Add channel-wise weights observers (#1697)

paddleslim/quant/observers/__init__.py

@@ -17,7 +17,16 @@ from .kl import KLObserver
 from .mse import MSEObserver
 from .emd import EMDObserver
 from .avg import AVGObserver
+from .mse_weight import MSEChannelWiseWeightObserver
+from .abs_max_weight import AbsMaxChannelWiseWeightObserver
 
 __all__ = [
-    "HistObserver", "KLObserver", "MSEObserver", "EMDObserver", "AVGObserver"
+    "HistObserver",
+    "KLObserver",
+    "MSEObserver",
+    "EMDObserver",
+    "AVGObserver",
+    "MSEWeightObserver",
+    "MSEChannelWiseWeightObserver",
+    "AbsMaxChannelWiseWeightObserver",
 ]

paddleslim/quant/observers/abs_max_weight.py

+# Copyright (c) 2023 PaddlePaddle Authors. All Rights Reserved.
+#
+# 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.
+
+import numpy as np
+import paddle
+from .channel_wise import ChannelWiseObserver
+from paddle.quantization.factory import ObserverFactory
+
+
+class AbsMaxChannelWiseWeightObserver(ObserverFactory):
+    r"""
+    It collects channel-wise maximum absolute values of target weights.
+    Args:
+        bit_length(int, optional): Number of bits to represent an quantized integer in binary.
+        dtype(str, optional): The data type of input tensor.
+        name (str, optional): This parameter is used by developers to print debugging information. \
+            For details, please refer to :ref:`api_guide_Name`. Default is None.
+    Examples:
+       .. code-block:: python
+            from paddle.quantization import QuantConfig
+            from paddle.quantization.quanters import AbsMaxChannelWiseWeightObserver
+            quanter = AbsMaxChannelWiseWeightObserver()
+            q_config = QuantConfig(activation=None, weight=quanter)
+    """
+
+    def __init__(self, quant_bits=8):
+        super(AbsMaxChannelWiseWeightObserver, self).__init__(
+            quant_bits=quant_bits)
+
+    def _get_class(self):
+        return AbsMaxChannelWiseWeightObserverLayer
+
+
+class AbsMaxChannelWiseWeightObserverLayer(ChannelWiseObserver):
+    def __init__(self, layer, quant_bits=8):
+        super(AbsMaxChannelWiseWeightObserverLayer, self).__init__(
+            layer,
+            quant_bits=quant_bits,
+            sign=True,
+            symmetric=True, )
+        self.quant_bits = quant_bits
+        self.calibration_loss = float('inf')
+        self.qmin, self.qmax = self.qmin_qmax
+        self._max = None
+        self._scale = None
+        self._zero_point = None
+
+    def forward(self, inputs):
+        if self._max is None:
+            self._max = self._cal_abs_max(inputs)
+        return inputs
+
+    def _cal_abs_max(self, inputs):
+        reduce_axis = tuple(
+            [i for i in range(len(inputs.shape)) if i != self.quant_axis()])
+        abs_max_values = paddle.max(paddle.abs(inputs), axis=reduce_axis)
+        abs_max_values = paddle.where(abs_max_values == np.float32(0.0),
+                                      np.float32(1e-8), abs_max_values)
+        minimum_loss = paddle.full(abs_max_values.shape, float('inf'))
+        result = abs_max_values
+        factor = 0.3
+        while factor <= 1.0:
+            scales = factor * abs_max_values
+            factor += 0.02
+            expand_scales = paddle.unsqueeze(scales, axis=reduce_axis)
+            quant_var = paddle.clip(
+                paddle.round(inputs / expand_scales * self.qmax), self.qmin,
+                self.qmax)
+            quant_dequant_var = quant_var / self.qmax * expand_scales
+
+            mse_loss = ((inputs - quant_dequant_var)**2).mean(axis=reduce_axis)
+            result = paddle.where(mse_loss < minimum_loss, scales, result)
+            minimum_loss = paddle.minimum(mse_loss, minimum_loss)
+
+        return result
+
+    def min_value(self) -> float:
+        return 0.
+
+    def max_value(self) -> float:
+        return self._max
+
+    def cal_thresholds(self):
+        """ Compute thresholds for MAX function.
+        """
+        self._scale = self._max
+        self._zero_point = paddle.zeros_like(self._scale)
+
+    def scales(self):
+        """ Return output scales.
+        """
+        if self._scale is None:
+            self.cal_thresholds()
+        return self._scale
+
+    def zero_points(self):
+        """ Return output zero points.
+        """
+        if self._zero_point is None:
+            self.cal_thresholds()
+        return self._zero_point

