!7372 [quant] add QuantDtype and Observer

Merge pull request !7372 from yuchaojie/quant2

cmake/package.cmake

@@ -263,6 +263,7 @@ install(
         ${CMAKE_SOURCE_DIR}/mindspore/ops
         ${CMAKE_SOURCE_DIR}/mindspore/communication
         ${CMAKE_SOURCE_DIR}/mindspore/profiler
+        ${CMAKE_SOURCE_DIR}/mindspore/compression
     DESTINATION ${INSTALL_PY_DIR}
     COMPONENT mindspore
 )

mindspore/compression/__init__.py

@@ -0,0 +1,17 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+"""
+MindSpore compression module.
+"""

mindspore/compression/common/__init__.py

@@ -0,0 +1,19 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+"""
+Compression common module.
+"""
+
+from .constant import *

mindspore/compression/common/constant.py

@@ -0,0 +1,85 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+"""Constant module for compression"""
+import enum
+import re
+from types import DynamicClassAttribute
+
+
+__all__ = ["QuantDtype"]
+
+
+@enum.unique
+class QuantDtype(enum.Enum):
+    """
+    For type switch
+    """
+    INT2 = "INT2"
+    INT3 = "INT3"
+    INT4 = "INT4"
+    INT5 = "INT5"
+    INT6 = "INT6"
+    INT7 = "INT7"
+    INT8 = "INT8"
+
+    UINT2 = "UINT2"
+    UINT3 = "UINT3"
+    UINT4 = "UINT4"
+    UINT5 = "UINT5"
+    UINT6 = "UINT6"
+    UINT7 = "UINT7"
+    UINT8 = "UINT8"
+
+    FLOAT16 = "FLOAT16"
+    FLOAT32 = "FLOAT32"
+
+    def __str__(self):
+        return f"{self.name}"
+
+    @staticmethod
+    def is_signed(dtype):
+        return dtype in [QuantDtype.INT2, QuantDtype.INT3, QuantDtype.INT4, QuantDtype.INT5,
+                         QuantDtype.INT6, QuantDtype.INT7, QuantDtype.INT8]
+
+    @staticmethod
+    def switch_signed(dtype):
+        """switch signed"""
+        type_map = {
+            QuantDtype.INT2: QuantDtype.UINT2,
+            QuantDtype.INT3: QuantDtype.UINT3,
+            QuantDtype.INT4: QuantDtype.UINT4,
+            QuantDtype.INT5: QuantDtype.UINT5,
+            QuantDtype.INT6: QuantDtype.UINT6,
+            QuantDtype.INT7: QuantDtype.UINT7,
+            QuantDtype.INT8: QuantDtype.UINT8,
+            QuantDtype.UINT2: QuantDtype.INT2,
+            QuantDtype.UINT3: QuantDtype.INT3,
+            QuantDtype.UINT4: QuantDtype.INT4,
+            QuantDtype.UINT5: QuantDtype.INT5,
+            QuantDtype.UINT6: QuantDtype.INT6,
+            QuantDtype.UINT7: QuantDtype.INT7,
+            QuantDtype.UINT8: QuantDtype.INT8
+        }
+        return type_map[dtype]
+
+    @DynamicClassAttribute
+    def value(self):
+        """The value of the Enum member."""
+        return int(re.search(r"(\d+)", self._value_).group(1))
+
+    @DynamicClassAttribute
+    def num_bits(self):
+        """The num_bits of the Enum member."""
+        return self.value

mindspore/nn/layer/quant.py

@@ -24,6 +24,7 @@ from mindspore.common.parameter import Parameter
 from mindspore.common.initializer import initializer
 from mindspore.common.tensor import Tensor
 from mindspore._checkparam import Validator, Rel, twice
+from mindspore.compression.common import QuantDtype
 import mindspore.context as context
 from .normalization import BatchNorm2d, BatchNorm1d
 from .activation import get_activation, ReLU, LeakyReLU
@@ -277,13 +278,233 @@ class BatchNormFoldCell(Cell):
         return batch_mean, batch_std, running_mean, running_std
 
