Nigel
/
mindspore
forked from mindspore-Ecosystem/mindspore
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

!35259 extract ema param from quant.py to golden_stick layer_policy 

Merge pull request !35259 from cjh9368/static_check
This commit is contained in:
i-robot 2022-06-02 11:10:47 +00:00 committed by Gitee
parent 1267c54604 4b02866eef
commit fe4f3f2708
No known key found for this signature in database
GPG Key ID: 173E9B9CA92EEF8F
1 changed files with 5 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
mindspore/python/mindspore/nn/layer/quant.py
View File

@ -759,8 +759,7 @@ class Conv2dBnFoldQuantOneConv(Cell):
requires_grad=False)
# initialize fake ops
self.fake_quant_weight = quant_config.weight(ema=False,
channel_axis=channel_axis,
self.fake_quant_weight = quant_config.weight(channel_axis=channel_axis,
num_channels=out_channels)
self.freeze_bn = False
self.bn_train = P.BatchNorm(is_training=True, epsilon=self.eps,
@ -1025,8 +1024,7 @@ class Conv2dBnFoldQuant(Cell):
requires_grad=False)
# initialize fake ops
self.fake_quant_weight = quant_config.weight(ema=False,
channel_axis=channel_axis,
self.fake_quant_weight = quant_config.weight(channel_axis=channel_axis,
num_channels=out_channels)
self.batchnorm_fold = BatchNormFoldCell(epsilon=eps, momentum=momentum, freeze_bn=freeze_bn)
self.correct_mul = Q.CorrectionMul(channel_axis)
@ -1257,8 +1255,7 @@ class Conv2dBnWithoutFoldQuant(Cell):
weight_shape = [out_channels, in_channels // group, *self.kernel_size]
channel_axis = 0
self.weight = Parameter(initializer(weight_init, weight_shape), name='weight')
self.fake_quant_weight = quant_config.weight(ema=False,
channel_axis=channel_axis,
self.fake_quant_weight = quant_config.weight(channel_axis=channel_axis,
num_channels=out_channels,
quant_dtype=quant_dtype)
self.batchnorm = BatchNorm2d(out_channels, eps=eps, momentum=momentum)
@ -1402,8 +1399,7 @@ class Conv2dQuant(Cell):
dilation=self.dilation,
group=self.group)
channel_axis = 0
self.fake_quant_weight = quant_config.weight(ema=False,
channel_axis=channel_axis,
self.fake_quant_weight = quant_config.weight(channel_axis=channel_axis,
num_channels=out_channels)
@classmethod
@ -1555,8 +1551,7 @@ class DenseQuant(Cell):
f"but got {activation}.")
self.activation_flag = self.activation is not None
self.fake_quant_weight = quant_config.weight(ema=False,
channel_axis=0,
self.fake_quant_weight = quant_config.weight(channel_axis=0,
num_channels=out_channels)
@classmethod