paddleslim/quant/observers/channel_wise.py

+# Copyright (c) 2023 PaddlePaddle Authors. All Rights Reserved.
+#
+# 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.
+
+from typing import Dict
+import numpy as np
+import paddle
+from .mse import MSEObserverLayer
+from .uniform import UniformObserver
+from paddle.quantization.factory import ObserverFactory
+
+CHANNEL_AXIS: Dict[type, int] = {paddle.nn.Conv2D: 0, paddle.nn.Linear: 1}
+
+
+class ChannelWiseObserver(UniformObserver):
+    def __init__(
+            self,
+            layer,
+            quant_bits=8,
+            sign=True,
+            symmetric=True, ):
+        super(ChannelWiseObserver, self).__init__(
+            quant_bits=quant_bits,
+            sign=sign,
+            symmetric=symmetric, )
+        self._channel_axis = CHANNEL_AXIS[type(layer)]
+        self._quant_bits = quant_bits
+
+    def quant_axis(self):
+        """ Return quantization axis.
+        """
+        return self._channel_axis
+
+    def bit_length(self):
+        """ Return the bit length of quantized data.
+        """
+        return self._quant_bits

paddleslim/quant/observers/mse_weight.py

+# Copyright (c) 2023 PaddlePaddle Authors. All Rights Reserved.
+#
+# 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.
+
+import numpy as np
+import paddle
+from paddle.quantization.factory import ObserverFactory
+from .abs_max_weight import AbsMaxChannelWiseWeightObserverLayer
+
+
+class MSEChannelWiseWeightObserver(ObserverFactory):
+    r"""
+    It collects channel-wise maximum absolute values and calculates the quantization scales by minimizing
+    the quantization MSE error.
+    Args:
+        bit_length(int, optional): Number of bits to represent an quantized integer in binary.
+        dtype(str, optional): The data type of input tensor.
+        name (str, optional): This parameter is used by developers to print debugging information. \
+            For details, please refer to :ref:`api_guide_Name`. Default is None.
+    Examples:
+       .. code-block:: python
+            from paddle.quantization import QuantConfig
+            from paddle.quantization.quanters import MSEChannelWiseWeightObserver
+            quanter = MSEChannelWiseWeightObserver()
+            q_config = QuantConfig(activation=None, weight=quanter)
+    """
+
+    def __init__(self, quant_bits=8):
+        super(MSEChannelWiseWeightObserver, self).__init__(
+            quant_bits=quant_bits)
+
+    def _get_class(self):
+        return MSEChannelWiseWeightObserverLayer
+
+
+class MSEChannelWiseWeightObserverLayer(AbsMaxChannelWiseWeightObserverLayer):
+    def __init__(self, layer, quant_bits=8):
+        super(MSEChannelWiseWeightObserverLayer, self).__init__(
+            layer, quant_bits=quant_bits)
+
+    def _cal_abs_max(self, inputs):
+        reduce_axis = tuple(
+            [i for i in range(len(inputs.shape)) if i != self.quant_axis()])
+        abs_max_values = paddle.max(paddle.abs(inputs), axis=reduce_axis)
+        abs_max_values = paddle.where(abs_max_values == np.float32(0.0),
+                                      np.float32(1e-8), abs_max_values)
+        return abs_max_values