 
+def _partial_init(cls_or_self, **kwargs):
+    """
+    Wrapper that allows creation of class factories.
+
+    This can be useful when there is a need to create classes with the same
+    constructor arguments, but different instances.
+
+    Example::
+        >>> Foo.partial_init = classmethod(_partial_init)
+        >>> foo_builder = Foo.partial_init(a=3, b=4).partial_init(answer=42)
+        >>> foo_instance1 = foo_builder()
+        >>> foo_instance2 = foo_builder()
+        >>> id(foo_instance1) == id(foo_instance2)
+        False
+    """
+
+    class _PartialWrapper:
+        r"""
+        class of wrapper that allows creation of class factories.
+        """
+
+        def __init__(self, p):
+            self.p = p
+
+        def __call__(self, *args, **keywords):
+            return self.p(*args, **keywords)
+
+        def __repr__(self):
+            return self.p.__repr__()
+
+        partial_init = _partial_init
+
+    r = _PartialWrapper(partial(cls_or_self, **kwargs))
+    return r
+
+
+class Observer(Cell):
+    """
+    Base class of Observer. Observer is used to calculate the statistics of specific layer.
+
+    Notes:
+        This class is an abstract class.
+
+    Args:
+        quant_dtype (QuantDtype): The type of FakeQuant data.
+    """
+
+    def __init__(self, quant_dtype):
+        super(Observer, self).__init__()
+        self.quant_dtype = quant_dtype
+
+    def extend_repr(self):
+        s = f"dtype={self.dtype}"
+        return s
+
+    def construct(self):
+        pass
+
+    partial_init = classmethod(_partial_init)
+
+
+class UniformQuantObserver(Observer):
+    """
+    The base class of Uniform Quantization Observer.
+
+    Args:
+        quant_dtype (QuantDtype): The type of FakeQuant data. Default: QuantDtype.INT8.
+        per_channel (bool):  Quantization granularity based on layer or on channel. Default: False.
+        symmetric (bool): Whether the quantization algorithm is symmetric or not. Default: False.
+        narrow_range (bool): Whether the quantization algorithm uses narrow range or not. Default: False.
+        num_channels (int): declarate the min and max channel size, Default: 1.
+
+    Returns:
+        Tensor.
+    """
+
+    min_max_map = {
+        QuantDtype.INT2: (-2, 1),
+        QuantDtype.INT3: (-4, 3),
+        QuantDtype.INT4: (-8, 7),
+        QuantDtype.INT5: (-16, 15),
+        QuantDtype.INT6: (-32, 31),
+        QuantDtype.INT7: (-64, 63),
+        QuantDtype.INT8: (-128, 127),
+
+        QuantDtype.UINT2: (0, 3),
+        QuantDtype.UINT3: (0, 7),
+        QuantDtype.UINT4: (0, 15),
+        QuantDtype.UINT5: (0, 31),
+        QuantDtype.UINT6: (0, 63),
+        QuantDtype.UINT7: (0, 127),
+        QuantDtype.UINT8: (0, 255)
+    }
+
+    def __init__(self, quant_dtype=QuantDtype.INT8, per_channel=False, symmetric=False, narrow_range=False,
+                 num_channels=1):
+        super(UniformQuantObserver, self).__init__(quant_dtype)
+        self.per_channel = per_channel
+        self.symmetric = symmetric
+        self.narrow_range = narrow_range
+        self.num_channels = num_channels
+
+
+class FakeQuantWithMinMaxObserver(UniformQuantObserver):
+    r"""
+    Quantization aware op. This OP provides the fake quantization observer function on data with min and max.
+
+    Args:
+        min_init (int, float): The initialized min value. Default: -6.
+        max_init (int, float): The initialized max value. Default: 6.
+        ema (bool): The exponential Moving Average algorithm updates min and max. Default: False.
+        ema_decay (float): Exponential Moving Average algorithm parameter. Default: 0.999.
+        per_channel (bool):  Quantization granularity based on layer or on channel. Default: False.
+        channel_axis (int): Quantization by channel axis. Default: 1.
+        num_channels (int): declarate the min and max channel size, Default: 1.
+        quant_dtype (QuantDtype): The datatype of quantization, supporting 4 and 8bits. Default: QuantDtype.INT8.
+        symmetric (bool): Whether the quantization algorithm is symmetric or not. Default: False.
+        narrow_range (bool): Whether the quantization algorithm uses narrow range or not. Default: False.
+        quant_delay (int): Quantization delay parameters according to the global step. Default: 0.
+
+    Inputs:
+        - **x** (Tensor) - The input of FakeQuantWithMinMaxObserver.
+
+    Outputs:
+        Tensor, with the same type and shape as the `x`.
+
+    Examples:
+        >>> fake_quant = FakeQuantWithMinMaxObserver()
+        >>> input_x = Tensor(np.array([[1, 2, 1], [-2, 0, -1]]), mindspore.float32)
+        >>> result = fake_quant(input_x)
+    """
+
+    def __init__(self,
+                 min_init=-6,
+                 max_init=6,
+                 ema=False,
+                 ema_decay=0.999,
+                 per_channel=False,
+                 channel_axis=1,
+                 num_channels=1,
