!1963 bug fix in fake quant training in r0.3

Merge pull request !1963 from chenzhongming/r0.3

mindspore/nn/layer/quant.py

@@ -118,7 +118,6 @@ class FakeQuantWithMinMax(Cell):
         quant_delay (int): Quantization delay parameters according by global step. Default: 0.
         symmetric (bool): Quantization algorithm use symmetric or not. Default: False.
         narrow_range (bool): Quantization algorithm use narrow range or not. Default: False.
-        training (bool): Quantization algorithm training or not. Default: True.
 
     Inputs:
         - **x** (Tensor) - The input of FakeQuantWithMinMax.
@@ -143,8 +142,7 @@ class FakeQuantWithMinMax(Cell):
                  out_channels=1,
                  quant_delay=0,
                  symmetric=False,
-                 narrow_range=False,
-                 training=True):
+                 narrow_range=False):
         """init FakeQuantWithMinMax layer"""
         super(FakeQuantWithMinMax, self).__init__()
         self.min_init = min_init
@@ -158,7 +156,6 @@ class FakeQuantWithMinMax(Cell):
         self.quant_delay = quant_delay
         self.symmetric = symmetric
         self.narrow_range = narrow_range
-        self.training = training
         self.is_ascend = context.get_context('device_target') == "Ascend"
 
         # init tensor min and max for fake quant op
@@ -208,7 +205,7 @@ class FakeQuantWithMinMax(Cell):
         return s
 
     def construct(self, x):
-        if self.ema and self.is_ascend:
+        if self.is_ascend and self.training:
             min_up, max_up = self.ema_update(x, self.minq, self.maxq)
             out = self.fake_quant(x, min_up, max_up)
             P.Assign()(self.minq, min_up)

mindspore/train/quant/quant.py

@@ -247,7 +247,7 @@ def convert_quant_network(network,
         network (Cell): Obtain a pipeline through network for saving graph summary.
         quant_delay (int): Number of steps after which weights and activations are quantized during eval. Default: 0.
         bn_fold (bool): Flag to used bn fold ops for simulation inference operation. Default: False.
-        freeze_bn (bool): Number of steps after which BN parameters used total mean and variance. Default: 0.
+        freeze_bn (int): Number of steps after which BN parameters used total mean and variance. Default: 0.
         weight_bits (int): Number of bits to use for quantizing weights. Default: 8.
         act_bits (int): Number of bits to use for quantizing activations. Default: 8.
         per_channel (bool):  Quantization granularity based on layer or on channel. Default: False.