor()
             exe.run(start_prog)
             for epoch in range(20):
-                for batch_id in range(2):
+                for batch_id in range(5):
                     out = exe.run(
                         main_prog,
                         feed={