+                 quant_dtype=QuantDtype.INT8,
+                 symmetric=False,
+                 narrow_range=False,
+                 quant_delay=0):
+        """Initialize FakeQuantWithMinMax layer"""
+        super(FakeQuantWithMinMaxObserver, self).__init__(quant_dtype=quant_dtype, per_channel=per_channel,
+                                                          symmetric=symmetric, narrow_range=narrow_range,
+                                                          num_channels=num_channels)
+        Validator.check_type("min_init", min_init, [int, float])
+        Validator.check_type("max_init", max_init, [int, float])
+        Validator.check("min_init", min_init, "max_init", max_init, rel=Rel.LT)
+        Validator.check_integer('quant_delay', quant_delay, 0, Rel.GE)
+        self.min_init = min_init
+        self.max_init = max_init
+        self.quant_dtype = quant_dtype
+        self.ema = ema
+        self.ema_decay = ema_decay
+        self.per_channel = per_channel
+        self.num_channels = num_channels
+        self.channel_axis = channel_axis
+        self.quant_delay = quant_delay
+        self.symmetric = symmetric
+        self.narrow_range = narrow_range
+        self.is_ascend = context.get_context('device_target') == "Ascend"
+
+        # init tensor min and max for fake quant op
+        if self.per_channel:
+            min_array = np.array([self.min_init] * self.num_channels).astype(np.float32)
+            max_array = np.array([self.max_init] * self.num_channels).astype(np.float32)
+        else:
+            min_array = np.array([self.min_init]).astype(np.float32)
+            max_array = np.array([self.max_init]).astype(np.float32)
+        self.minq = Parameter(Tensor(min_array), name='quant_min', requires_grad=False)
+        self.maxq = Parameter(Tensor(max_array), name='quant_max', requires_grad=False)
+
+        # init fake quant relative op
+        if self.per_channel:
+            quant_fun = partial(Q.FakeQuantPerChannel, channel_axis=self.channel_axis)
+            ema_fun = partial(Q.MinMaxUpdatePerChannel, channel_axis=self.channel_axis)
+        else:
+            quant_fun = Q.FakeQuantPerLayer
+            ema_fun = Q.MinMaxUpdatePerLayer
+
+        self.ema_update = ema_fun(ema=self.ema, ema_decay=self.ema_decay)
+        if self.is_ascend:
+            self.fake_quant_train = quant_fun(num_bits=self.quant_dtype.num_bits,
+                                              symmetric=self.symmetric,
+                                              narrow_range=self.narrow_range,
+                                              quant_delay=self.quant_delay)
+            self.fake_quant_infer = self.fake_quant_train
+        else:
+            quant_fun = partial(quant_fun,
+                                ema=self.ema,
+                                ema_decay=ema_decay,
+                                num_bits=self.quant_dtype.num_bits,
+                                symmetric=self.symmetric,
+                                narrow_range=self.narrow_range,
+                                quant_delay=self.quant_delay)
+            self.fake_quant_train = quant_fun(training=True)
+            self.fake_quant_infer = quant_fun(training=False)
+
+    def extend_repr(self):
+        s = 'quant_dtype={}, symmetric={}, narrow_range={}, ema={}({}), per_channel={}({}, {}), ' \
+            'quant_delay={}, min_init={}, max_init={}'.format(self.quant_dtype, self.symmetric, self.narrow_range,
+                                                              self.ema, self.ema_decay, self.per_channel,
+                                                              self.channel_axis, self.num_channels, self.quant_delay,
+                                                              self.min_init, self.max_init)
+        return s
+
+    def construct(self, x):
+        if self.training:
+            min_up, max_up = self.ema_update(x, self.minq, self.maxq)
+            P.Assign()(self.minq, min_up)
+            P.Assign()(self.maxq, max_up)
+            out = self.fake_quant_train(x, self.minq, self.maxq)
+        else:
+            out = self.fake_quant_infer(x, self.minq, self.maxq)
+        return out
+
+
 class FakeQuantWithMinMax(Cell):
     r"""
     Quantization aware op. This OP provides the fake quantization observer function on data with min and max.
 
     Args:
-        min_init (int, float): The dimension of channel or 1(layer). Default: -6.
-        max_init (int, float): The dimension of channel or 1(layer). Default: 6.
+        min_init (int, float): The initialized min value. Default: -6.
+        max_init (int, float): The initialized max value. Default: 6.
         ema (bool): The exponential Moving Average algorithm updates min and max. Default: False.
         ema_decay (float): Exponential Moving Average algorithm parameter. Default: 0.999.
         per_channel (bool):  Quantization granularity based on layer or on channel. Default: False.