tests/quantization/test_observers_acc.py

@@ -31,6 +31,8 @@ from paddleslim.quant.observers.mse import MSEObserverLayer
 from paddleslim.quant.observers.avg import AVGObserverLayer
 from paddleslim.quant.observers.emd import EMDObserverLayer
 from paddleslim.quant.observers.kl import KLObserverLayer
+from paddleslim.quant.observers.mse_weight import MSEChannelWiseWeightObserver
+from paddleslim.quant.observers.abs_max_weight import AbsMaxChannelWiseWeightObserver
 from paddle.nn.quant.format import LinearDequanter, LinearQuanter
 
 import logging
@@ -70,10 +72,14 @@ class ImperativeLenet(paddle.nn.Layer):
 
 
 class TestPTQObserverAcc(unittest.TestCase):
-    def __init__(self, observer, observer_type, *args, **kvargs):
+    def __init__(self,
+                 activation_observer,
+                 weight_observer=None,
+                 *args,
+                 **kvargs):
         super(TestPTQObserverAcc, self).__init__(*args, **kvargs)
-        self.observer = observer
-        self.observer_type = observer_type
+        self.act_observer = activation_observer
+        self.weight_observer = weight_observer
 
     def setUp(self):
         paddle.set_device("cpu")
@@ -100,7 +106,9 @@ class TestPTQObserverAcc(unittest.TestCase):
     def init_case(self):
         self.q_config = QuantConfig(activation=None, weight=None)
         self.q_config.add_type_config(
-            paddle.nn.Conv2D, activation=self.observer, weight=self.observer)
+            paddle.nn.Conv2D,
+            activation=self.act_observer,
+            weight=self.weight_observer)
 
     def _count_layers(self, model, layer_type):
         count = 0
@@ -202,7 +210,7 @@ class TestPTQObserverAcc(unittest.TestCase):
         quant_model = ptq.quantize(model, inplace=False)
 
         ptq_sample(quant_model)
-        converted_model = ptq.convert(quant_model, inplace=False)
+        converted_model = ptq.convert(quant_model, inplace=True)
         top1_2, top5_2 = test(converted_model)
 
         _logger.info(
@@ -220,20 +228,26 @@ class TestPTQObserverAcc(unittest.TestCase):
 
 
 observer_suite = unittest.TestSuite()
-observer_suite.addTest(
-    TestPTQObserverAcc(
-        observer=HistObserver(sign=True, symmetric=True),
-        observer_type=PercentHistObserverLayer))
-observer_suite.addTest(
-    TestPTQObserverAcc(
-        observer=KLObserver(bins_count=256), observer_type=KLObserverLayer))
-
-observer_suite.addTest(
-    TestPTQObserverAcc(observer=AVGObserver(), observer_type=AVGObserverLayer))
-observer_suite.addTest(
-    TestPTQObserverAcc(observer=EMDObserver(), observer_type=EMDObserverLayer))
-observer_suite.addTest(
-    TestPTQObserverAcc(observer=MSEObserver(), observer_type=MSEObserverLayer))
+
+for _observer in [
+        AVGObserver(),
+        EMDObserver(),
+        MSEObserver(),
+        KLObserver(bins_count=256),
+        HistObserver(sign=True, symmetric=True),
+]:
+    observer_suite.addTest(
+        TestPTQObserverAcc(
+            activation_observer=_observer, weight_observer=_observer))
+
+for _weight_observer in [
+        MSEChannelWiseWeightObserver(),
+        AbsMaxChannelWiseWeightObserver(),
+]:
+    observer_suite.addTest(
+        TestPTQObserverAcc(
+            activation_observer=MSEObserver(),
+            weight_observer=_weight_observer))
 
 if __name__ == '__main__':
     runner = unittest.TextTestRunner(verbosity=2)